Why AI adoption in finance stalls, and the cultural fix no vendor will sell you. Three walls, what the 17% do differently, and what to do this week.
Why AI adoption in finance stalls, and the cultural fix no vendor will sell you
Let me give you a number that should embarrass every CFO in the room.
88% of finance leaders believe AI will be the most transformative technology in their field over the next two years. Only 8% say their organisation is “very well prepared” for it.
That is an 80-point gap between belief and readiness. And it has been sitting in survey after survey, year after year, without meaningfully closing.
The State of AI in Finance 2026 report puts it even more starkly: 45% of finance teams are still in limited pilot mode. Only 17% are using AI in core workflows. And — the number that stops me cold every time I read it — 68% of CFOs say they have been slow to adopt because they don't know where to start.
Not because the technology is unavailable. Not because the budgets aren't there. Not because leadership doesn't believe in it. Because they don't know where to start.
I have spent 25 years inside finance functions across IT and ITES, GCCs and analytics. I have watched transformations succeed and quietly die. And I can tell you with certainty: the vendors are selling you the wrong diagnosis.
They will tell you the problem is tools. It isn't. They will tell you it's data. Partly true, but not the whole story. They will tell you it's a skills gap that their training programme will fix for ₹X per seat. It isn't any of these things. Not primarily.
The problem is cultural. And no vendor will sell you the fix, because culture doesn't come in a subscription tier.
What the numbers are actually telling us
Finance has the lowest AI adoption of any business function — lower than marketing, operations, HR and product. This is remarkable, given that finance is one of the most data-intensive functions in any organisation.
Rize's 2026 data from 30,000 users puts finance teams at 3–5 hours of AI tool usage per week per employee. Engineering and creative teams average 8–12. The gap is described as reflecting “differences in workflow compatibility, not interest.”
That framing is polite. Let me be blunter.
Finance lags because finance has built its identity around precision, control and accountability — and AI, at first glance, threatens all three. When a marketing team uses AI to draft campaign copy and it's 80% good, that's a win. When a finance team uses AI to produce a cash flow forecast and it's 80% accurate, that's a governance failure waiting for an audit.
The zero-error culture that makes finance trustworthy is the exact same culture that makes finance slow to change.
That tension is real. It is not irrational. And it is not solved by buying better software.
The three cultural walls I have walked into, in every organisation
Wall 1: the middle-manager identity crisis. Every finance transformation I have led or advised has hit the same obstruction: the senior finance manager who has spent fifteen years mastering the manual process you are now trying to make obsolete.
This person is not a villain. They are often your most capable, most loyal team member. They know every exception, every quirk in the data, every reason why the “standard” approach breaks in Q3. They are the institutional memory of your finance function.
And they are terrified. Not of AI, exactly. Of irrelevance. Of a future where the expertise they spent two decades building no longer commands the respect it once did.
The Hackett Group's 2026 Finance Key Issues Study found that finance workloads are projected to rise 3.2% this year even as headcount falls 2.1% and budgets drop 1.7%. That 5.3% productivity gap is being closed with technology. Every finance professional in your team can feel that pressure, even if nobody has said it out loud.
When you roll out an AI pilot without addressing this fear directly, what happens? The pilot gets technically “implemented.” The tool gets used for low-stakes tasks — formatting reports, summarising documents, writing email drafts. The core workflows stay manual. Six months later, leadership asks why adoption is low. The answer, buried in the data: nobody wanted to be the person who automated themselves out of the room.
The fix: before you launch any AI initiative, have the honest conversation about what changes and what doesn't. Map out explicitly which tasks are moving to AI and what the humans in those roles will now own instead. People don't resist change. They resist uncertainty. Give them a clear picture of the future that includes them.
Wall 2: the pilot that proves nothing. Here is a pattern I have seen repeat across organisations of every size.
A CFO attends a conference. Comes back energised. Asks the finance team to pilot an AI tool for, say, variance analysis. The team runs the pilot for sixty days. It produces some interesting outputs. Nobody quite knows how to evaluate whether it worked. The pilot ends. The learnings get written into a slide. The slide gets presented to the leadership team. Everybody nods. Nothing changes.
Six months later, a different CFO attends a different conference. The cycle repeats.
This is not a technology problem. It is a design problem. The pilot was never set up to succeed because nobody defined what success looked like before it started.
The Thomson Reuters Institute's 2025 research found that firms with a documented AI strategy are twice as likely to see AI-driven revenue growth as firms without one. Not twice as likely to use more tools. Twice as likely to see an actual return.
A documented strategy means: here is the specific workflow we are changing, here is the metric we are improving, here is how we will measure it in 90 days, and here is who is accountable for the outcome. Without that, a pilot is just an expensive proof of concept that proves nothing.
