The foundation is what's missing. Not the model.

Observation

Most organisations evaluate AI on model quality — which vendor, which version, which benchmark. Boards ask whether the technology is good enough. Delivery teams run pilots to prove it can work.

The assumption is structural: better models will close the gap between pilot and production.

Why that’s incomplete

Models draw on world knowledge, not yours. They have your data and your systems, but not how your firm screens a deal, prices an exception, or decides when to walk away.

Upgrading the model does not supply judgement your organisation never wrote down. It widens access to generic reasoning — not to how you work.

The insight

AI capability is constrained by organisational operating logic, not model intelligence.

Why it matters

Strategy: AI investment without encoded operating logic produces tools that work in demos and fail in production — because production is where tacit judgement matters.

Organisation: The bottleneck moves from “can the model do it?” to “does the model know how we do it?” That is an ownership question, not a procurement question.

Operating model: Revenue per employee improves when encoded logic lets a small team run more volume at the same standard — not when a larger model runs faster on the same missing context.

Example

A fund manager reviews four hundred opportunities to execute four. A better model reviews four hundred faster — still applying generic criteria. Encode the firm’s deal criteria first, and the same team reviews four thousand with principals engaged only when judgement is required.

A question for leaders

If your best employee left tomorrow, what judgement would disappear with them — and is any of it written down where an agent could use it?