Hype Cycle for Artificial Intelligence in Banking, 2025
9 July 2025 - ID G00831999 - 105 min read
By Jasleen Kaur Sindhu
Banks are investing in AI deployment and governance to boost revenue, streamline operations, manage risk and enhance customer experience. This Hype Cycle helps banking CIOs identify high-impact AI trends and prioritize those with the most potential for near-term strategic gains.
Analysis
What You Need to Know
AI adoption continues to accelerate in banking, with 77% of banking CIOs reporting active or planned AI deployments in the current year, according to the 2025 Gartner CIO and Technology Executive Survey. As the broader AI landscape shifts from experimentation to scalable implementation, banks are also moving beyond pilots to embed AI into core operations.
This year’s Hype Cycle reflects a maturing understanding of AI’s potential and limitations. Generative AI (GenAI), while still prominent, is no longer the sole focus. Instead, attention is expanding to foundational enablers like AI-ready data, AI engineering, and AI governance — all critical for sustainable, scalable AI in banking. These shifts mirror broader industry trends, where GenAI has entered the Trough of Disillusionment, and the spotlight is now on operational readiness and responsible deployment.
For banks, this means a strategic pivot from chasing hype to building resilient AI ecosystems. Innovations such as AI agents, composite AI, and multimodal AI are gaining traction, but their success depends on robust data infrastructure, governance, and cost management — areas where banking CIOs must lead with discipline.
New entries like AI observability, FinOps for AI, and banking-specific GenAI models underscore the sector’s focus on transparency, financial efficiency, and domain-specific intelligence. Meanwhile, long-term innovations like quantum AI and artificial general intelligence (AGI) remain on the horizon, offering potential but requiring cautious, long-term investment strategies.
To stay competitive, banks must align AI initiatives with business outcomes, prioritize innovations with near-term value, and invest in the capabilities that will enable AI to scale responsibly and effectively across the enterprise.
The Hype Cycle
This Hype Cycle explores four critical themes essential for banking CIOs to maximize returns from their AI investments. These themes are:
AI foundational elements in banking: Includes the prerequisites necessary for successful AI implementation in banking, including underlying data preparation, engineering practices, computational approaches and cost management strategies. Relevant innovations are:
- AI engineering in banking
- AI-ready data in banking
- Banking-specific GenAI models
- Edge AI
- Federated machine learning
- FinOps for AI
- Multimodal AI
- Synthetic data in banking
Advanced AI techniques and capabilities in banking: Explores current and emerging AI techniques, capabilities and computational approaches that help with development of intelligent systems and processes in banking. Relevant innovations are:
- AI agents in banking
- Artificial general intelligence
- Composite AI in banking
- Generative AI in banking
- Knowledge graphs in banking
- Model distillation
- Neurosymbolic AI
- Quantum AI
Applied AI functionalities in banking: Focuses on the orchestration of multiple advanced AI techniques and capabilities to create new, human-centric functionalities and experiences in banking. These innovations enable intelligent simulations, emotion-aware interactions, advanced decision intelligence, and new approaches for cyber-risk security. Relevant innovations are:
- AI simulation in banking
- AI in cyber-risk security for banking
- Causal AI in banking
- Decision intelligence
- Emotion AI in banking
Responsible and ethical AI in banking: Focuses on ensuring AI systems are transparent, accountable and aligned with ethical standards. This involves managing risks, maintaining system observability and establishing governance structures to guide AI use in banking. Relevant innovations on this Hype Cycle are:
- AI governance in banking
- AI observability in banking
- AI TRiSM in banking
The Priority Matrix
The Priority Matrix is a companion to the Hype Cycle. It seeks to communicate some key attributes contained within the Hype Cycle, namely:
- How much value could an organization realize from the effective implementation of a particular technology?
- When will the technology be mature enough to help deliver expected value?
Two features of this Hype Cycle stand out:
- Almost all the innovations, except for Quantum AI, are high-benefit or transformational.
- Almost half of the featured innovations are expected to reach the Plateau of Productivity in five years or less. Artificial general intelligence (AGI) and quantum AI are the only two entries expected to take more than 10 years to reach mainstream adoption.
High-benefit innovations bring business efficiencies, but require ongoing training and education. Similarly, transformational innovations are game changers, but they require new skills and present high risk and reward.
To achieve practical efficiencies, it is crucial to prioritize innovations expected to achieve mainstream adoption within the next two to five years. Notable examples include the application of GenAI and AI agents, alongside robust AI governance and AI observability solutions.
It is important to recognize that, within the banking sector’s AI Hype Cycle, many innovations are currently at the Peak of Inflated Expectations. As such, banking CIOs must remain vigilant, continuously benchmarking their efforts against market developments to avoid being swept up in the prevailing hype. Investments should be grounded in a clear assessment of which innovations are most likely to deliver strategic or financial value in the context of their organization.
Quantum AI is also noteworthy. Although it is expected to take over a decade to reach mainstream adoption, it offers a unique competitive edge for banks. As validated techniques mature, quantum AI will provide distinct advantages across various industries. Specifically in banking, it will benefit use cases such as portfolio optimization, customer experience enhancement, and fraud prevention.
Artificial general intelligence (AGI) is another innovation expected to take more than a decade to achieve mainstream adoption. In the short term, banking CIOs should remain cautious of the hype surrounding AGI, which often fuels fears and unrealistic expectations about the current capabilities of AI. In the long term, AI capabilities will continue to evolve, with or without AGI, influencing banks through developments such as the emergence of machine customers and autonomous business operations.
Priority Matrix for Artificial Intelligence in Banking, 2025
| Benefit | Years to Mainstream Adoption | |||
|---|---|---|---|---|
| Less Than 2 Years | 2 to 5 Years | 5 to 10 Years | More Than 10 Years | |
Transformational | ||||
High | ||||
Moderate | ||||
Low | ||||
Source: Gartner (July 2025)
On the Rise
Quantum AI
Analysis By: Chirag Dekate, Soyeb Barot
Benefit Rating: Low
Market Penetration: Less than 1% of target audience
Maturity: Embryonic
Definition:
Quantum AI is an embryonic field of research emerging at the intersection of quantum technologies and AI. Quantum AI aims to exploit unique properties of quantum mechanics to develop new and more powerful AI algorithms that deliver better than classical performance, potentially resulting in new types of AI algorithms designed to run on quantum systems.
Why This Is Important
Quantum AI is an area of active research. Once commercialized, quantum AI could potentially help in:
- Enabling organizations to use quantum systems to address advanced AI analytics faster while using a fraction of the resources used in conventional AI supercomputing.
- Developing new AI algorithms that exploit quantum mechanics to deliver capabilities beyond ones that can be executed on classical systems.
- Unlocking disruptive applications that include drug discovery, energy industry and logistics.
Business Impact
While the business impact of the embryonic quantum AI field today is low, when validated techniques mature, quantum AI will enable competitive advantage across industries; for instance:
- Life sciences: Transform drug discovery by shortening timelines, lowering costs and improving outcomes.
- Finance: Optimize portfolios, minimize risk and improve fraud detection systems.
- Material science: Revolutionize energy transportation, manufacturing and create new revenue streams by discovering new materials.
Drivers
- Progress is steady in scaling quantum systems and improving error correction schemes.
- Hype around quantum technologies is driving more businesses and researchers to explore the intersection of quantum and AI.
- The accelerated pace of innovation in quantum systems (including a larger volume of higher quality qubits, and greater stability and reliability of quantum systems) is driving greater interest in applicability in areas, including quantum AI.
- Access to quantum computing as a service is lowering the barrier to entry, encouraging greater collaboration among researchers and enabling exploration of new algorithms and techniques.
- Governments and enterprises globally are increasing funding for quantum (and quantum AI) research, resulting in accelerated innovation.
- The halo effect of increased hype around GenAI is driving new focus on alternative research techniques, including quantum AI, that could potentially deliver new disruptive results.
- Universities and training programs are developing programs and curricula to develop a quantum-ready workforce.
Obstacles
- Hardware limitations: Current quantum systems, while getting stabler, are still error-prone and inherently noisy, limiting their utility and impact on practical quantum AI.
- Algorithm limitations: While several quantum AI algorithms have been proposed, very few have been vetted and proven, and they are nowhere close to being enterprise-ready.
- Cost: Despite their limited utility and widespread accessibility, rapidly evolving noisy intermediate-scale quantum (NISQ) systems are relatively expensive, which could inhibit research and development efforts needed to devise quantum AI algorithms.
- Scalability of systems: Scaling quantum systems to the level necessary for enterprise-ready quantum AI continues to be a major technical hurdle.
- Compute paradigms: Integrating traditional data and analytics pipelines with quantum is inherently challenging because quantum systems operate on a fundamentally different paradigm both from a data representation perspective and from a compute (non-von Neumann model) perspective.
User Recommendations
- Prioritize investments in AI and GenAI over any quantum AI investments. Quantum AI is too nascent to warrant focused investments and unlikely to yield material gains in the next two to three years.
- Partner with local universities by sponsoring academic research as a means of derisking your quantum AI investments and create a university-to-industry talent pipeline.
- Create a quantum AI opportunity radar that enables you to track progress of underlying technologies and quantum AI algorithms, enabling you to maximize value creation as the embryonic field of quantum technologies evolves.
- Diversify quantum use cases beyond a narrow AI context into other domains including materials simulations, search, optimization and other emerging algorithmic domains.
Sample Vendors
Amazon Web Services; Google; IBM; IonQ; Microsoft; Multiverse Computing; Pasqal; SandboxAQ
Gartner Recommended Reading
AI in Cyber-Risk Management in Banking
Analysis By: Lopa Sinha
Benefit Rating: Transformational
Market Penetration: 1% to 5% of target audience
Maturity: Emerging
Definition:
AI in cyber-risk management optimizes processes by analyzing data and providing predictive insights for real-time decision making. AI in cyber-risk management in banking enhances control implementation, identifies deficiencies and scales risk monitoring through automated compliance validation and issue detection.
Why This Is Important
AI in cyber-risk management in banking can:
- Analyze banking data such as transactions to flag suspicious activities, thus minimizing financial losses.
- Ensure adherence to law by mapping controls, automating checks and flagging noncompliance.
- Predict potential vulnerabilities or attack patterns based on historical data and emerging trends.
- Automate routine processes, risk management and incident responses.
- Support applications such as AI agents to orchestrate policy drafts and risk assessment.
Business Impact
With digital transformation, banks generate massive amounts of data across transactions, user sessions and third-party and cloud services. AI reduces dependency on manual operation while increasing effectiveness of security operation. AI also helps stress-test the resilience of critical banking processes by simulating attack paths and proactively identifying business impact. By parsing and correlating millions of events in real time, it enables banks to build automated incident management processes.
Drivers
AI transforms cybersecurity risk management in banking from static, reactive and siloed processes into data-driven, time-sensitive and integrated defenses, and it enables these drivers:
- The increasing prevalence of advanced and AI-powered cyberthreats, including advanced persistent threats (APTs), AI-powered phishing and sophisticated malware, is a primary driver compelling banks to adopt AI for real-time threat detection and response.
- Improvements in the accessibility and applicability of AI and GenAI technology are making these solutions more practical for banking institutions to deploy.
- The necessity for enhanced detection and prevention capabilities across various security domains, such as behavior analytics to identify fraud and insider threats and facilitate adaptive access and authentication, is driving AI adoption.
- The need exists to improve security operations efficiency and effectiveness by enabling security operations center (SOC) teams to prioritize critical alerts. Another improvement approach is to enhance the prioritization and risk scoring of open and self-identified issues to effectively manage cyber risk. Both of these drive the leveraging of capabilities potentially enhanced by AI.
- Strategic forecasting and resource allocation in cybersecurity budgeting, utilizing predictive analytics that consider emerging threats, organizational data and industry trends to anticipate future risks and propose best next actions and optimal security investments, promotes use of AI-driven approaches.
- Maintaining continuous compliance and dynamically managing security policies that align with existing and evolving regulations, industry standards and business-relevant emerging threats are challenges. They drive the need for capabilities like dynamic policy tailoring and comprehensive analysis of policy control gaps, which AI can support.
- The need to manually analyze cyber issues with complex data landscapes while dealing with a shortage of talent is a formidable task without AI augmentation.
Obstacles
- Many core banking systems and mainframes may be incompatible with AI tools. Retrofitting AI into legacy systems will require costly overhaul or custom middleware.
- AI accuracy declines without unified, clean data; however, data relevant to banking security is generally scattered across legacy systems, third-party SaaS and on-premises and cloud infrastructure.
- AI-driven cyber-risk management solutions often require access to sensitive data, such as network logs, user behavior patterns and security incident reports, increasing privacy and security concerns.
- AI algorithms are susceptible to biases inherent in the training data, which can lead to inaccurate recommendations. This creates hesitation among risk owners and cyber-risk teams to trust and further adopt these technologies.
- Cost is a barrier, particularly for small and midsize banks with limited IT capabilities.
- AI tools often don’t plug and play with existing SOAR, SIEM, IAM or risk management platforms.
User Recommendations
- Strategically position AI as a core technology for banking enterprise resilience, specifically by leveraging it to improve security operations benchmarks like mean time to detect (MTTD) and mean time to respond (MTTR). This demonstrates the tangible value of AI in strengthening the organization’s defensive posture.
- Align strategic goals such as customer engagement by connecting real-time threat mitigations to brand trust and operational uptime.
- Quantify cyber-risk exposure with AI-enhanced scenarios to calculate potential loss across data breaches, ransomware and other compromises.
- Leverage AI to map regulatory requirements and ensure technology usage aligns with compliance standards to reduce future audit gaps.
- Develop a comprehensive data strategy that addresses critical considerations like data quality, accessibility, security and privacy, which are foundational for compliant AI deployment in banking.
