After 25 years of Finance Transformation work, the pattern is clear: teams fail at GenAI adoption not because of the technology — but because of three foundational gaps that AI can't fix on its own.
Every Finance leader I talk with is under pressure to "do something with AI." Boards ask about it. Budgets get earmarked for it. Vendors promise it will transform the close, FP&A, and reporting practically overnight. And yet, in engagement after engagement, I see the same story: a promising pilot, genuine early enthusiasm, and then a quiet stall six to nine months in.
The technology, in most of these cases, was never the problem. The models work. The tools are increasingly capable. What fails is the foundation they're supposed to stand on. That pattern isn't just anecdotal: MIT's Project NANDA found that roughly 95 percent of enterprise generative AI pilots fail to produce a measurable P&L impact — and pinned the cause not on model quality, but on the organizational and workflow gaps around it. That tracks with everything I see in the field.
GenAI is only as reliable as the data it uses and reasons over. Most Finance organizations have spent decades accumulating spreadsheets, shadow systems, and tribal knowledge. Layering AI on top of that doesn't clean it up — it amplifies the inconsistency, at scale and with more confidence than the inconsistency deserves.
I've watched teams deploy an AI-assisted reporting tool only to discover it was confidently generating different answers depending on which system or file fed it that week. The fix isn't a better prompt. It's the unglamorous, foundational work of data governance: clear ownership, a single source of truth for key metrics, and validation logic that doesn't rely on a human quietly double-checking everything behind the scenes.
You cannot automate — or meaningfully augment — a process that only exists in someone's head. A shocking number of core Finance processes are undocumented in any real sense; they run because one or two people know the steps, the exceptions, and the workarounds. When you try to bring AI into that picture, you're not augmenting a process. You're guessing at one.
The organizations that succeed with AI adoption almost always did the process documentation work first — not as a separate initiative, but as a deliberate, upfront phase of the AI rollout itself. It's slower at the start. It's dramatically faster from month three onward.
This is the gap I find hardest to talk Finance leaders into acknowledging, because it's the least technical and the most personal. Finance, as a function, has spent generations optimizing for control — precision, audit trails, sign-offs, and a healthy institutional skepticism of anything that moves fast. That instinct isn't wrong. It built the credibility Finance has today.
But AI-enabled work rewards a different posture: iterate, validate, adjust, and trust the system to flag anomalies rather than manually re-checking everything. Teams that don't consciously address this cultural tension end up with a strange hybrid — AI tools bolted onto old control habits — where every AI-generated output still gets manually re-verified line by line. At that point, you haven't gained speed. You've just added a very expensive first draft.
The engagements that produce lasting GenAI adoption share a common sequence, and it's rarely the one vendors pitch:
None of this is a reason to slow down or wait for "perfect" data. It's a reason to sequence the work correctly. The Finance organizations pulling ahead right now aren't the ones with the fanciest AI tools — they're the ones who did the unglamorous groundwork first, so the technology had something solid to stand on.