Which previously outsourced IT functions are you bringing in-house using GenAI tools, and what’s been your actual ROI experience versus initial projections?

We have not done much outsourcing in general, as we have been able to meet most needs internally or through select vendors. On the construction side, we are exploring optimizations, but nothing significant yet in terms of insourcing with GenAI.

I cannot point to an example where GenAI has enabled us to bring outsourced services in-house. In our experience, ROI with GenAI is more about productivity and efficiency gains than direct labor cost savings. In healthcare, we are still in the early stages of leveraging GenAI, ensuring it does not conflict with clinical care.

We have not used AI to insource previously outsourced functions, but we are optimizing and automating certain processes. For example, we use AI for Python development in analytics and have attempted to automate back-office primary care functions, though our experiment with Copilot in this area was unsuccessful. We are also using chatbots to automate some service desk functions. Adoption and literacy with AI tools remain emerging themes, and we recognize the need for further investment in training.

We have automated several use cases with GenAI, including customer care requests such as tracking parcels, invoices, and payments. Content marketing is another area where we have invested heavily in AI automation. Additionally, we leverage AI for dynamic pricing and promotions by analyzing competitor pricing, which is crucial as we are a pure digital player. We have also moved away from external translation vendors by deploying our own trained LLM models for translation as a service across the organization.

Three years ago, we had an outsourced company providing help desk services in the Philippines. As a global company, we leveraged our footprint and insourced this function to our office in Pune, India, still at a lower cost. We also increased Level 2 and Level 3 support for infrastructure, networks and systems, cybersecurity, and applications. Recently, we began using AI bots for Level 1 support. While I was initially hesitant—having disliked bot-driven customer service myself—we are now piloting this to see if it makes a difference. Our experience with Microsoft Copilot over the past year and a half has been mixed: those who use it find it beneficial, but adoption is not universal. We are closely monitoring usage to evaluate its impact.