- Avoid implementing duplicate AI functions across different tools and platforms. Establish a clear reference architecture for AI deployment.
Sample Vendors
Cisco (Splunk); IBM; LexisNexis Risk Solutions; Microsoft; NICE; Palo Alto Networks; ThreatQuotient; Tufin; Visa (Featurespace); Zscaler
Gartner Recommended Reading
Banking-Specific GenAI Models
Analysis By: Sudarshana Bhattacharya
Benefit Rating: High
Market Penetration: Less than 1% of target audience
Maturity: Emerging
Definition:
Domain-specific GenAI models are tailored to specific industry, business function or workflow needs. They enhance accuracy, privacy and compliance while minimizing hallucinations and prompt engineering. Built from scratch or fine-tuned on domain data, these models outperform general-purpose models on targeted use cases. In banking, these models are trained on specific financial data providing deep contextual understanding, enhancing accuracy and regulatory compliance for financial tasks.
Why This Is Important
While general-purpose models perform well across a broad set of applications, they may fall short for many banking applications that require domain-specific data and knowledge. GenAI models tailored to banking and investment services are growing and can improve alignment and trustworthiness with industry use cases while delivering more accurate, compliant and contextualized responses. Through targeted training, these models can lower hallucination risks associated with large models.
Business Impact
Banking-specific GenAI models:
- Outperform general-purpose models on domain-specific tasks by understanding financial language, context and nuances, while maintaining robust model benchmarks.
- Increase precision by reducing hallucination and inaccuracies through banking-specific dataset training.
- Accelerate GenAI project deployment in banks by providing a more advanced starting point with RAG, chain of thought and fine-tuning.
- Offer greater compliance and trust with banking regulations, which increases adoption within the industry.
Drivers
- The proliferation of open-source foundation models: The banking sector is capitalizing on open-source GenAI models for domain-specific solutions. Capital One’s Chat Concierge tool, built on Llama and integrated into car dealership websites, facilitates vehicle comparisons, explores financing options, offers trade-in estimates and assists in scheduling test drives.
- Increased specialization in GenAI model use: AI models are increasingly trained on private, organization-specific financial data that is not available in open-source models. This exclusive data allows for focused training, which enables models to excel in specialized tasks like customer service automation (Ally Bank), document analysis, financial forecasting and advisory services. As a result, banking-specific GenAI models offer precision and efficiency by using proprietary data and sidestepping the extensive customization needed with general-purpose solutions.
- Enhanced data privacy: Banking-specific GenAI models address challenges of confidentiality and privacy when handling PII data.
- Reduced hallucination: Transparent and proprietary training data creates a narrower knowledge base, which reduces unexpected outputs.
- Growing interest in domain-specific AI agents: The current hype around agentic AI — systems that can autonomously plan, reason and take actions — is fueling rapid growth in domain-specific GenAI models. As banks realize the value of specialized agents for processes like portfolio management, which require deep understanding of market trends, they push strongly to use tailored models optimized for those verticals, combining general AI capabilities with domain-specific expertise.
Obstacles
- Compliance and security: Open-source models may conflict with the bank’s internal IT standards, posing compliance threats from bad actors.
- Model proliferation and reduced versatility: Optimizing models like robo advisors and customer assistants for specific tasks limits their broader applicability. On the other hand, increasing the numbers of models complicates governance and management.
- Model maintenance: Managing domain-specific models requires frequent updates to avoid drift and adds operational complexity to the model risk management process.
- Data scarcity and quality: Domain-specific models require high-quality, relevant data, which is often scarce because of the prevalence of legacy systems in banking. Data silo and data quality issues require data cleanups that make model training challenging.
- Computational cost and in-house expertise: Banks have challenges in hiring and retaining skilled AI engineering and DevSecOps resources because of the technology sector’s competitive salaries and benefits.
User Recommendations
- Gather representative, high-volume and high-quality data for model pretraining to encompass all possible scenarios like lending, risk management, fraud detection, customer service, compliance or competitive differentiation.
- Establish life cycle management practices like model drift and data drift monitoring to avoid model degradation and ensure continuous compliance with regulatory or industry standards.
- Inquire about intended use cases and transparency details, such as training data size, quality, relevance, domain knowledge tests and data privacy measures, when evaluating banking-specific AI models. Assess accuracy benchmarks on real-world banking tasks.
- Assess the model for regulatory and ethical compliance by examining its model card, benchmarks and metrics, and confirm the vendor’s domain expertise in banking to ensure it understands the industry’s unique challenges.
Sample Vendors
AI4Finance Foundation; Ant Group; Broadridge; BUSINESSNEXT; Squirro
Gartner Recommended Reading
Causal AI in Banking
Analysis By: Benjamin Seesel, Priyanka Shukla, Ben Yan, Leinar Ramos
Benefit Rating: High
Market Penetration: 1% to 5% of target audience
Maturity: Emerging
Definition:
Causal AI identifies and utilizes cause-and-effect relationships to go beyond correlation-based predictive or generative models and toward AI systems that can prescribe actions more effectively and act more autonomously. It includes different techniques, such as causal graphs and simulation, that help uncover causal relationships to improve decision making. A deeper understanding of events is critical for several banking functions, such as customer engagement, operations and risk management.
Why This Is Important
Causal AI goes beyond correlation-based generative and predictive approaches to help understand the underlying factors that cause an outcome, so banks can take proactive, targeted actions. Banks can use causal AI in many ways, such as to analyze customer behavior and predict future needs, identify loan default drivers, improve financial forecasting, detect fraud or understand how new events or policy changes will impact business outcomes.
Business Impact
Causal AI helps banks:
- Devise data-driven customer engagement and retention strategies, driving revenue growth
- Improve credit risk assessment and prediction of default by differentiating between correlation and causation
- Identify the root cause of disruptions in business processes, improving operational efficiency and cost saving
- Unlock the power of explainable and understandable AI, enabling reliable augmentation and autonomy in AI-driven decision making, even in changing economic and competitive environments
Drivers
- AI — in particular agentic AI — systems increasingly need to act autonomously, particularly for time-sensitive and complex use cases (such as fraud detection and credit assessment). This will only be possible if AI understands what impact actions will have and how to make effective interventions.
- Limited data availability for certain use cases requires more data-efficient techniques like causal AI, possibly combined with synthetic data. Causal AI leverages human domain knowledge of cause-and-effect relationships to bootstrap AI models in small-data situations.
- Growing complexity in banking and continuously evolving regulatory, competitive and technological landscapes requires more robust AI techniques. Correlation-based AI models, trained with historical data, are brittle and lose accuracy when faced with gradual, let alone disruptive, changes. Causal structure changes much more slowly than statistical correlations, making causal AI more robust and adaptable in fast-changing environments.
- Banks face significant pressure to ensure transparency, explainability and interpretability in decisions made by AI systems. Causal AI can help explain why the model arrived at a particular decision.
- Generative AI (GenAI) can accelerate causal AI implementation. GenAI is emerging as an aid to explore documents and other data sources for existing causal knowledge. This can then be used to generate candidate causal graphs, which, while still requiring human validation or completion, may reduce time-consuming manual work.
- The next step in AI requires causal AI. Current deep learning models, in particular large language models (LLMs) and “reasoning” models for GenAI and AI agents, have limitations in terms of reliability. A composite AI approach that complements, for example LLMs with causal AI — in particular, causal knowledge graphs — offers a promising avenue to bring AI to a higher level, especially helpful in use cases such as fraud management, AML and customer authentication.
Obstacles
- Causality is not trivial. Not every phenomenon is easy to model in terms of cause and effect, with many factors potentially being relevant. Causality might be delayed, circular, unknown or hard to validate, despite the growing use of AI for causal discovery.
- Identifying a complete list of causal factors is difficult in today’s banking landscape, which is impacted in unpredictable ways by geopolitical risks, macroeconomic uncertainties and widely varying regulations around the world.
- Causal AI requires high-quality data, but data in banking is often siloed, disparate and not AI-ready.
- Even when models appear to function well, flawed assumptions about cause and effect can lead to incorrect or risky conclusions.
- Top causal AI use cases in banking, such as fraud prevention and credit and liquidity risk assessment, carry significant regulatory, operational and reputational risk if the causal AI model is flawed.
- Limited experience, skill sets and technical knowledge with enterprise-scale applications make it difficult for banks to scale causal AI pilots to larger and more complex causal models.
User Recommendations
- Run proofs of concept (POCs) to understand how causal AI complements existing AI approaches, including machine learning (ML), GenAI, agentic AI, and impact reliability and transparency.
- Use causal AI when more augmentation and automation is required. Examples include use cases such as next best action advisory tools, credit risk assessment, fraud detection, portfolio optimization and loan origination.
- Select different causal AI techniques based on the complexity of the use case, including ML or LLMs for causal discovery, causal rule inferencing, causal graphs, Bayesian networks or simulation.
- Educate your AI teams on how causal AI differs from correlation-based AI and the full range of applicable techniques if you plan to scale causal AI following a successful POC.
- Involve domain experts across lines of business, such as retail and commercial banking and wealth management, as applicable, in causal AI initiatives to help create, maintain or validate causal models.
Sample Vendors
Actable AI; Bayes Server; causaLens; Causality Link; Geminos; Howso; Parabole.ai; Scalnyx; Vizuro; Xplain Data
Gartner Recommended Reading
AI Simulation in Banking
Analysis By: Jasleen Kaur Sindhu, Leinar Ramos
Benefit Rating: High
Market Penetration: 1% to 5% of target audience
Maturity: Adolescent
Definition:
AI simulation is the combined application of AI and simulation technologies to jointly develop AI agents and the simulated environments in which they can be trained, tested and sometimes deployed. It includes both the use of AI to make simulations more efficient and useful and the use of a wide range of simulation models to develop more versatile and adaptive AI systems. AI simulation has a wide application in banking, from stress-testing market scenarios to simulating new fraud categories.
Why This Is Important
AI simulation has a wide-ranging application in banking. By combining AI and simulation, banks can model new risk and market scenarios to optimize investment decisions and improve predictive insights. New fraud scenarios can be simulated to strengthen the existing fraud detection capabilities. Banks can improve customer and competitor intelligence by simulating customer preferences to new products or marketing campaigns, including developing new credit products that minimize customers’ default risk.
Business Impact
AI simulation can enable banks to:
- Optimize investment strategies and evaluate risk exposure by simulating different and new financial scenarios.
- Use stress-testing scenarios for financial crime prevention.
- Onboard frontline staff faster on how to respond to different scenarios through situational simulation training.
- Enhance custom retention strategies and new product development pipelines.
- Reuse simulation environments to train future AI models.
Drivers
- Limited availability of AI training data: The scarcity in available AI training data is increasing the need for synthetic data techniques such as simulation. This is particularly critical for fraud prevention or credit decision making that requires a diverse balanced dataset. Generative AI (GenAI) and agentic AI implementations will further necessitate the need for more training data.
- Data privacy requirements: AI simulation allows banks to safely generate and share data, without compromising customers’ personally identifiable information (PII) with fintechs and technology providers.
- IT testing: By automating and optimizing testing processes, AI can significantly speed up quality assurance and engineering testing cycles, enabling quicker releases and updates.
- Regulatory compliance: Simulation offers a unique opportunity for banks to run and model millions of market scenarios. Banks can run stress tests and avoid large-scale market contagions, optimize capital allocations and ensure regulatory compliance.
- Innovation opportunities: With the rise in digital banking, banks now have greater access to customer data. AI simulation can further enhance banks’ ability to use this data to experiment and innovate. For instance, creating novel customer-facing applications that provide financial coaching to customers, or using AI agents to simulate and test customer preferences.
- Increasing number of sophisticated cyber and fraud attacks: This is driving the usage of AI and simulation-generated synthetic data to identify new risk types and fraud scenarios. For example, banks are using synthetic data to identify new variables and clusters to improve the accuracy of their fraud detection models.
- AI reusability: Banks will increasingly deploy hundreds of AI models, which requires a shift in focus toward building persistent, reusable environments where many AI models can be trained, customized and validated. Simulation environments are ideal since they are reusable and scalable, and enable many AI models to be trained at once.
Obstacles
- Talent scarcity: A lack of awareness and skills among AI practitioners to use simulation techniques can be a major obstacle.
- High cost: Using AI simulation can be expensive for most banks; additional investment is required to improve accuracy of the simulated output.
- Gap between simulation and reality: Simulations can only emulate — not fully replicate — real-world systems. Given this gap, AI models trained in simulation might not have the same performance and accuracy.
- Lack of explainability: Banks need to ensure that users can understand, explain and trust the results generated through AI simulations. Further, the combination of AI and simulation techniques can result in more complex pipelines that are harder to test, validate, maintain and troubleshoot.
- Data quality and availability: AI simulation relies on good quality and quantity of data. This requires integrating data from disparate sources and legacy systems and removing underlying biases, which might be a challenge for most banks.
User Recommendations
- Start with tried and tested business applications, such as stress-testing market scenarios or investment portfolios, where AI simulation has proven to be a success.
- Ensure AI simulation models comply with regulations related to data and model privacy, fairness, transparency and explainability. Invest in techniques and tools to build trust.
- Create synergies between AI and simulation teams, projects and solutions to enable a next generation of more adaptive solutions for evermore complex use cases. Incrementally build a common foundation of more generalized and complementary models that are reused across different use cases, business circumstances and ecosystems.
- Prepare for the combined use of AI, simulation and other relevant techniques, such as graphs, natural language processing (NLP) or geospatial analytics, by prioritizing vendors that offer platforms that integrate different AI techniques (composite AI) and simulation.
- Identify skills and competencies required to train or hire AI talent.
