AI amplifies whatever foundation you give it

Most finance AI pilots I see are automating a process nobody fixed first.

That is the part the pilot deck never says. The tool works. The demo is convincing. And six months later the output is wrong in the same ways the manual process was wrong, arriving faster and with more confidence attached to it.

AI amplifies whatever foundation you give it. If you automate a broken process, you get bad results faster.

I ran a regional close where the effort dropped by about 90%. Automation gets the credit for that number. It should not. The work that made it possible came first, and none of it was clever: agreeing what the data actually meant across markets, writing down a process that until then lived in a handful of people's heads, and deciding who owned each control.

Only then was there anything worth automating.

So before an agent or a model goes anywhere near a close, three things have to exist. Data you trust. A process that is documented rather than remembered. Governance with real audit trails and a human checkpoint where it matters.

None of that is exciting. All of it is the difference between a system you can sign off on and one you have to explain afterwards.

None of this is an argument against AI in finance. I have spent the last stretch of my working life building with it, and the ceiling is high. The constraint is almost never the model.

If your close is messy today, AI will not clean it up. It will scale it.

What is the foundation work you had to do before the automation was worth anything?