Shipping AI into a regulated industry? The slow part isn't the AI

I used to think the model work would be the bottleneck in banking and insurance projects. Lol no. The slowest part of every rollout I've seen is everything around it:

- legal reviewing data flows line by line

- risk teams asking for an audit trail nobody designed yet

- procurement needing 3 rounds of questionnaires

- the "can it run inside our environment" conversation

None of that is bad, tbh. It's what keeps customers safe. But if you plan a 6 week pilot and ignore it, you'll be at week 14 wondering what happened.

What helped on our side was starting with the low-risk work that sits next to the core system, not inside it. We wrote up how that looks for AI agents in banking and financial services if anyone's in the same boat.

For folks in regulated industries: what's the step that ate the most calendar time for you?