Sample Vendors
Altair Engineering; The AnyLogic Company; FICO; FNA; Microsoft; NVIDIA; Simudyne; Zenarate
Gartner Recommended Reading
Decision Intelligence
Analysis By: David Pidsley, Pieter den Hamer, Erick Brethenoux
Benefit Rating: Transformational
Market Penetration: 5% to 20% of target audience
Maturity: Emerging
Definition:
Decision intelligence (DI) is a practical discipline that advances decision making by explicitly understanding and engineering how decisions are made, and how outcomes are evaluated, managed and improved via feedback. By digitizing and modeling decisions as assets, DI bridges the insight-to-action gap to continuously improve decision quality, actions and outcomes. DI is technology-agnostic and applies decision-centric frameworks like observe, orient, decide and act (OODA) and Gartner DI (GDI).
Why This Is Important
Agentic AI and generative AI (GenAI) hype, regulatory pressures on decision automation, and recent global crises have exposed the fragility of business processes and the predigital, implicit and suboptimal ways of decision making that remain incumbent. DI is positioned beyond the trigger, poised to address these challenges by making decisions more explicit, optimal, adaptable and auditable.
Business Impact
- Faster, higher quality decisions that are consistent, compliant and cost-effective while being complex, contextual and continuous, thus driving agility in facing opportunities and threats in domains like banking, healthcare and supply chain.
- Enduring, effective, efficient, explainable and ethical enterprisewide DI execution enhances timely stakeholder outcomes.
- Risk is mitigated through accurate, trustworthy, fair, privacy-protective and scalable decision-centric operationalization of AI to augment and automate decisions.
- Adaptability of decisions as assets strengthens decision governance and outcome predictability.
Drivers
- Dynamic business complexity: Unpredictable disruptions, chaotic environments and accelerating pace of digital competition demands near real-time decision models that can adapt. Decision services can be powered by the composition of multimodal data analysis, data science, optimization, expert knowledge and other AI techniques.
- Decision silos: DI curtails fragmented, localized and implicit decisions that undermine organizational efficacy and efficiency. It also addresses the demand for cross-functional alignment on decisions as assets, the need for harmonization on which action should be taken following a business decision, and outcome optimization that balances global efficiency and local adaptations.
- Deluge of dashboards not driving action: Despite proliferation of “data-driven” tools and interfaces, most of which fail to connect insights to actions, dashboard development delays create decision latency, ambiguous outcomes and inability to perceive a decision’s impact harming organization efficiency.
- Human-AI delegation and distrust: AI adoption requires transparent, auditable decision models to address ethical concerns and ensure accountability. Automating human decisions has promoted disquiet and requires monitoring.
- Regulatory scrutiny: Data protection, AI and socio-environmental mandates compel explicit decision documentation for tighter compliance, risk awareness and mitigation. Explicit decision modeling and decision stewardship drive the analysis, management and control of the operational processes and observations needed to enforce decision governance policies and standards applied to decisions as assets.
- Availability and innovation of enabling technologies: Convergence of rule engines, simulation and optimization in DI platforms practically enables DI prototypes and pilots to become scalable DI implementations.
- GenAI acceleration of DI: Enriched context awareness via LLMs is accelerating composite AI model development for low-code/no-code business decision analysts and pilots of agentic decision automation.
Obstacles
- Business stakeholder apathy, limited urgency and low cultural readiness, ineffective change management, and lack of DI skills and AI literacy hinder adoption.
- Bridging the insight-to-action gap to improve outcomes requires a decision-centric vision beyond the data-driven dogma and the data-to-insight workflow. Technology centricity overlooks psychological and sociological factors in decision making.
- Weak collaboration, inadequate operating, delivery and organizational models (i.e., a DI center of excellence), and disconnected decision-making silos hamper DI effectiveness. Even advanced cross-silo DI practitioners struggle to impartially reconsider key decision flows.
- Unselective or overly enthusiastic adoption of decision automation introduces risks, including unintended consequences, loss of context and bias amplification. This undermines trust in DI and limits effective use of DI platforms.
User Recommendations
- Define and model critical decisions involving resource allocation, uncertainty or competing alternatives. Use these as pilots to build DI momentum and demonstrate value for enterprisewide adoption.
- Inventory repetitive, high-impact decisions and their key inputs. Adopt decision-centric modeling by articulating outcomes, decision logic, alternative courses of actions and required observations to drive continuous learning, improvement and transparency.
- Maximize decision quality, resilience and traceability through cross-functional DI fusion teams, fostering collaboration and alignment across departments. Delegate decision-making authority to those with the most relevant expertise and context.
- Upskill staff in decision modeling, prescriptive analytics and optimization. Investigate the roles of decision engineers, decision scientists and decision stewards. Experiment with agentic, GenAI and other composite AI, and DecisionOps to support organizationwide decision centricity and excellence.
Gartner Recommended Reading
FinOps for AI
Analysis By: Jim Hare, Adam Ronthal, Andrei Razvan Sachelarescu
Benefit Rating: High
Market Penetration: 1% to 5% of target audience
Maturity: Emerging
Definition:
FinOps for AI is the application of financial operations best practices to help organizations increase visibility and manage the costs of AI services to ensure efficient usage and deliver maximum business impact. Using FinOps to track and measure AI spend and usage is crucial for optimizing costs, ensuring financial accountability and maximizing ROI.
Why This Is Important
Cost poses one of the greatest near-term threats to AI and GenAI success. AI workloads, especially in cloud environments, often use expensive GPU-based compute infrastructure and consume tokens in unforeseen ways, leading to unpredictable expenses if not monitored properly. Deploying and managing AI solutions generates other costs, including development, governance and change management. Using FinOps to track and measure AI spend and usage is crucial for optimizing costs, ensuring financial accountability and maximizing ROI.
Business Impact
FinOps helps businesses optimize AI spend by providing real-time cost visibility and control, enabling teams to allocate resources efficiently and prevent budget overruns while also preventing underprovisioning that can cause downtime or slowdowns. FinOps for AI also enhances collaboration between finance, engineering and operations teams, ensuring that AI investments align with business objectives while ensuring cost-efficiency. Using FinOps practices, organizations can maximize the ROI of AI initiatives and leverage cost-saving opportunities such as reserved instances, workload automation/optimization and usage-based pricing models.
Drivers
- AI adoption, especially AI applications and GenAI, is contributing to a spike in cloud costs for most enterprises. Hidden costs and unpredictable invoices make it difficult for organizations to deploy AI more broadly.
- Tracking AI costs and usage scaling is complex due to fluctuating computational demands, variable AI service pricing, hidden infrastructure costs and exponential scaling of model training and inference across users and applications.
- Organizations new to AI and/or the cloud are unlikely to be prepared for AI cost volatility and will need to adjust their legacy operating models and budget practices by adopting FinOps for AI. Many organizations face challenges in tracking and measuring AI costs against concrete business benefits.
- Engineering teams are often immature in their use of AI services and the many dynamic layers needed to achieve ongoing cost-effectiveness.
- The total cost of ownership (TCO) of AI use cases can differ from the cost of traditional software applications with fixed costs and purpose. Continuous training, switching to newer models, specialized infrastructure like GPUs and differences in processing costs for specific data types (text, image, video, audio) are part of ongoing AI costs.
- Many AI models and services are based on consumption pricing models and may be purchased in many versions or variants.
- Pricing may also fluctuate based on a variety of factors such as usage, model choice, accuracy and performance guarantees. The velocity of pricing volatility requires continuous and active assessments of price/performance and accuracy.
Obstacles
- Implementing FinOps for AI is challenging because of AI workloads’ unpredictable and dynamic nature and the complexity and variety of cost factors that make cost estimation, budgeting and optimization more challenging compared with traditional cloud operations.
- Balancing performance and cost-efficiency is difficult because AI models often require specialized compute infrastructure resources, GPUs and large datasets that can lead to excessive cloud spending if they are not monitored and optimized effectively.
- Many organizations struggle with cross-functional collaboration among finance, operations and engineering teams, as aligning AI-specific cost insights with business objectives requires a cultural shift and enhanced visibility into AI-driven expenditures.
User Recommendations
- Track the TCO of using AI: Implement real-time tracking of total running costs, including cloud, infrastructure and labor costs. Use tagging and cost allocation strategies to assign expenses to specific AI projects or departments. Assign budgets to AI-related resource and service groups, and trigger cost alerts when consumption exceeds budget goals.
- Optimize AI spend and workloads: Track AI spend across packaged and custom SaaS, AI-leveraging commercial models (tokens via API calls), and compute from hosted models.
- Understand the pros and cons of buying versus building models: Closed models built by model providers may be considered more expensive at first glance, but they reduce delivery time, upfront development costs and the need for more expensive skills. Build models for truly strategic types of use cases.
- Embrace an agile approach to model switching: Regularly compare models in use with alternative options to see whether the same or better accuracy can be achieved at lower cost.
- Invest in making data AI-ready: Control data preparation and processing costs by investing in data cleansing and curation to produce smaller training and retrieval-augmented generation datasets of higher quality.
- Implement proactive cost management controls and guardrails: Integrate real-time anomaly detection and alerts with demand-throttling options to guard against unexpected cost spikes.
Sample Vendors
Airia; Exostellar; FinOps Foundation; Finout; Flexera; IBM
Gartner Recommended Reading
Composite AI in Banking
Analysis By: Moutusi Sau, Erick Brethenoux, Pieter den Hamer
Benefit Rating: Transformational
Market Penetration: More than 50% of target audience
Maturity: Early mainstream
Definition:
Composite AI, or hybrid AI, refers to the combined application (or fusion) of different AI techniques to improve the efficiency of learning and expand the level of knowledge representations. It broadens AI abstraction mechanisms and provides a platform to solve a wide range of banking problems effectively. In banking, the ability to deeply contextualize data is applied in areas like detecting fraud, adhering to compliance or constructing a 360-degree customer view.
Why This Is Important
Composite AI recognizes that no single AI technique is a panacea. It aims to combine connectionist AI approaches, like machine learning (ML) and deep learning, with symbolic and other AI approaches, like rule-based reasoning, graph analysis or optimization techniques. A single technique can rarely solve banking problems. Integrating the strengths of different techniques can bring in nuances to the solution. Thus, composite AI is at the center of the generative AI (GenAI), decision intelligence (DI) and agentic AI markets.
Business Impact
With its emphasis on customer trust and responsibility to regulations, composite AI is critical to the banking industry being able to safely employ GenAI. We’ve seen a combination of techniques, including AI agents or ML models, conduct all regulated tasks around GenAI implementation. By cross-verifying results from multiple AI methods and models, composite AI can lead to more accurate predictions and fewer errors, which is crucial in risk-sensitive areas like credit scoring and fraud detection.
Drivers
- Enhancing financial crime prevention: Composite AI significantly augments existing rule-based systems with methods like behavioral, network and login analysis. For AML, it utilizes composite methods like knowledge graphs, label propagation, clustering and computer vision. It also aids the know-your-business process by analyzing transactions to uncover patterns revealing connected organizations and potential money-laundering activities.
- Simulating and future-proofing the business: Agent-based modeling, identified as the next wave of composite AI, uses multiple agents to represent actors in the ecosystem. This allows banks to simulate complex situations like financial crises or how investors interact during emergent behaviors. Agents will be able to determine the best AI tools for specific problems.
- Deepening customer hyperpersonalization: Composite AI facilitates truly personalized customer experiences by combining various AI methods to analyze diverse customer data, including transactions, interactions, preferences and sentiments. This ranges from tailored product recommendations to personalized financial advice.
- Addressing data scarcity with synthetic data creation: Enterprises are now complementing scarce, raw historical data with synthetic data created using composite AI techniques. Methods such as knowledge graphs and generative adversarial networks (GANs) generate this synthetic data, which is increasingly valuable for use cases like fraud detection and customer-facing applications.
- Increasing strategic advantage of GenAI: The acceleration of GenAI is actively driving the research and adoption of composite AI models. Such models are described as the foundation of the digital innovation platforms that are becoming prevalent across banks.
Obstacles
- Data scientists as decision makers: Applying multiple AI methods together is still in its early days. Ultimately, banking data scientists decide how to combine AI methods, which prevents applications from scaling fast.
- Lack of talent to leverage multiple AI methods: Hiring someone with data science experience is still expensive in banking, and getting talent to leverage multiple methods is still uncommon.
- Trust and risk barriers: The AI engineering discipline is also starting to take shape, but only mature banks have started to apply its benefits in operationalizing AI techniques. Organizations must first address security, ethical model behaviors, observability, model autonomy and change management practices across the combined AI techniques.
- Deploying ModelOps: The ModelOps domain in banking remains fragmented with multiple tooling. A robust ModelOps approach is required to efficiently govern composite AI environments and harmonize them with other areas, such as DevOps and DataOps.
User Recommendations
- Work with banking leaders to identify projects, including credit decisioning or fraud detection, in which an ML-only approach is inefficient or doesn’t work well. This includes cases where enough data is not available or when the pattern cannot be represented through current ML models.
- Capture domain knowledge and human expertise in areas like front office or risk management that require context for data-driven insights by applying decision management with business rules and knowledge graphs, in conjunction with ML and/or causal models.
- Capture domain expertise within banking areas like lending, credit decisioning workflows to provide context for data-driven insights by applying decision management with business rules and knowledge graphs, in conjunction with ML and/or causal models.
- Accelerate the development of DI projects by encouraging experimentation with GenAI, which in turn will accelerate the deployment of composite AI solutions.
Sample Vendors
4Paradigm; ACTICO Group (ACTICO); Aera; FICO; Frontline Systems; IBM; Indico Data; SAS; Simudyne
Gartner Recommended Reading
Neurosymbolic AI
Analysis By: Erick Brethenoux, Afraz Jaffri
Benefit Rating: High
Market Penetration: 5% to 20% of target audience
Maturity: Emerging
Definition:
Neurosymbolic AI is a form of composite AI that combines probabilistic reasoning methods and symbolic systems to create more robust and trustworthy AI models. This fusion enables the combination of probabilistic models with logic-based techniques (such as rules and knowledge graphs) to enable AI systems to better represent, reason and generalize concepts. This approach provides a reasoning infrastructure for solving a wider range of business problems more effectively.
