What are some strategies for managing the costs associated with building and rolling out AI models?
President, CEO, & CDAO in Services (non-Government), Self-employed
(1) Time box how much time is spent trying to reach a particular threshold (e.g., F1 score, R2, etc.). The amount of time dedicated to building the model should be based on a cost to benefit ratio.(2) Apply the law of parsimony - use the simplest model that works. So many times people want to get fancy and build the coolest model, but that doesn't always get you the best results from a model performance and/or an ROI perspective.
(3) This one is probably the most obvious - prioritize which models you plan to build and deploy based on feasibility and costs.
(4) Start with the lowest hanging fruit and build from there. For example, if your organization does not have the bandwidth to process more loans, deals, customers, etc. with the current resources, then focus on AI models that help automate that workload first before you try to build and deploy models that are geared towards increasing demand.
(5) Ensure you have the appropriate guardrails in place to avoid data breaches or ethical issues because that could result in significant unexpected costs to your organization.
Chief Technology Officer in Software, 11 - 50 employees
We go wide first with managing costs a secondary objective to efficiency, solution and scale. As an organisation, where cost containment is of course important, we must first think of the value that a model can bring. If that model allows our business to increase by X or this model making us $Y then the cost associated with building it are relative to the projected benefit. Creativity first! Content you might like
Too Much Hype25%
Amazing Engineering - Great step forward38%
Still lacking judgement17%
Too Dangerous4%
Whats is GPT-3 ? How do i learn more?17%
270 PARTICIPANTS
Providing more accurate service31%
Personalized service46%
Faster service51%
Analyse customer data37%
Provide insights to help businesses better understand their customers34%
Automate customer service task26%
Identify customer sentiment12%
Detect customer issues more quickly13%
123 PARTICIPANTS
CTO in Software, 11 - 50 employees
No, we haven't published corporate guidance establishing guardrails for use of commercial generative AI services.
In our case, we implemented a Medical AI project, and we all agreed that there would not only be the known benefits, but quite a few UNknown benefits of AI, and we added a WAG (Wild Ass Guess) factor to the end result. What we actually found was that we got FAR more insight and savings, both time and monetarily than we had added in the WAG section. We found various patterns that would probably not have been caught by human eyes, and it led to some real steps forward for diagnosis and causal factors in our user/patient group.
I'd try to model it all out with your vendor's help, since they have probably got some pretty good data from existing customers that they can at least point you in the right direction with.