Case study · AI frontier lab · Series C
Teaching AI how the real economy works.
A Series C AI frontier lab wanted its models to do genuinely useful work on sustainability — inside the policy, economic and business context that decides what a good answer looks like. This work is under NDA, so what follows is the shape of it, not the specifics.
3 areas
Risk management, strategy and operations
Build ↻ evaluate
A loop for improving model output
New sectors
Opened up for go-to-market
The challenge
Models are fluent. But real decisions in a business sit inside regulation, markets, budgets and the way an organisation actually signs things off. Without that context, AI output can sound right and still be wrong — and enterprise customers notice. The lab needed to understand how real enterprise work happens, and to make its models perform on it.
What I did
- 01Mapped real enterprise workflows. How risk, strategy and operations work actually gets done, reviewed and decided — not how a textbook describes it.
- 02Connected AI to real-world scenarios. Built realistic sustainability situations grounded in policy, economics and commercial reality, the way a client would bring them to a consultant.
- 03Ran a build-and-evaluate loop. Designed tasks, defined what expert-quality output looks like, scored model responses and analysed where they failed — feeding reinforcement learning and model improvement.
- 04Sharpened the context. Pinpointed where models lacked domain understanding, and how to close the gap so the work became more specific, accurate and usable.
- 05Brought new sectors into view. Surfaced industries and use cases the lab hadn't been looking at, and what customers there would need from AI.
What changed
- A stronger go-to-market motion, reaching into sectors the lab hadn't been targeting.
- Better output for end customers: work that reflects how decisions are really made, not just how they're described.
- A repeatable way to keep improving — build, evaluate, learn, and build again.
Why it matters to you
AI is only as useful as the context it understands. I bring over a decade of that context — audit, policy, economics and running real businesses — and I know how to turn it into something a model, or a team, can actually use.
Intelligence is cheap. Context is the edge.