How AI fits together

There's so much talk about AI and where it can be used in a business. I wanted a quick visual sketch that could help me think it through from the top — what business outcome are you trying to achieve — through to capability and the intelligence behind it, and finally the value delivered. This is my first attempt and I'm still pondering how to make it better.

I wanted an easy to understand map, so when I think about solving a business problem I could quickly reference the capability. A retail bank wanting to reduce cost to serve in the contact centre, and the same bank wanting to catch fraud, will often sit on the same AI roadmap. They need completely different builds. One needs retrieve and remember — find the right policy across hundreds of documents, hold context through a conversation. That's a language model with RAG behind it. The other needs predict — score a transaction against patterns in structured data. Throwing a bigger or better language model at fraud scoring won't beat a machine learning model, because it's the wrong kind of tool for structured transaction data.

Or a large retailer with thousands of products. Helping someone work out what they need to tile a bathroom is search and reason. Forecasting stock by store is predict. Different capabilities, different intelligence, different builds — and you only see that if you stop at the capability layer on the way down.

It matters because the two need different things to work. The fraud model needs labelled historical data, a data scientist, retraining as the patterns shift, and a risk owner to sign off the thresholds. The contact centre one needs its policies and procedures to be current and findable, retrieval design, evaluation, and guardrails on what it can say to a customer. Different people, different data, different definitions of working. Planned as one AI project with one team and one budget, one of them will disappoint.