Why This Is Important
Neurosymbolic AI addresses limitations in current AI systems, such as incorrect outputs, lack of generalization to a variety of tasks and an inability to explain the steps that led to an output. The neurosymbolic approach leads to more powerful, versatile and interpretable AI solutions and allows AI systems to reason through more complex tasks. Generative AI systems are starting to leverage neurosymbolic methods to overcome their reasoning shortcomings.
Business Impact
Neurosymbolic AI will have an impact on the efficiency, adaptability and reliability of AI systems used across business processes. The integration of logic and multiple reasoning mechanisms brings down the need for ever larger AI models and their supporting infrastructure. These systems will rely less on the processing of huge amounts of data, making AI agile and resilient. Neurosymbolic approaches can augment and automate decision making with less risk of unintended consequences.
Drivers
- Neurosymbolic AI addresses the limitations of large reasoning models (LRMs), which are still plagued with a lack of symbolic abstraction when exclusively based on deep learning techniques.
- The need for explanation and interpretability of AI outputs is especially important in regulated industry use cases and in systems that use private data.
- Understanding the meanings behind words, not just their arrangement (semantics over syntax), is an increasing priority in systems that deal with real-world entities to ground meaning to words and terms in specific domains.
- The set of tools available to combine different types of AI models is increasing and becoming easier to use for developers and end users. The dominant approach is to chain together results from different models (composite AI) rather than using single models.
- The integration of multiple reasoning mechanisms necessary to provide agile AI systems eventually leads to adaptive AI systems, notably through blackboardlike mechanisms.
- Agentic AI advances also participate in advancing neurosymbolic methods, while agents using various composite AI techniques collaborate to solve problems.
Obstacles
- Most fundamental neurosymbolic AI methods and techniques are being developed in academia or industry research labs. Despite the increased availability of tools, implementations in business or enterprise settings are still limited.
- No agreed-upon techniques exist for implementing neurosymbolic AI, and disagreements continue between researchers and practitioners on the effectiveness of combining approaches, despite the emergence of real-world use cases.
- The commercial and investment trajectories for AI startups allocate almost all capital to deep-learning approaches, leaving only those willing to bet on the future to invest in neurosymbolic AI development.
- Currently, despite increasing exposure, popular media and academic conferences do not give as much exposure to the neurosymbolic AI movement as compared to other approaches (such as generative AI).
User Recommendations
- Adopt composite AI approaches when building AI systems by using a range of techniques that increase the robustness and reliability of AI models. Neurosymbolic AI approaches will fit into a composite AI architecture.
- Dedicate time to learning about neurosymbolic AI approaches, and to identifying use cases that can benefit from applying these approaches.
- Invest in data architecture that can leverage the building blocks for neurosymbolic AI techniques, such as knowledge graphs and agent-based techniques.
- Consider neurosymbolic AI architectures when the limitations of generative AI models prevent their implementation in the organization.
- Educate developers on the potential of neurosymbolic models by exploring the capabilities of neurosymbolic approaches while building learning AI agents.
Sample Vendors
Franz; Google DeepMind; IBM; Microsoft; RelationalAI; Wolfram|Alpha
Gartner Recommended Reading
Artificial General Intelligence
Analysis By: Pieter den Hamer, Philip Walsh
Benefit Rating: Transformational
Market Penetration: Less than 1% of target audience
Maturity: Embryonic
Definition:
Artificial general intelligence (AGI) is the (currently hypothetical) capability of a machine that can match or surpass the capabilities of humans across all cognitive tasks. In addition, AGI will be able to autonomously learn and adapt in pursuit of predetermined or novel goals in a wide range of both physical and virtual environments.
Why This Is Important
With AI’s growing sophistication — including the recent advances in generative AI (GenAI) and agentic AI — a growing number of AI experts have shortened their predicted timelines for achieving AGI in the future or view AGI as no longer purely hypothetical. A clear, shared definition of AGI is necessary for evidence‑based governance and realistic expectations. Achieving AGI would be a transformative tipping‑point with profound consequences for productivity, employment, geopolitical power, legal, ethical and cultural norms — and society at large.
Business Impact
In the near term, anticipation of AGI fuels both overly optimistic expectations and existential fears, skewing investment, distorting trust and accelerating the emergence of new AI regulations. Over the longer horizon, the question of who builds and controls AGI — or other forms of increasingly powerful AI — looms large. Many experts see public stewardship as essential, a prospect that could upend private advantage and redraw entire markets.
Drivers
- Recent advances and growing interest in multimodal large language models (LLMs), so-called reasoning models and AI agents drive considerable hype about AGI. The massive scaling of deep learning and the availability of huge amounts of data and compute power largely have enabled these advances.
- AI’s further evolution toward AGI, as defined here, is increasingly complemented by other partially new approaches, such as knowledge or causal graphs, world models, adaptive AI, embodied AI, composite and neurosymbolic AI, and likely other innovations yet unknown.
- A number of AI vendors are openly discussing and actively researching the field of AGI, creating the impression that AGI lies within reach. However, their definitions of AGI vary greatly and are often open to multiple interpretations. Moreover, other leading AI vendors and experts have dismissed AGI as hype, urging focus on the real impact of AI’s growing capabilities.
- Humans’ innate desire to set lofty goals is also a major driver for AGI. At one point in history, humans wanted to fly by mimicking bird flight. Today, airplane travel is a reality. The inquisitiveness of the human mind, taking inspiration from nature and from itself, is not going to fizzle out.
- People’s tendency to anthropomorphize nonhuman entities also applies to AI-powered machines. The humanlike responses of LLMs and the reasoning-like capabilities of recent AI models have fueled this tendency. Although many philosophers, neuropsychologists and other scientists consider this attribution highly uncertain or going too far, it has created a sense that AGI is within reach or at least is getting closer. In turn, this has triggered massive media attention, several calls for regulation to manage the risks of AGI and a great appetite to invest in AI for economic, societal and geopolitical reasons.
Obstacles
- Little scientific consensus exists on the meaning of “human intelligence.” Any claims about AGI are hard to validate in the face of the enormous complexity of the human brain and mind and such a limited understanding of them.
- Unreliability, lack of transparency and limited abstraction and reasoning of pattern-based capabilities in current AI are not easy to overcome with deep learning’s intrinsically probabilistic approach. More data or more compute power for ever-bigger models is unlikely to resolve these issues, let alone to achieve AGI. To realize (and control) AGI will require further technological innovations. Therefore, AGI as defined here is unlikely to emerge in the near future.
- If AGI materializes, autonomous actors likely will emerge that, in time, will be attributed with full self-learning, agency, identity and perhaps even morality. This will open a bevy of considerations about AI’s legal rights and trigger profound ethical and even religious discussions. AGI also brings the risk of negative impacts on humans, from job losses to a new, AI-triggered arms race and more. A serious backlash may result, and regulations to ban or control AGI are likely to emerge in the near future.
User Recommendations
- Engage with stakeholders to address excessive optimism or unwarranted pessimism, and create or maintain realistic expectations around AGI. Ground AI strategy in concrete business problems rather than speculative AGI forecasts. Recalibrate the AI portfolio periodically as AI capabilities evolve, while leveraging the complementary strengths of human and artificial intelligence.
- Stay apprised of scientific and innovative breakthroughs that may indicate AGI’s possible emergence; however, be aware of the broad range of definitions and views regarding AGI, some strict and some less strict. Meanwhile, keep applying current AI to learn, reap its benefits and develop practices for its responsible use.
- Assess whether AI systems truly meet their specific use-case needs, rather than relying on generic measures of intelligence.
- Prepare for emerging AI regulations and promote internal AI governance to manage current and emerging AI risks. Because although AGI as defined here is not a reality now, current AI already poses significant risks regarding ethics, reliability and other areas.
Sample Vendors
Aigo; Amazon; Anthropic; Butterfly Effect; DeepSeek; Google; Microsoft; OpenAI
Gartner Recommended Reading
Federated Machine Learning
Analysis By: Tong Zhang, Bart Willemsen, Svetlana Sicular
Benefit Rating: High
Market Penetration: Less than 1% of target audience
Maturity: Emerging
Definition:
Federated machine learning (FedML) is a decentralized approach to machine learning that enables multiple clients to collaboratively train a shared model without sharing their raw data, enhancing privacy and overcoming data transfer constraints. This process involves clients performing local training on their data and sending updates to a central server, which aggregates these updates to refine the global model.
Why This Is Important
FedML highlights an important innovation in (re)training ML algorithms in a decentralized environment without disclosing sensitive business information. It enhances model personalization and contextualization by allowing local data processing in smartphones, softbots, autonomous vehicles or Internet of Things (IoT) edge devices. It also facilitates organizations to build collaborative learning models across data silos.
FedML unlocks access to diverse datasets, improving model accuracy and robustness while overcoming challenges related to data gravity and sovereignty.
Business Impact
FedML offers transformative benefits across industries by enabling collaborative machine learning while preserving data privacy and security. It allows organizations to enhance model accuracy and operational efficiency without transferring sensitive data, reducing costs and ensuring compliance with data regulations. FedML is applicable in sectors such as healthcare, finance, telecommunications, IoT and manufacturing, providing solutions for diagnostics, fraud detection, network optimization and personalized services. It also facilitates training and fine-tuning AI agents and large language models using private data.
Drivers
- The proliferation of privacy and legislative regulations requires protection of local data. FedML is one type of solution to protect data privacy.
- The training of AI agents and LLMs requires the use of private, distributed datasets without exposure.
- Growth of edge computing and IoT requires data processing and model training directly and in real time on distributed devices, reducing the need for data centralization.
- Exploding data volumes and data gravity make large-scale data transfers challenging, but FedML can resolve these challenges by processing data locally and avoiding the need for centralization.
- Collaboration is essential for organizations to gain valuable data insights FedML facilitates this collaboration by enabling shared model training without directly exchanging sensitive or proprietary data.
- Centralized architectures of machine learning have certain limitations, including scalability, power consumption and latency issues. FedML can be one effective solution to address these challenges.
- Advancements in enabling technologies, such as differential privacy and blockchain, extend the adoption of FedML by enhancing privacy and facilitating decentralized coordination.
Obstacles
- FedML adoption is hindered by lack of awareness, trust issues, incentive design, collaboration and infrastructure maturity, requiring comprehensive solutions for enterprise integration.
- Enabling FedML requires a complex pipeline that integrates capabilities across DataOps, ModelOps, deployment and continuous tracking/retraining, necessitating a high degree of implementation maturity.
- Creating a new, more accurate and unbiased central model from local model improvements can be nontrivial, as the diversity or overlap between local learners and their data may be hard to assess and may vary greatly.
- Diverse device capabilities cause challenges like stragglers and client dropout, necessitating asynchronous training and adaptive algorithms to handle variability in computation and connectivity.
- FedML’s model updates can leak sensitive information, requiring privacy-enhancing technologies like differential privacy and secure multiparty computation to safeguard against inference attacks.
- Ensuring equitable model performance across diverse clients is challenging due to statistical heterogeneity, necessitating fairness-aware algorithms to address disparities during training and aggregation.
User Recommendations
- Focus on FedML for scenarios where data decentralization is required by privacy regulations, competitive sensitivities or logistical challenges, leveraging its advantage in creating decentralized smart services with diverse user data.
- Use centrally pretrained models to start FedML processes, capturing general patterns and focusing federated rounds on fine-tuning and personalization with local data.
- Begin with small pilot projects or simulations to gain experience and understand algorithm and framework nuances before large-scale deployments.
- Establish governance, trust and incentive mechanisms early to facilitate partnerships with suppliers, customers or peers, enhancing operational efficiency and product offerings through shared insights.
- Implement strategies to prevent excessive divergence from the global model, using central reference models or algorithms like FedProx to maintain cognitive cohesion.
Sample Vendors
Alibaba Group; Devron; Eder Labs; FedML; Google; Intel; NVIDIA; Owkin; WeBank; WithSecure
Gartner Recommended Reading
At the Peak
Multimodal AI
Analysis By: Nick Ingelbrecht, Sushovan Mukhopadhyay, Yogesh Bhatt
Benefit Rating: Transformational
Market Penetration: 1% to 5% of target audience
Maturity: Adolescent
Definition:
Multimodal AI models are trained with multiple types of data (also known as modalities) simultaneously, such as images, video, audio and text. This enables them to create a shared data representation to improve performance in different tasks. At runtime, they can handle more than one modality, either in their inputs, their outputs or both.
Why This Is Important
Multimodal AI adds significant new technology capabilities such as greater accuracy to existing software. It spurs new specialized applications, enables new use cases such as visual question answering of image frames, manufacturing optimization, and fraud detection in banking and finance, and creates new value outcomes. The physical world and the data it generates are inherently multimodal. By integrating and analyzing diverse data sources, a more comprehensive evaluation of complex environments and tasks can be achieved compared with unimodal models, helping users make sense of the world and opening up new avenues for AI applications.
Business Impact
Gartner forecasts that:
- Over the next five years, multimodal AI will become increasingly integral to capability advancement in every application and software product across all industries.
- By 2027, 40% of generative AI (GenAI) solutions will be multimodal (text, images, audio and video), up from 1% in 2023.
Drivers
Multimodal AI adoption will generate cross sector transformational opportunities. Key drivers include:
- A paradigm shift from traditional, linear processes to dynamic, AI-driven systems where humans and machines collaborate seamlessly. And further evolution of agentic AI will involve increasing integration with multimodal AI techniques to handle the complexity and richness of real-world data and tasks.