The fix: stop running pilots. Start running experiments with clear hypotheses. “We believe that AI-assisted three-way matching will reduce invoice processing time by 40% and the error rate by 25% in 60 days. Here is our baseline. Here is our measurement method. Here is our decision rule for scaling or stopping.” That is a business case. A pilot is just a test drive with no intention to buy.
Wall 3: the governance vacuum. Finance functions are governed environments. Every number has a trail. Every process has a control. Every exception has an approver. AI, as most organisations have deployed it, exists outside all of that.
I have walked into finance teams where an analyst is using ChatGPT to help build financial models, saving hours of work. When I ask whether those AI-assisted outputs are disclosed in the review process, whether the assumptions have been independently validated, and whether the audit trail documents which elements were AI-generated, the answer is almost always the same: we haven't really figured that out yet.
This is not a failure of individuals. It is a failure of governance frameworks to keep pace with tool adoption.
And here is the consequence: when something goes wrong — a model error, a misclassified input, a forecast that proves wildly off — the organisation doesn't know whether the failure came from the AI, the data, the human reviewing it, or the process it was embedded in. You can't fix what you can't locate.
The irony is that finance teams are more worried about AI errors than they should be, and less worried about AI governance than they should be. The technology is increasingly reliable. The question of who owns the output, who validates it, and what happens when it's wrong is dangerously underexplored.
The fix: build your AI governance framework before you scale any AI deployment. It does not need to be elaborate. It needs to answer four questions.
Four questions to answer before you scale
- Who approved this AI tool for this use case?
- What human validation is required before the output is used in a decision?
- Where is the output documented and traceable?
- What is the escalation path when the AI is wrong?
Put them in writing. Assign ownership. Then scale.
What separates the 17% from the 83%
I've had the privilege of seeing both ends of this. The finance functions that are genuinely AI-native in 2026 — not just AI-interested — share a set of behaviours that have nothing to do with budget or technology stack.
- They started with a problem, not a tool
They identified a specific, painful, high-stakes workflow — often month-end close, FX exposure reporting or intercompany reconciliation — and built AI into the solution for that problem. They did not buy a platform and then search for use cases.
- They made AI a finance outcome, not an IT initiative
The worst AI deployments I have seen are ones where finance said “yes” to a tool that IT bought. The best are ones where a senior finance leader owned the outcome, defined the metrics and reported progress to the CFO quarterly.
- They trained for judgement, not just usage
The most important skill in an AI-enabled finance function is not prompt engineering. It is knowing when to trust the output and when to question it. That is a financial judgement skill. It requires understanding what the model is doing well enough to catch it when it goes wrong, and it cannot be learned in a two-hour vendor onboarding session.
- They were transparent about the direction of travel
The 17% didn't pretend AI wasn't going to change roles. They said: here is what is changing, here is what we are building toward, and here is how we are investing in your development so you grow with it rather than against it.
The conversation no vendor will start
The software companies selling AI to finance functions have a product to move. Their pitch will always emphasise ease of adoption, speed to value and return on investment. Some of those claims are real. Many are optimistic.
What they will not tell you is that the tool is the easy part.
They will not tell you that your biggest obstacle is a 52-year-old controller who has been running month-end close the same way for twelve years and whose cooperation you absolutely cannot afford to lose. They will not tell you that your pilot failed because you never defined success. They will not tell you that your governance framework needs to exist before you scale, not after.
Those conversations are uncomfortable. They are also the only ones worth having.
The 80-point gap between believing in AI and being ready for it will not close because better models are released. It will close when CFOs stop treating AI adoption as a technology rollout and start treating it as an organisational change programme — with the same rigour, the same investment in people, and the same CEO-level sponsorship they would give any other transformation that touches every corner of their finance function.
The technology is ready. The question is whether the organisation is.
Three things to do this week, not this quarter
- Have the conversation with your top three senior finance managers about what AI means for their roles. Not a presentation. A conversation. Ask them what worries them. Listen before you explain.
- Diagnose your most recent failed or stalled AI pilot honestly. Was the problem the tool, or was it that nobody owned the outcome, nobody defined success, and nobody was accountable when it drifted?
- Write down the one workflow in your finance function that, if AI-enabled, would create the most material improvement in quality, speed or risk. Not ten workflows. One. Start there.
The gap between 63% and 17% is not a technology problem waiting for a technology solution.
It is a leadership problem waiting for a leader.
Sources
The 88% and 8% survey figures, the Rize usage data and the Thomson Reuters Institute finding are as cited by the author in the original article.
- ChatFin, ‘State of AI in Finance 2026: adoption, workforce and the numbers that matter’: about 45% of finance teams in limited pilot mode, about 17% using AI in core workflows, and 68% of CFOs slow to adopt because they don't know where to start.
- The Hackett Group, 2026 Finance Key Issues Study (press release, 19 March 2026): finance workloads projected to rise 3.2% in 2026 as headcount falls 2.1% and budgets fall 1.7%, a 5.3% productivity gap.
First published on LinkedIn on 10 June 2026. Read the original and join the discussion
