As a CIO, how have you approached building a business case for GenAI? How are you making sure that you're accurately forecasting the costs and benefits associated with any use cases you're pursuing?

I will suggest to avoid a full business case until a reasonable proof of value has been delivered and tested in a real business scenario, remember that there is an aspect of adoption (from employees and/or customers) and many readiness aspects to scale up any serious GenAI product. the business case will have so many assumptions that wmight not stand in real life. Start very small, monitor, prove value and adoption and fund incrementally; do not attach to a plan that has "no real legs".

GenAI and AI/ML are a core part of our services and technology stack.
The landscape is still rapidly changing, so our general strategy is to build in some basic abstraction around the specific technology.

Instead of assuming GPT-X or Anthropic 3/3.5 models are going to remain consistent, we have enough abstraction to make switching specific LLM's pretty easy.  We also take some time to experiment with the major public models because they have pros/cons and have changed how they work or what they are great at over the past year.

Perhaps the simplest version is, we're not going "all in" on any one vendor, or tech in the space right now, we're also not trying to compete with the big players and train our own models when fine-tuning, RAG and other concepts work.