- Recent AI breakthroughs, particularly in the realm of large language models (LLMs) and vision language models (VLMs), are highly relevant to multimodal AI. These advancements have catalyzed a renaissance in natural language processing and computer vision.
- Intelligent applications, by their nature, are context-rich and designed to adapt to constantly changing scenarios. This makes multimodal AI a crucial component for their development and evolution.
- World models are a significant driver for multimodal AI because they inherently require the ability to process and understand information from various modalities to accurately represent and simulate the complexity of the real world.
- Broader availability of AI/GenAI multimodal models, both proprietary and open-source, lowers the barriers to entry and adoption via AI marketplaces.
- There is a demand for multimodal domain-specialized models in areas such as healthcare, where multimodality extends or enriches use cases.
Obstacles
Multimodal AI is powerful in understanding and processing from various modalities, but faces several primary obstacles to adoption:
- Integrating diverse data types — such as text, images, audio and video — is challenging due to differences in format and time stamps, risking inaccurate interpretations. Multimodal AI models are complex, combining various modality-specific subnetworks, which can obscure transparency and explainability.
- Architectural complexity, increased data volume and the need for data fusion lead to inference latency, hindering reliable operation where immediate decision making is crucial.
- Dataset bias originates from leveraging training datasets like text, images, videos and speech, which may inadvertently reflect societal or cultural biases. This can result in making unfair or inaccurate predictions/decisions.
- Handling sensitive data across modalities increases breach risks and privacy violations. This complicates compliance with regulations like General Data Protection Regulation (GDPR) or Health Insurance Portability and Accountability Act (HIPAA), as multimodal AI exposes new attack surfaces and heightens privacy risks with diverse data types.
User Recommendations
Organizations looking to implement multimodal AI should:
- Identify AI use cases where multimodal AI can enhance business value beyond unimodal AI foundation models.
- Run pilots with off-the-shelf multimodal models to demonstrate not only technical feasibility but the business value.
- Build a strong model evaluation by assessing the quality of relationships between modalities such as comparing generated captions from images to ground-truth labels/descriptions.
- Prioritize building or accessing robust data infrastructure supporting the collection, storage and processing of diverse data types (text, images, audio and video).
- Build or acquire expertise to handle the technical complexities of processing and integrating multimodal data with legacy and existing workflows.
- Create or extend AI governance strategies and policies to address challenges with multimodal datasets and ensure compliance.
- Incorporate multimodality into technology roadmaps and create migration paths for multimodal AI in systems procurement or product development plans.
Sample Vendors
Aimesoft; Google; Hugging Face; Jina AI; Meta; Midjourney; NVIDIA; OpenAI; Stability AI; TwelveLabs
Gartner Recommended Reading
AI Agents in Banking
Analysis By: Jasleen Kaur Sindhu, Lopa Sinha, Sophia Palmstedt
Benefit Rating: High
Market Penetration: 5% to 20% of target audience
Maturity: Emerging
Definition:
AI agents are autonomous or semiautonomous software entities that use AI techniques to perceive, make decisions, take actions and achieve goals in their digital or physical environments. In banking, AI agents offer new opportunities to automate workflows requiring complex decisioning such as fraud investigation, sales, trade financing and lending. Governance, however, remains crucial for mainstream adoption of AI agents in the banking industry.
Why This Is Important
Banks are beginning to explore AI agents but have made less progress than other industries. Unlike GenAI assistants, AI agents have the ability to make decisions and take actions for complex tasks. Banks are currently considering AI agents for internal operations like HR, IT, risk and compliance, and customer service, but they also present opportunities to augment complex processes such as lending and fraud investigation.
Business Impact
- Hyperpersonalized banking experiences that adapt based on customer behavior, spending pattern or lifestage.
- Real-time anomaly investigation with self-improving AI agents trained on evolving fraud patterns.
- Continuous monitoring of compliance postures to flag policy violations in real time.
- Improved employee productivity by removing or reducing time spent on tedious, complex banking processes such as account opening, statement processing, reconciliation and settlement.
Drivers
- Early implementations in banking: Banks are beginning to implement AI agents from simple applications to complex systems. BNY Mellon uses multiagent architecture for tailored sales recommendations, enhancing client interactions. Capital One’s proprietary AI agent tool assists car buyers. Top use cases reported include IT support, call center support and internal administrative tasks like HR.
- GenAI breakthroughs: Advances in reasoning models, large action models (LAMs) and domain-specific small language models enhance the planning of complex actions and banking industry-specific implementations.
- Data-rich landscape: Banking’s extensive customer and financial data and complex decision-making processes offer opportunities for AI agents to automate operations and improve delivery.
- Multimodal understanding: AI’s ability to use vision, audio and language allows for flexible agents, reducing development time and effort for automation.
- Composite AI, including neurosymbolic models: Advances in planning and problem-solving models enable complex AI agents. These agents use diverse AI practices for forecasting, decision making and planning. For instance, Digital Credit Union’s AI agent aids fraud investigation by gathering relevant account information, summarizing details and sharing insights with business units.
Obstacles
- Potential negative impact on employees working alongside AI agents, risking increased job security fears.
- Workflows that can be fully automated by AI agents in banking are not yet feasible. Governance challenges and the need for oversight will likely delay the adoption of fully autonomous AI systems.
- AI agents can suffer from issues like hallucinations, lack of traceability and explainability, which can complicate compliance and lead to incorrect decisions.
- Many banks operate with outdated legacy systems and fragmented data silos. Poor quality data can lead to suboptimal decisions by AI, reducing its effectiveness.
- High implementation costs are an obstacle; developing, training and maintaining AI agents can be expensive.
- Governance and security tools are still emerging that can automate real-time remediation of incorrect decisions made by AI agents in the process workflow. Maintenance and monitoring of underlying models and multiagentic frameworks are still evolving within the banking industry.
User Recommendations
- Incorporate AI agents into strategic planning by understanding their capabilities and potential applications in various environments, considering their increasing autonomy and wide-ranging usability. Ensure that AI agent use cases match value expectations — whether they help in defending, extending or upending.
- Engage vendors and partners early to understand their AI agent initiatives underway.
- Investigate the possibilities of utilizing multiagent systems, collectives of AI agents that can operate both collaboratively and independently, and enhance adaptability and flexibility in response to different tasks and scenarios.
- Establish clear implementation and operational guardrails for usage of AI agents, including legal and ethical guidelines.
- Engage your employees early on to define the future state of work, including process reengineering opportunities.
Sample Vendors
Amazon Web Services; ARQA AI; Auquan; Bud Financial; CrewAI; Druid AI; Google; Microsoft; Norm Ai; Salesforce
Gartner Recommended Reading
AI-Ready Data in Banking
Analysis By: Sudarshana Bhattacharya
Benefit Rating: Transformational
Market Penetration: 1% to 5% of target audience
Maturity: Adolescent
Definition:
AI-ready data is determined through the ability to prove its fitness for specific AI use cases. Proof of readiness comes from the assessment of its representativeness evaluated by its alignment to the use case, support for continuous data qualification, and ensuring data and AI governance. As a result, AI-ready data can only be determined contextually to the AI use case and the AI technique used, which forces new approaches to data management.
Why This Is Important
Journey from AI aspiration to impactful, scaled deployment is often stalled by a significant foundational challenge of achieving AI data readiness to support AI use cases. Based on the 2024 Gartner AI Mandates for the Enterprise Survey results, data availability or quality is a top barrier for AI implementation among banking organizations. Banks not investing in AI-ready data practices can increasingly fail to deliver business values and face AI governance and compliance issues. (The survey was conducted to understand how enterprises are adopting AI and GenAI. It was conducted October through December 2024 among 432 respondents in various countries.)
Business Impact
Banks are expanding AI use cases at scale and deploying bankwide. This necessitates the need to evolve their data management practices and capabilities not only to preserve the classical principles of data management, but also to extend them to AI with AI-ready data. It enables banks with:
- Better regulatory compliance with traceable and unbiased data
- Accurate credit risk assessment, efficient banking assistant, optimized KYC/KYB processes
- Cost-efficient, timely and effective model implementation and maintenance
Drivers
- Data is becoming the main source of differentiation and value from both in-house and pretrained models used in the bank.
- Banks possess a wealth of both structured and unstructured data, which, when made AI-ready, can unlock valuable insights and support AI and specifically GenAI applications.
- Adoption of domain-specific LLMs in banking pivots on AI ready data.
- AI-ready data is foundational to leverage the investments in advanced AI applications like agentic AI and large action models in the bank.
- AI-ready data in the bank is a building block for internal data monetization.
- Protecting sensitive customer data is paramount and requires robust data security measures and strict adherence to data protection regulations like GDPR and CCPA, which AI-ready data frameworks facilitate.
- AI’s rapid advancement challenges data management, requiring augmented techniques suited for AI data needs, enabled by data fabric architecture. These ecosystems benefit from AI-driven approaches, including automated feature engineering and assisted data engineering, leveraging retrieval-augmented generation (RAG) for enhanced data-centric solutions.
Obstacles
- Lack of clarity about the use case and poor data quality is a major challenge to define the scope of AI-ready data in banking.
- Legacy systems and decentralized applications in banks cause data fragmentation and silos, impeding scalable AI solutions due to interoperability and quality.
- Skill gap and lack of executive buy-in for AI observatory tools essential for AI systems, ensuring accurate data interpretation, resolving semantic inconsistencies to monitor data or model drifts.
- AI applications, especially GenAI, require leveraging information from semistructured and unstructured data, which presents integration and processing challenges.
- Ensuring compliance, managing risks and building trust require effective data governance frameworks for AI. The process with model risk management is still maturing in banks and thus takes longer times for deploying models or implementing incremental updates.
- Significant shortage of professionals in ML engineering, UX designing and data architecture to support AI use case deployment, monitoring and enablement within business functions.
User Recommendations
- Prioritize all the data assets in scope for the relevant banking use case and proactively address concerns about perceived risk regarding AI, data governance, privacy and security as the first step of AI-ready data.
- Formalize AI-ready data as a dedicated practice as part of your overall data management strategy.
- Implement active metadata management, data quality, observability, integration and data fabric as foundational components of this strategy.
- Investigate data management tools rich in augmented data management capabilities that can integrate well with AI tools that have created disruptive data-centric AI capabilities, like multimodal data fabric.
- Facilitate traceable, unbiased AI-driven decisions to comply with AI policies, building customer trust through secure data practices.
- Utilize data management expertise, AI engineering, DataOps and MLOps approaches to support the AI observatory.
- Prioritize process documentation with MRM, compliance and legal to streamline deployment and upgrade of AI models in a timely and cost-effective way.
Gartner Recommended Reading
AI TRiSM in Banking
Analysis By: Jasleen Kaur Sindhu, Lopa Sinha, Avivah Litan
Benefit Rating: High
Market Penetration: 5% to 20% of target audience
Maturity: Early mainstream
Definition:
AI trust, risk and security management (TRiSM) comprises four layers of technical capabilities that support enterprise policies for all AI use cases and help assure AI governance, trustworthiness, fairness, safety, reliability, security, privacy and data protection. The top two layers — AI governance and AI runtime inspection and enforcement — are new to AI and are, in part, consolidating into a distinct market segment. The bottom two layers represent traditional technology focused on AI.
Why This Is Important
Banks handle sensitive data and operate under strict regulations, demanding data and AI model explainability, interpretability, transparency, fairness and protection of customer interests. AI TRiSM helps banks leverage AI while adhering to regulations and safeguarding against financial and reputational risks. It also ensures compliance and enhances trust, positioning banks to integrate AI into their operations responsibly.
Business Impact
Adopting AI TRiSM ensures banks can use AI for business results while complying with the existing and new regulatory focus on responsible AI. This is critical as banks adopt AI in a wide range of use cases, including customer experience, credit decisions and financial crime prevention that have heightened compliance requirements. Considerations such as proportionality, efficacy fairness, explainability and accountability are essential to ensure customers are not disadvantaged as AI adoption increases.
Drivers
- Banking is one of the leading adopters of AI and generative AI (GenAI). In the 2025 Gartner CIO and Technology Executive Survey, 77% of banking respondents reported their organization had already deployed AI or would do so in 2025, while 75% indicated the same for GenAI. To fully capture value from AI efforts, banks must prioritize AI TRiSM tools to pursue with confidence revenue generation and innovative AI use cases.
- Banks must ensure AI decisions are explainable, unbiased and nondiscriminatory. For instance, Fair Lending laws in the U.S. require banks to provide fair and uniform services and credit decisions to all customers (see Fair Lending, Office of the Comptroller of the Currency).
- Lack of compliance can lead to severe regulatory and operational risks, resulting in massive fines and reputational harm.
- TRiSM allows for proactive privacy protection of customer information, often used by banks to train and build AI models and to develop new products and solutions. AI TRiSM ensures banks can maintain customer trust when deploying these solutions in the market.
- With the rise of GenAI and AI agents, the risks associated with hosted, cloud-based GenAI applications or with autonomous AI agents are also significant and rapidly evolving. For example, data leakages, adversarial attacks and agency to act without human control.
- Regulations for AI risk management — such as the EU AI Act and other governance frameworks in North America, China and India — are driving businesses to institute measures for managing AI model application risk. Such standards define new compliance requirements for organizations, on top of existing ones, such as those pertaining to privacy protection and model risk assessment (e.g., SR 11-7 defined by the Federal Reserve and Office of the Comptroller of the Currency).
- Rapid development of AI TRiSM technologies and the possibility to make AI risk and security management more robust.
Obstacles
- Navigating diverse and evolving regulatory requirements across jurisdictions is challenging, with AI regulations often lagging behind advancements, creating compliance uncertainty for banks.
