CFOs aren’t wrong to be skeptical of AI in finance. They’ve been promised transformation and delivered complexity. But the problem isn’t the technology it’s what the technology is sitting on top of.
There is a pattern playing out in finance organizations across the middle market right now. A CFO approves an AI initiative. The vendor demos are impressive. The use cases are compelling; automated variance analysis, predictive cash flow, AI-assisted close. The implementation begins.
Six months later, the results are underwhelming. The AI tool surfaces insights that don’t match what the team knows to be true. The forecasts it generates require manual correction before anyone will act on them. The close process is marginally faster, but the output is no more trusted than it was before. The CFO is left wondering whether the problem was the vendor, the implementation, or the technology itself. In most cases, it was none of those things. The problem was the foundation.
AI in finance is not a technology problem. It is a data quality, process integrity, and systems configuration problem — with a technology interface on top.
The Skepticism Is Earned — But Directed at the Wrong Target
CFO skepticism about AI is well-documented and entirely rational. Finance leaders have watched automation promises underdeliver for years. ERP implementations that were supposed to eliminate manual work created new categories of manual work. Dashboards that were supposed to provide real-time visibility required weekly maintenance to stay accurate. Planning tools that were supposed to replace spreadsheets became systems that fed into spreadsheets.
So, when AI arrived with a new set of promises, a significant portion of the CFO community responded the way any experienced operator would: with cautious skepticism. Show me the ROI. Show me a real use case that worked at a company like mine. Show me that this isn’t the next thing I’ll have to explain away to my board in 18 months.
That skepticism is healthy. What it sometimes misses are the root cause of why previous technology investments underdelivered and why AI is likely to repeat the same pattern if that root cause is not addressed first.
THE PATTERN
Why every finance technology wave disappoints
ERP systems were supposed to eliminate manual reconciliation. They didn’t, because the data going in was inconsistent. Planning tools were supposed to replace spreadsheets. They didn’t, because the underlying actuals weren’t reliable enough to build on. AI is supposed to surface insight automatically. It won’t deliver reliable output if the data it draws from has the same structural problems that defeated every previous technology. The tool changes. The root cause stays the same.
What AI Actually Needs to Work in Finance
AI in finance is not magic. It is pattern recognition applied to data at scale. Which means its output is only as good as the data it is working with, the processes that generated that data, and the systems that structured and stored it.
When finance leaders describe AI failing them, they are almost always describing one of three underlying conditions:
Dirty or Inconsistent Data
AI cannot reconcile a chart of accounts that has been inconsistently applied, entities that have been coded differently across periods, or actuals that required manual adjustment before they were posted. It will find patterns in that data; they just won’t be the right ones.
Broken or Undocumented Processes
If the close process is different every month (different cutoffs, different accrual timing, different treatment of edge cases) AI has no stable baseline to work from. It will generate outputs that look precise but are not repeatable or trustworthy.
Misconfigured Systems
Most ERP systems are configured for the company that existed at go-live, not the company that exists today. When AI draws on system data that doesn’t reflect the current business model, cost structure, or reporting hierarchy, the results are systematically off in ways that are hard to detect and harder to correct.
None of these are AI problems. They are accounting, process, and systems problems. AI simply makes them more visible and more consequential, because decisions are being made on outputs that feel authoritative but are not.
The CFOs getting real value from AI right now have one thing in common: they fixed the foundation before they deployed the technology.
The Foundation AI Needs Is the Same One Your Finance Function Needs
Here is what makes this moment genuinely interesting: the work required to make AI effective in finance is the same work that makes the finance function itself more effective with or without AI.
A clean, consistent accounting function produces data that humans can trust and AI can learn from. A disciplined FP&A process produces forecasts that are repeatable, driver-based, and structured in ways that AI can augment rather than replace. ERP and EPM systems configured for the business as it operates today produce the kind of structured, consistent data that AI tools require to deliver reliable output.
This is not a complicated idea, but it has an important implication: organizations that try to shortcut the foundation to get to AI faster will get AI that does not work. And organizations that invest in the foundation, for any reason, will find that AI adoption becomes straightforward, because the conditions for success are already in place.
THE PRACTICAL REALITY
What ‘fixing the foundation’ actually means
It means a month-end close that runs on a consistent schedule with consistent methodology so the actuals AI analyzes are comparable period over period. It means a forecasting process built on clearly defined drivers, so AI can model the relationships between inputs and outputs rather than just extrapolating from historical noise. And it means an ERP configured to reflect how the business is organized today so the data AI draws on is structurally sound, not just numerically populated.
The Sequence That Actually Works
The organizations getting real, measurable value from AI in finance today did not start with AI. They started with the accounting foundation, built a reliable planning and analysis layer on top of it, optimized the systems that connect the two and then found that AI adoption was the easy part.
The sequence matters more than the technology selection. It matters more than the vendor. It even matters more than the budget. An AI tool deployed on a well-structured finance function will outperform a more sophisticated AI tool deployed on a broken one, every time.
01
Stabilize the accounting foundation.
Clean data starts with a clean close. Standardized methodology, consistent entity treatment, and timely reconciliations are not just accounting hygiene they are the data infrastructure AI depends on.
02
Build a disciplined FP&A layer.
Driver-based forecasting models create the structured relationships between business inputs and financial outputs that AI is genuinely good at analyzing and extending. Without that structure, AI is just pattern-matching on noise.
03
Optimize the systems that connect them.
An ERP and EPM environment configured for the current business, with clean hierarchies, automated data flows, and current reporting structures, gives AI the consistent, well-labeled data it needs to produce reliable output.
04
Then deploy AI.
At this point, AI adoption is not a transformation initiative. It is an incremental improvement on a foundation that is already working. The wins are real, the results are trustworthy, and the ROI is measurable.
What This Means for Finance Leaders Right Now
If your AI initiative is underperforming, the most useful question is not “which tool should we try next?” It is “what is the state of the foundation that tool is operating on?”
If your close is inconsistent, your actuals are regularly adjusted after posting, your ERP requires manual workarounds, or your FP&A team spends most of its time gathering data rather than analyzing it, AI will not fix those things. It will inherit them.
The good news is that fixing the foundation is not a multi-year transformation program. It is a sequenced set of capability improvements that organizations of almost any size can execute and that deliver value independent of AI, with or without it.
The finance functions that will get the most from AI in the next 12 to 24 months are not the ones that are investing most aggressively in AI tools today. They are the ones that are building the conditions for those tools to work.