- Many AI threats are not fully understood, leading to ineffective risk management.
- Off-the-shelf AI software that are often susceptible to data leakages, are often closed, lacking API support for third-party products to enforce enterprise policies.
- Implementing AI TRiSM requires a multidisciplinary team and a mature operating model for IT, business and functional collaboration, which can be challenging.
- Relying on third-party vendors for TRiSM solutions introduces risks such as vendor lock-in and security vulnerabilities.
- TRiSM solutions for managing autonomous AI agents in runtime environments are still in the beginning stages.
- Rapid AI advances necessitate that banks and TRiSM solution providers continuously update systems and practices.
User Recommendations
- Partner with other functional leaders to set up an organizational task force, including legal and compliance, to collectively manage and govern AI TRiSM efforts. Collectively define acceptable use policies.
- Make an inventory of relevant AI models and applications in production or in use. Implement solutions that enable red teaming and protect data used. Prepare to use different methods for different use cases and components.
- Identify users and implementers of AI/GenAI/agentic AI applications and models, and the role they must play in surfacing and mitigating AI risks.
- Set up a system of record for users to request permission to develop or use AI-based applications, outline the data that will be used and define how it will be used. Use this system to gain periodic user attestation that the system is being used according to preset intentions.
- Revisit, visualize and implement technologies that provide AI TRiSM controls.
Sample Vendors
Airia; Cranium AI; CredoAI; Fairly AI; HiddenLayer; IBM Granite; Mindgard AI; ModelOp; MostlyAI; TrustWise
Gartner Recommended Reading
AI Engineering in Banking
Analysis By: Lopa Sinha, Anthony Mullen, Leinar Ramos, Cuneyd Kaya
Benefit Rating: Transformational
Market Penetration: 5% to 20% of target audience
Maturity: Early mainstream
Definition:
AI engineering is the foundational discipline for enterprise delivery of AI and generative AI (GenAI) solutions at scale. The discipline unifies DataOps, ModelOps and DevOps pipelines to create a coherent development, deployment (hybrid, multicloud, edge) and operationalization framework for AI-based systems. It facilitates delivery of AI and GenAI solutions in banks at scale by building secure, explainable and compliant AI systems that operate within regulatory and risk management frameworks.
Why This Is Important
Banking technology must meet the unique demands for security, regulatory compliance and customer trust. Well-engineered AI systems that ensure scalability, fault tolerance and explainability are critical for high-stakes operations such as loan processing, KYC or personalized customer engagement. These systems must handle sensitive data securely, maintain auditability and integrate seamlessly with legacy systems.
Business Impact
- By designing governable, traceable, and explainable AI systems, AI engineering allows banks to innovate within the regulatory boundaries, meeting expectations of Basel accords and local regulators.
- Banks can scale AI capabilities across business units, such as AML, customer service and IT risk, without reinventing infrastructure or pipeline each time.
- AI engineering fosters early collaboration with business stakeholders, allowing banks accelerated delivery of high-impact use cases.
Drivers
- With global regulators moving toward AI-specific mandates and regulations such as the EU AI Act, banks must demonstrate control over AI system development and design. AI engineering offers a centralized control plane to comply with regulations.
- Banks generate massive datasets from transactions, customer interactions and external sources. AI engineering enables discoverable, composable and reusable data and AI artifacts across the enterprise technical architecture, enabling scaling AI enterprisewide.
- Due to operational complexity, many banks operate in departmental silos. AI engineering unlocks efficiency by offering a horizontal collaboration model that connects data engineers, data scientists, and risk teams, deduplicating efforts and streamlining operation.
- Standardization across data and model pipelines accelerates the delivery of AI solutions irrespective of the approach, retrieval-augmented generation (RAG), chain of thoughts, fine-tuning techniques or models built using diverse AI techniques.
- To gain competitive parity, banks should accelerate GenAI adoption. AI engineering practices, processes and tools accelerate GenAI-specific adaptations including support for prompt engineering, vector DBs and knowledge graphs, architecting and deploying multiagent systems and interactive deployment models.
- AI engineering tools can be subdivided into model-centric and data-centric tools. Terms such as DataOps and LLMOps, or broader terms such as ModelOps and MLOps, are used frequently. However, they are all a subset of AI engineering employing DevOps best practices to operationalize specific portions of the AI development life cycle.
- Finally, AI engineering makes it possible to orchestrate solutions across on-premesis, hybrid, multicloud, edge AI or Internet of Things (IoT).
Obstacles
- In the banking operating model, often data scientists, software engineers, risk and business teams work in isolation, with limited shared accountability for the AI system’s full life cycle.
- Legacy DevOps and data teams may resist as it tackles the end-to-end flow instead of breaking down accountability within departmental and management boundaries. The talent gap in tackling end-to-end flow is also a challenge.
- The rapid expansion of the Ops family has led to an influx of newer, yet marginally nuanced understanding. Without C-level support, AI engineering initiatives get stuck at the infrastructure level and fail to scale bankwide, leading to an inferior AI operation.
- Many core banking systems are still legacy tech stacks, extending AI workflow and thereby AI engineering into core banking can technically not be feasible.
- AI engineering requires simultaneous development of pipelines across domains as well as maintaining maturity across the platform infrastructure.
User Recommendations
- Invest in runtimes observability by instrumenting AI systems with risk scoring, system telemetry and anomaly monitoring.
- Maximize business value from ongoing AI initiatives with an AI engineering practice that streamlines data, models and implementation pipelines.
- Simplify data and analytics pipelines by building composable AI development platforms and build AI-specific toolchains. Promote reuse across lines of businesses.
- Implement policy-as-code into AI toolchain, integrating MRM, audit, compliance and internal controls framework.
- Leverage cloud service provider environments as foundational to build AI engineering. At the same time, rationalize data, analytics and AI portfolios.
- Adopt software delivery best practices by establishing a common understanding of the various aspects of solution, avoiding silos and engaging with business stakeholders.
- Upskill data and platform engineering teams to adopt tools and processes for end-to-end AI life cycle.
Sample Vendors
Amazon Web Services; Anyscale; Dataiku; DataRobot; Domino Data Lab; Google; Microsoft; NVIDIA (OctoAI); Unstructured; Weights & Biases
AI Governance in Banking
Analysis By: Jasleen Kaur Sindhu, Lopa Sinha, Svetlana Sicular
Benefit Rating: High
Market Penetration: 5% to 20% of target audience
Maturity: Emerging
Definition:
AI governance is the process of creating policies, assigning decision rights, and ensuring organizational accountability for risks and decisions for the application and use of AI techniques. Enterprises make decisions on the appropriate, safe use of AI to achieve business outcomes within the governance guardrails that guide the adoption and application of AI techniques. AI governance addresses the predictive and generative nature of AI.
Why This Is Important
AI governance is crucial in banking as AI investments surge, with innovations like GenAI and agentic AI, alongside emerging regulations. Lacking proper governance exposes banks to fines and reputational and financial harm. It can also lead to data privacy breaches and biased algorithms, affecting competitiveness. The 2024 Gartner AI Mandates for the Enterprise Survey shows 72% of AI leaders claim their organizations have clear AI governance structures, yet only 14% view themselves as leaders in AI governance.
Business Impact
AI governance establishes a robust framework for banks to safely adopt and innovate with both current and emerging AI technologies. It supports banks by:
- Ensuring compliance with evolving AI regulations and governance standards
- Effectively managing and mitigating risks throughout the AI life cycle
- Building and maintaining customer trust, explainability and transparency in AI systems
- Enhancing accountability through well-defined governance structures that outline clear roles and responsibilities
Drivers
- Innovation: Banks are rapidly integrating AI technologies, with 77% of banking respondents to the 2025 Gartner CIO and Technology Executive Survey reporting their enterprise had already deployed or would deploy them by the end of 2025. Despite this, 55% of senior banking leaders admit their AI governance practices are fragmented or still emerging, as highlighted in the 2025 Q1 Gartner Financial Services AI and Priority Survey. This leads to a mismatch between the bank's risk appetite and its ambition to pursue AI.
- Regulation: As AI technologies advance, governance is essential for banks to meet regulatory requirements, avoiding penalties, and liabilities related to data privacy and governance regulations like GDPR, DPDPA, CCPA and the EU AI act. Emergence of regional AI regulations drives the need for interoperable governance procedures.
- Risk management: Governance frameworks help manage and mitigate risks associated with AI adoption, ensuring adherence to ethical considerations such as fairness, transparency and accountability. This enables banks to innovate and explore new AI techniques without compromising regulatory requirements.
- Stakeholder expectation/sentiment: Investors, customers and regulators expect banks to demonstrate responsible AI use. Governance frameworks help meet these expectations by providing guidelines on disclosures and communications that reinforce trust.
- Time to market: Effective governance structures reduce inefficiencies and accelerate time to market for AI-enabled innovations, enhancing operational efficiency.
- Governance tools: The emergence of specialized platforms and tools for AI governance, including the emerging concept of AI guardian agents that oversee other AI systems, simplifies and accelerates the adoption of governance practices, providing banks with scalable solutions tailored to their specific needs.
Obstacles
- Model risk management (MRM) expansion: Current practices focus on in-house models, but there is a need to include external vendor models from applications, APIs and marketplaces, which are often hard to track.
- Slow adoption of AI governance tools: Banks often build their own governance processes and are slow to adopt scalable tools like model inventory systems or AI risk dashboards, complicating management as AI evolves.
- Rapid evolution of AI and regulations: The fast-paced development of AI techniques and regulations makes it challenging to fully understand risks and adopt appropriate governance measures.
- Inadequate traditional MRM frameworks: Many frameworks do not address adaptive or self-learning AI, leading banks to separate MLOps governance from GenAI governance.
- Data traceability and quality issues: Poor data traceability, lineage and quality make it difficult for banks to effectively monitor and audit AI systems.
- Investment and resistance challenges: Implementing AI governance requires significant investments in technology and personnel, defining decision rights and roles, and overcoming internal resistance and siloed operations, which can impede consistent framework adoption.
User Recommendations
- Establish comprehensive AI governance frameworks that include robust inventory and monitoring systems to track all models, regardless of origin, ensuring consistent oversight and risk management across both in-house and third-party AI tools.
- Align your governance approach with your pace of AI investments. For an accelerated AI investment pace, governance tools and platforms will be critical. Consider emerging capabilities like AI agents or the concept of AI guardians.
- Establish an AI governance operating model, assigning clear roles, responsibilities and objectives that align with your bank’s specific AI goals and vision.
- Define levels of use-case criticality to focus AI governance on what matters the most and allow freedom for innovation.
- Involve diverse experts (ethics, legal and compliance, security, risk, procurement, etc.) in AI governance to collectively resolve AI issues. Establish and refine processes for making AI-related decisions.
- Ensure that a feedback loop is in place to allow users to report issues and mitigate deficiencies of AI automation.
Sample Vendors
Airia; Cranium; Credo AI; Domino Data Lab; FICO; Holistic AI; IBM; ModelOp; Monitaur; SolasAI
Gartner Recommended Reading
Synthetic Data in Banking
Analysis By: Sudarshana Bhattacharya
Benefit Rating: High
Market Penetration: 5% to 20% of target audience
Maturity: Adolescent
Definition:
Synthetic data is generated artificially rather than from real-world observations, and plays a crucial role for banking. It is used for data anonymization, AI and machine learning development, data sharing, and data monetization. In banking, synthetic data offers PII protection and model development for financial crime prevention, credit scoring, stress testing, backtesting trading algorithms and strategies by generating diverse market scenarios and open banking initiatives.
Why This Is Important
A major problem with AI development today is the burden of obtaining and labeling real-world data, which is a time-consuming and expensive task. Synthetic data remediates this by generating faster and cheaper data, without risking any personally identifiable information (PII) or intellectual property (IP) being exposed or compromised. It supports open banking and data monetization with safe data sharing. This enables banks to accelerate innovation and testing without privacy, security and regulatory concerns.
Business Impact
Banks benefit from using synthetic data as it:
- Provides sufficient volumes of payment data for fraud and real-time payment testing and ML model development.
- Enables accelerated software quality testing and delivery.
- Enhances credit decisioning by plugging gaps in existing data in training.
- Improves existing customer and market intelligence through simulation and scenario planning.
- Promotes safe data sharing for open banking initiatives.
- Supports audio/video content for diverse customer segments.
Drivers
- Obtaining and labeling real-world data to train AI models is expensive and time-intensive. Synthetic data can enrich the training dataset with on-demand, faster and cheaper data.
- Generative adversarial networks (GANs), transformer architectures, variational autoencoders (VAE) and agent-based models are enabling synthetic data generation at a level of quality and precision not seen before.
- Synthetic data is used to stress-test existing fraud models to new attack vectors, or identify new risk typologies that models need to be trained on. Swedbank, for instance, used synthetic data to identify variables and new clusters that correlate with fraud and financial crime, improving the labeling of fraud and suspicious activities.
- Synthetic tabular data can be applied to address gaps, outliers or underrepresented customers such as the financially vulnerable. Synthetic data can be applied to improve data quality, mitigate biases and improve samples of specific segments. It can also support simulations to test new business models such as rollout of real-time payments.
- Synthetically created images, videos, text and speech data can enable banks to generate content that meets the needs of diverse customers and employees. One example of this is reading text aloud for people with vision impairments in a naturalistic simulacrum of a banker’s voice.
- Banks can take advantage of open banking opportunities or data monetization as synthetic data makes it easier and safer to share information with fintechs and other institutions. Nationwide Building Society uses synthetic data to build a secure sandbox environment that allows for quick and secure data sharing.
- R&D labs are expanding the concept of synthetic data to graphs. Synthetically generated graphs will resemble, but not overlap the original. As organizations begin to use graph technology more, we expect this method to mature and drive adoption.
Obstacles
- Synthetic data can exacerbate biases in data, miss natural anomalies, add complexity to development or not contribute any new information.
- The GenAI models that generate synthetic data lack transparency, which means any subsequent AI models trained on that data, in whole or in part, will lack explainability. This is a critical concern for heavily regulated industries like banking.
- Synthetic datasets may look realistic and accurate regardless of whether they’ve accurately captured the underlying real-world environment.
- Bad actors can inject synthetic identity data with fictitious information, resulting in both fraudulent transactions and denial of legitimate ones.
- New business models lack valid historical data with which to validate synthetic data.
- The lack of expertise in synthetic data generation and the challenges associated with implementing advanced techniques like GANs challenges rapid progress, requiring substantial resources and knowledge to overcome.
User Recommendations
- Understand the pros and cons of synthetic data in banking, including development, testing and analysis (see Synthetic Data Will Transform How Banks Operate, Innovate and Compete).
- Ensure synthetic data generated and used adheres to relevant data privacy regulations. New regulations’ compliance requirements may influence the deployment of a synthetic data solution.
- Vet the efficacy of synthetic data relative to real-world data and determine use cases where its privacy persevering attributes can be a powerful differentiator.
- Educate internal stakeholders through training programs on the benefits and limitations of synthetic data and institute guardrails to mitigate challenges such as user skepticism and inadequate data validation.
- Measure the value through ROI and net present value modeling to justify the use of synthetic data (see Use Synthetic Data to Improve Software Quality).
Sample Vendors
Anonos; K2view; MOSTLY AI; NVIDIA (Gretel); Parallel Domain; Rendered.ai; SAS (Hazy); Syntheticus; Tonic.ai; YData
Gartner Recommended Reading
Sliding into the Trough
Edge AI
Analysis By: Eric Goodness
Benefit Rating: High
Market Penetration: 20% to 50% of target audience
Maturity: Adolescent
Definition:
Edge AI is the use of AI techniques embedded in non-IT products (consumer/commercial, industrial), Internet of Things (IoT) endpoints, gateways and edge servers. Capabilities span consumer, commercial and industrial uses, such as mobile devices, autonomous vehicles, enhanced medical diagnostics and streaming analytics. While predominantly focused on AI inference, more sophisticated systems include local training capabilities to optimize models at the edge.
Why This Is Important
Many edge computing use cases are latency-sensitive and data-intensive, and require a level of autonomy and data sovereignty, for local decision intelligence. Such needs have resulted in AI deployment in a wide range of edge computing solutions. Edge AI allows industries, in hazardous and/or highly regulated environments, to apply various AI and machine learning (ML) techniques in operational environments. These applications include distributed, resource-constrained assets, an ability to benefit from improved decision support providing a close-feedback loop, with automated machine tasks, that enables asset reliability.
Business Impact
- Real-time data analysis and decision intelligence.
- Improved operational efficiency, such as visual inspection systems for quality management, output and process efficiency.
- Enhanced customer experience (CX) from AI feedback embedded within products.
- Reduced connectivity costs with fewer data journeys between the edge and cloud.
- Persistent functionality, independent of connectivity.
- Reduced storage demand as only prioritized data is passed to core systems.
- Preserved data privacy at the endpoint.
Drivers
Overall, edge AI has benefited from improvements in the capabilities of AI, including:
- The maturation of MLOps and ModelOps tools and processes that support ease of use across a broader set of features spanning the wider MLOps functions. Initially, many companies came to market with a narrowcast focus on model compression.
- The improved performance of combined ML techniques and an associated increase in data availability (such as time-series data from industrial assets).
Business demand for new and improved outcomes, solely achievable from the use of AI at the edge, include:
- Reducing full-time equivalents with vision-based solutions used for surveillance or inspections.
- Improving manufacturing production quality by automating various processes.
- Optimizing operational processes across industries.
- New approaches to CX, such as personalization on mobile devices or changes in retail from edge-based smart check-out points of sale.
- Privacy-preserving edges.
Additional drivers include:
- An increasing number of users are upgrading legacy systems and infrastructure in “brownfield” environments. By using MLOps platforms, AI software can be hosted within an edge computer or a gateway (aggregation point) or embedded within a product with the requisite compute resources.
- More manufacturers are embedding AI in the endpoint as an element of product servitization. In this architecture, IoT endpoints, such as in automobiles, home appliances and commercial building infrastructure, are capable of running AI models to interpret data captured by the endpoint and drive some of the endpoints’ functions.
- Rising demand for R&D in training has decentralized AI models at the edge for adaptive AI. These emerging solutions are driven by explicit needs such as privacy preservation or the requirement for machines and processes to run in disconnected (from the cloud) scenarios.
Obstacles
- Edge AI is constrained by the limitations of the equipment deployed, such as form factor, power budget, data volume, decision latency and security.
- Systems deploying AI techniques can be nondeterministic. This will impact edge AI applicability in certain use cases, especially where safety and security requirements are primary.
- The autonomy of edge AI-enabled solutions, built on some ML and deep learning techniques, often presents questions of trust, especially where the inferences are not readily explainable. As adaptive AI solutions increase, these issues will increase if identical models deployed to equivalent endpoints begin to evolve diverging behaviors.
- The lack of quality and sufficient data for training is a universal challenge.
- Deep learning in neural networks is a compute-intensive task, often requiring the use of high-performance chips with corresponding high-power budgets. This limits deployment locations where small-form factors and low-power requirements are paramount.
User Recommendations
- Determine whether the use of edge AI provides suitable cost-benefit improvements or whether traditional centralized data analytics and AI methodologies are adequate and scalable.
- Evaluate when to consider AI at the edge versus a centralized solution. Good candidates for edge AI are applications that have high communications costs, are sensitive to latency, require real-time responses or ingest high volumes of data at the edge.
- Assess the different technologies available to support edge AI and the viability of the vendors offering them. Many potential vendors are startups that may have interesting products but limited support capabilities.
- Use edge gateways and servers as the aggregation and filtering points to perform most of the edge AI and analytics functions. Make an exception for compute-intensive endpoints, where AI-based analytics can be performed on the devices themselves.
Sample Vendors
Edge Impulse; IFS (Falkonry); Johnson Controls; Pratexo; Synadia Communications
Gartner Recommended Reading
Emotion AI in Banking
Analysis By: Jonathan Jackson
Benefit Rating: High
Market Penetration: 1% to 5% of target audience
Maturity: Emerging
Definition:
Emotion AI technologies, also called affective computing, use AI and software techniques to analyze the emotional state of a user via computer vision, audio/voice input, sensors and/or software logic. Emotion AI facilitates responses by performing specific, personalized actions to fit the mood of the customer. While emotion AI is not yet a mature technology, it has many potential applications to improve customer experience (CX) at key moments in their financial life.
Why This Is Important
Emotion AI turns human behavioral attributes into data and insights that will significantly impact human-machine interface (HMI). By analyzing customer sentiment, leveraging data most banks already collect, banks can better respond to each customer’s financial context and improve their experiences. Furthermore, the proliferation of multimodals drives combinatorial systems that, for example, summarize text, video or audio, and provide additional emotional insights.
Business Impact
Banks regularly engage customers with major financial decisions. Emotion AI has the potential to personalize these engagements and enable banks to create more empathetic experiences in both human-assisted and digital channels, which may, in turn, improve rates of customer acquisition and retention. As a result, we see the technology expanding to use cases such as risk assessment in lending, personalized advice and conversational assistants. The rise of AI agents is another source of demand.
Drivers
- The emergence of multimodal generative AI (GenAI) allows organizations to train single generative models on multiple types of data (e.g., images, video, audio, numerical data, etc.). This means that emotion AI solutions can leverage larger multimodal datasets of emotion indicators to best achieve personalization of services. Examples include GenAI avatars exploiting multimodal data to adaptively empathize with the user’s emotional state, or use cases in marketing and customer service to generate emotionally compelling campaigns, content and interactions.
- One of the drivers for detecting emotions/states is the need for a system to act more empathetically. This is true even in human-assisted interactions. For example, Gartner found that retail banking customers rated the human assistance they received from their providers as below average in empathizing with their situation (see Opportunities for Generative AI in the Retail Banking Experience).
- As banks advance their digital banking capabilities, emotion AI can help banks deliver more empathetic digital experiences. A person’s daily behavior, communication and decisions are based on emotions — our nonverbal responses in a one-to-one communication are an inseparable element from our dialogues and need to be considered in the HMI concept.
- In some geographies, there is growing regulatory scrutiny on how customers are treated and serviced by banks. Emotion AI can help monitor and review customer conversations with banking staff in places such as the contact center, where voice-based emotion analysis supports multiple use cases. For example, real-time analysis on voice conversations, emotion detection in chat conversations, emotional chatbots and more.
- In creating virtual beings in customer service or advisory services, such as financial planning, emotional responses are a critical element; insights can guide when to involve senior advisors.
Obstacles
- Privacy and regulatory concerns are the main obstacles to rapid enterprise adoption, as laws (such as the EU Artificial Intelligence Act) contain provisions on the use of biometric data.
- The amount and type of personal data collected to enable emotion recognition can appear unsettling to banking customers. Furthermore, inaccurate responses or recommendations could damage the trust customers have with their banks.
- AI models could be biased and lack nuance. When using facial expression analysis, models are likely to be retrained in different geographies to get the system to detect the different nuances present due to different cultural backgrounds.
- Variations could exist across modalities. Certain emotions can be better detected with one technology mode than with another.
- Integration with CRMs, core banking systems and service platforms face challenges around system complexity, data compatibility and privacy concerns, raising maintenance costs.
User Recommendations
- Monitor the development around multimodal GenAI models, as multimodality can increase output accuracy.
- Implement fairness-aware algorithms to continually check for bias in training datasets. Use synthetic data to test models.
- Conduct pre- and postmarket impact assessments to ensure emotion AI technologies promote equitable outcomes (e.g., mortgage eligibility).
- Scope out use cases where better emotion detection would lead to improved CX outcomes in contact centers, such as first-call resolution, cross-selling and customer satisfaction.
- Partner with risk, legal and compliance leaders from the start to outline how the use of emotion recognition complies with organizational risk frameworks.
- Build customer trust by incorporating consent management tools into dashboards.
- Establish communication with customers about the limitations of emotion AI technology for financial journeys where customers are at risk of developing overly optimistic expectations from the technology.
Sample Vendors
Affectiva; Behavioral Signal Technologies; Cogito; DAVI; Grammarly; kama.ai; MorphCast; Soul Machines; Uniphore; Verint Systems
Gartner Recommended Reading
Generative AI in Banking
Analysis By: Jasleen Kaur Sindhu, Moutusi Sau
Benefit Rating: Transformational
Market Penetration: 20% to 50% of target audience
Maturity: Adolescent
Definition:
Generative AI (GenAI) technologies can generate new derived versions of content, strategies, designs and methods by learning from large repositories of original source content. GenAI has profound business impacts, including on content discovery, creation, authenticity and regulations; automation of human work; and customer and employee experiences. In banking, there are demonstrated impacts on workforce productivity, operational efficiencies and risk management.
Why This Is Important
GenAI implementations in banking are increasing, fueled by hype and the availability of banking-specific tools, applications and models. Banks using GenAI report tangible business outcomes, boosting industry confidence. Vendors are innovating rapidly and enhancing GenAI governance, observability, engineering tools and small models for on-premesis deployment. Emerging agentic capabilities, multimodality and reasoning models open new opportunities for GenAI in complex banking workflows.
Business Impact
GenAI in banking is primarily used for internal low-risk tasks like code conversion, document processing and customer service, focusing on productivity and efficiency. Over time, banks are expected to pursue complex processes such as loan origination or integrate GenAI with AI techniques like graph analytics for improved risk management. Multimodality will boost customer service use cases, but costs, ROI, governance and change management remain critical for scaling GenAI deployments.
Drivers
- Seventy-five percent of banking CIOs have deployed or are deploying GenAI in 2025, up from 42% in 2023, according to results of the 2025 Gartner CIO and Technology Executive Survey. This reflects a strong commitment to integrating GenAI into banking operations.
- Banks are now turning to more complex internal tasks, either combining GenAI with other AI techniques or exploring LLM-based agentic capabilities for greater operational efficiency. For example, China Construction Bank uses graph algorithms for identifying suspicious groups, while LLMs describe suspicious characteristics and transaction types (see AI and Generative AI Use Cases in Banking and Investment Services). Capital One launched a proprietary multiagentic AI tool that assists car buyers (see Driving the Future of Car Buying With Agentic AI).
- GenAI offers opportunities to differentiate through innovative banking products and services. For example, Banco da Brasil’s ARI provides recommendations and insights to small business clients using GenAI and analytics (see Olá, Sou a ARI). Bank of Baroda’s GenAI virtual relationship manager, ADITI, enhances the digital banking experience (see Introducing ADITI).
- Ongoing research in model quality, small language models, sophisticated reasoning and multimodality enhances usability and accuracy for specific tasks.
- Open-source LLMs democratize access to GenAI and foster ecosystem innovation.
- Technology vendors are providing tooling, platforms and banking models to standardize and scale GenAI deployments. Advancements in GenAI governance tools and emerging regulations offer a structured path for financial services to explore GenAI innovations.
- Infrastructure innovations are reportedly accelerating, with projects like The Stargate Project’s planned $500 billion investment in AI infrastructure in the U.S. Hyperscalers and enterprises are building supercomputing systems with innovations in computational accelerators, high-speed networks, sovereign cloud and performance-optimized storage. Additionally, innovations like DeepSeek demonstrate efficiency with less advanced chips and lower costs.
Obstacles
- GenAI causes new ethical and societal concerns such as misinformation and deepfakes. Ongoing GenAI research is essential to navigate evolving governance and compliance requirements in banking.
- Hallucinations, bias, a black-box nature and inexperience with a full AI life cycle limit the use of GenAI in customer-facing banking use cases for now.
- Proving GenAI’s value over high-usage costs and energy consumption is challenging, raising affordability and sustainability concerns.
- Banks with low GenAI maturity face difficulty with lack of AI-data readiness, costs and management of technical stack. More advanced banks face issues on change management, operating model and establishing new ways of working that stall efforts for full-scale deployment.
- Most banks rely on outdated IT systems, making the integration of GenAI solutions costly, time-consuming and challenging.
User Recommendations
- Define your GenAI strategy and identify banking-specific use cases in the front office and middle office that align with business objectives and offer tangible results.
- Consider purchased capabilities or partner with consultancies. Consult vendor roadmaps to avoid developing similar solutions in-house.
- Architect GenAI solutions that allow for flexible model selection, as models and data tooling are advancing quickly.
- Use synthetic data for specific use cases to innovate without exposing sensitive data.
- Mitigate GenAI risks by working with legal, procurement, security and fraud experts to build GenAI governance policies and guidelines. Consider both your and your vendors’ ethical and responsible AI practices.
- Optimize the cost and efficiency of AI solutions by employing composite AI approaches to combine GenAI with other AI techniques.
- Invest in AI literacy and upskill your talent for working with GenAI tools and technologies.
Sample Vendors
Amazon Web Services; Bloomberg; DeepSeek; Google; Hapax; IBM; Kasisto; Meta; Microsoft; OpenAI
Gartner Recommended Reading
Climbing the Slope
Knowledge Graphs in Banking
Analysis By: Sudarshana Bhattacharya
Benefit Rating: High
Market Penetration: 5% to 20% of target audience
Maturity: Early mainstream
Definition:
Knowledge graphs (KGs) are machine-readable representations of the physical and digital worlds. They include entities (people, companies, digital assets) and their relationships, which adhere to a graph data model — a network of nodes (vertices) and links (edges/arcs). KGs are relevant to banking by providing an interconnected view of entities within complex financial data and their relationships, enabling extraction of meaningful insights and improving decision-making processes.
Why This Is Important
Knowledge graphs are crucial for modern banking, offering superior data insights by visualizing complex relationships. They enhance AI applications like credit underwriting and anti-money-laundering (AML) investigations, addressing challenges such as data silos and regulatory compliance. By converting structured and unstructured data into KGs, banks improve data accessibility and AI reliability, making them essential in today’s data-driven financial landscape for use cases like financial crime prevention.
Business Impact
KGs can drive significant business impact for banks across a variety of critical use cases:
- Identify patterns and connections between transactions and entities that indicate fraud and support advanced analytics like portfolio news
- Improve assessment of credit risk, market risk and operational risk
- Gain a 360-degree view of customers by integrating data from structured and unstructured sources
- Assist in tracking and understanding complex regulatory requirements
- Strengthen AML detection and personalized product recommendations
Drivers
- Demand for deeper insights: The integration of KGs enhances GenAI and AI implementations by offering a detailed understanding of complex data relationships, driving deeper analytical insights and decision-making processes.
- Potential of utilization of unstructured data: The challenge of leveraging unstructured data from diverse sources is pushing banks toward KGs for easy integration with improved data accessibility for reliable, actionable insights.
- Demand for improved customer experience: Banks are motivated to match the success of KGs in consumer technology, to enable better customer insights and services.
- Complex regulatory compliance: The intricate nature of regulatory requirements, such as ISO 20022, necessitates KGs for streamlined compliance, reducing operational risks and enhancing transparency.
- Increasing data security and privacy: KGs helps to bolster sensitive data protection and implement data deletion controls and regulations like “Right to be Forgotten” by implementing a transparent and auditable framework for data management.
- Integration with legacy systems: KGs facilitate seamless interoperability with existing systems, ensuring smooth transitions and continuity in data management across diverse banking platforms.
- The imperative to break down data silos: Enhanced information flow across banking operations is a significant driver for KG adoption.
- Complementing GenAI enhancement: KGs enhance retrieval-augmented generation (RAG), which are used with GenAI, by grounding information retrieval in enterprise-maintained data sources, improving response explainability, accuracy and contextual relevance in AI-driven applications.
- Drive for improved predictive models: KGs can enhance predictive models in banks, allowing them to analyze data relationships and patterns more effectively. This, in turn, enables banks to have more targeted outreach strategies.
- GraphRAG synergy: The integration of generative AI with KGs through GraphRAG offers precise, contextually rich and explainable answers, surpassing traditional methods and direct LLM usage.
Obstacles
- Early value demonstration: Difficulty proving business value and relevance in initial stages limits stakeholder buy-in and slows adoption.
- Scalability and maintenance: Mature methods for managing KGs as they grow are lacking, including performance reliability, data duplication and data quality preservation.
- Data interoperability: Technical challenges affect the interoperability of internal banking data with external KGs, which is crucial for integrating customers, transactions and regulations.
- Integration with banking systems: Incorporating KGs into existing banking systems is complex, requiring significant adjustments and resource allocation.
- Expertise shortage: In-house KG expertise is scarce and finding qualified third-party providers is challenging. Skills in scalability and optimization are also hard to acquire.
- Stringent security: Ensuring data within KGs complies with banking’s stringent privacy and security requirements is a major concern.
- Resource constraints: High initial costs and significant resources needed for development and maintenance can deter adoption, especially for smaller banks. End-user training is required for adoption of KGs.
User Recommendations
- Create a working group of knowledge graph practitioners and sponsors by assessing the skills of data and analytics (D&A) leaders, practitioners and business domain experts. Factors like use-case requirements, data characteristics, scalability expectations, query flexibilities and domain knowledge of knowledge graphs should be addressed.
- Initiate a pilot focusing on a banking-specific problem, like fraud detection, to evaluate the practical benefit of KGs.
- Start with a foundational KG that captures key data related to a specific banking service or function.
- Collaborate with technology vendors and consultants for their expertise in setting up KGs, but also prioritize building internal expertise to sustain long-term growth.
- Align KG initiatives with the bank’s existing data management policies to ensure they contribute to the overarching strategy without creating isolated data pools.
- Continuously evaluate the effectiveness of KG applications in banking contexts.
Sample Vendors
Cambridge Semantics; eccenca; Fluree; Neo4j; Ontotext; SAS; Squirro; Stardog; TigerGraph
Gartner Recommended Reading
Model Distillation
Analysis By: Birgi Tamersoy, Yogesh Bhatt
Benefit Rating: High
Market Penetration: 5% to 20% of target audience
Maturity: Adolescent
Definition:
Model distillation is a technique where a smaller, simpler model (the “student”) is trained to replicate the behavior of a larger, more complex model (the “teacher”). This is achieved by having the student model learn from the predictions of the teacher model. The goal is to retain the performance of the teacher model while benefiting from the efficiency and reduced resource requirements of the student model.
Why This Is Important
Organizations often face a trade-off between accuracy and efficiency when deploying AI models. Larger models, or model ensembles, tend to have higher accuracy but need more computational resources and may have latency issues. Smaller models are more efficient and require fewer computational resources, but tend to be less accurate when trained from scratch. Model distillation helps by transferring the information content of large models into smaller ones, maintaining performance while improving efficiency.
Business Impact
Model distillation can provide:
- Cost reduction and energy efficiency: Smaller models require fewer computational resources during deployment, resulting in reduced capital investments, operational costs and energy consumption.
- Faster inference times: Efficient models result in quicker inference times, enabling real-time applications and improving user experience.
- Compliance and privacy: Smaller models can be deployed on-premises or on a device, which can help in meeting data privacy regulations.
Drivers
- Developments in foundation models: As foundation models grow larger and more capable, they offer the ability to generate high-quality synthetic data, but at the same time, pose cost challenges for deployment. These developments create both an enabler and a need for model distillation.
- Growth in edge computing and mobile applications: The rise of edge computing and AI-powered mobile applications require AI models that can operate efficiently on devices with limited computational power. Model distillation helps create smaller models that are suitable for deployment on such devices.
- Increased focus on sustainability: As organizations strive to reduce their carbon footprint, there is a growing emphasis on energy-efficient AI solutions. Model distillation contributes to sustainability by reducing the computational resources and energy required for model deployment and operation.
- Advancements in transfer learning: Developments in transfer learning have enhanced the ability to transfer knowledge from large models to smaller ones, improving the effectiveness of model distillation techniques. This allows distilled models to achieve higher accuracy while remaining efficient.
- Regulatory pressure and data privacy: Increasing regulations around data privacy and protection are driving the need for AI models that can be deployed on-premises or on a device, minimizing data transfer and exposure. Model distillation supports compliance by enabling the deployment of efficient models in secure environments.
Obstacles
- Loss of model accuracy: Student models may lose some accuracy compared to teacher models, especially when the student model has significantly fewer parameters compared to the teacher model.
- Complexity of the distillation process: The process may require higher levels of technical expertise, which can be a barrier for some organizations.
- Limited generalization across domains: Distilled models might not generalize well across various domains, limiting their applicability in diverse tasks.
- Dependence on high-quality teacher models: The success of distillation depends on the quality of the teacher model; any biases, inefficiencies or potential intellectual property infringement risks can be passed to the student model.
- Licensing restrictions on teacher models: Some providers impose restrictions on the use of teacher model outputs, prohibiting their use in the training of other models, which can limit the feasibility of distillation efforts.
User Recommendations
- Optimize deployment costs: Use model distillation to reduce computational resources and lower operational expenses, while maintaining model performance.
- Prioritize high-quality teacher models: Ensure that your teacher models are well-aligned for your target task, well-optimized and free from biases. This will improve the quality of distilled models.
- Invest in expertise: Invest in skilled personnel or training programs for implementing effective model distillation. This will ensure that your AI solutions are accurate and efficient.
- Align with regulatory compliance: Use model distillation to create efficient models that can be deployed on-premises or on a device. This approach reduces privacy and safety concerns and eases regulatory compliance.
Sample Vendors
Amazon Web Services; Google; IBM; Microsoft; OpenAI; Predibase; Snorkel AI
Gartner Recommended Reading
Appendixes
See the previous Hype Cycle: Hype Cycle for Banking Data, Analytics and AI, 2024
Hype Cycle Phases, Benefit Ratings and Maturity Levels
Hype Cycle Phases
Phase | Definition |
Innovation Trigger | A breakthrough, public demonstration, product launch or other event generates significant media and industry interest. |
Peak of Inflated Expectations | During this phase of overenthusiasm and unrealistic projections, a flurry of well-publicized activity by technology leaders results in some successes, but more failures, as the innovation is pushed to its limits. The only enterprises making money are conference organizers and content publishers. |
Trough of Disillusionment | Because the innovation does not live up to its overinflated expectations, it rapidly becomes unfashionable. Media interest wanes, except for a few cautionary tales. |
Slope of Enlightenment | Focused experimentation and solid hard work by an increasingly diverse range of organizations lead to a true understanding of the innovation’s applicability, risks and benefits. Commercial off-the-shelf methodologies and tools ease the development process. |
Plateau of Productivity | The real-world benefits of the innovation are demonstrated and accepted. Tools and methodologies are increasingly stable as they enter their second and third generations. Growing numbers of organizations feel comfortable with the reduced level of risk; the rapid growth phase of adoption begins. Approximately 20% of the technology’s target audience has adopted or is adopting the technology as it enters this phase. |
Years to Mainstream Adoption | The time required for the innovation to reach the Plateau of Productivity. |
Source: Gartner (June 2025)
Benefit Ratings
Benefit Rating | Definition |
Transformational | Enables new ways of doing business across industries that will result in major shifts in industry dynamics |
High | Enables new ways of performing horizontal or vertical processes that will result in significantly increased revenue or cost savings for an enterprise |
Moderate | Provides incremental improvements to established processes that will result in increased revenue or cost savings for an enterprise |
Low | Slightly improves processes (for example, improved user experience) that will be difficult to translate into increased revenue or cost savings |
Source: Gartner (July 2025)
Maturity Levels
Maturity Levels | Status | Products/Vendors |
Embryonic | In labs | None |
Emerging | Commercialization by vendors Pilots and deployments by industry leaders | First generation High price Much customization |
Adolescent | Maturing technology capabilities and process understanding Uptake beyond early adopters | Second generation Less customization |
Early mainstream | Proven technology Vendors, technology and adoption rapidly evolving | Third generation More out-of-box methodologies |
Mature mainstream | Robust technology Not much evolution in vendors or technology | Several dominant vendors |
Legacy | Not appropriate for new developments Cost of migration constraints replacement | Maintenance revenue focus |
Obsolete | Rarely used | Used/resale market only |
Source: Gartner (July 2025)
2025 Gartner CIO and Technology Executive Survey. This survey tracked how senior IT leaders worldwide prioritize strategic business, technical and management objectives. It was conducted online from 1 May through 28 June 2024. The survey includes respondents who lead an IT function, with a total of 3,186 CIOs and technology executives participating, including 322 from banking. The survey participants are representative of various geographies, revenue bands and industry sectors, including both public and private organizations. Disclaimer: The results of the survey do not represent global findings or the market as a whole, but reflect the sentiments of the respondents and companies surveyed.
Methodology Statement: P-24039 2025 Gartner CIO and Technology Executive Survey
2025 Gartner CIO and Technology Executive Survey. This survey tracked how senior IT leaders worldwide prioritize strategic business, technical and management objectives. It was conducted online from 1 May through 28 June 2024. The survey includes respondents who lead an IT function, with a total of 3,186 CIOs and technology executives participating. The survey participants are representative of various geographies, revenue bands and industry sectors, including both public and private organizations. Disclaimer: The results of the survey do not represent global findings or the market as a whole, but reflect the sentiments of the respondents and companies surveyed.