Where AI Belongs in the CFO Stack (and Where It Doesn't)

The AI conversation in finance has become noise.

Vendor decks promise to replace your FP&A team. LinkedIn is full of "10 AI tools every CFO needs." Most of it is wrong, or at least incomplete. The result is a generation of finance leaders trying to figure out, in real time, where this technology actually belongs in their stack.

I've used AI tools as part of how I do CFO work, including through the diligence process on a $95M sale. Some uses are real and durable. Others are dangerous in ways that won't show up until a buyer, auditor, or board member asks the right question.

Three places AI belongs in the CFO stack. Three places it doesn't.

Where it works

01Driver-based modeling and scenario analysis

This is the highest-leverage use. AI compresses the time it takes to build, stress, and iterate on a financial model from days to hours. Generating multiple scenarios, running sensitivities, suggesting variables to test... all of it is faster.

The catch: AI is fast at executing the math. Choosing which drivers matter is a CFO judgment call. The model is only as good as the assumptions you put in, and those assumptions still come from understanding the business. AI accelerates the building. It doesn't replace the thinking.

Use it for: building a first draft of a complex model in a fraction of the time, generating downside scenarios you wouldn't have thought to test, stress-testing covenant compliance at multiple revenue levels.

02Narrative drafting and board prep

CFOs spend a non-trivial amount of time turning numbers into words. Variance commentary. Board memos. Investor updates. Lender narratives. AI is genuinely good at first drafts of this kind of writing, especially when given the underlying numbers and a structure.

The catch: the CFO still owns the "so what." AI can describe a 12% revenue miss. It can't decide what to commit to, what message lands with a particular board member who pushed back last quarter, or how to position the recovery story. The judgment is yours.

Use it for: first drafts of variance commentary, board memo structure, investor email templates that you then edit. Saves real hours per board cycle.

03Anomaly detection on transactions

Where AI quietly outperforms humans. Run it across thousands of AR or AP transactions and it will flag duplicate vendor payments, expense miscoding, revenue timing issues, and patterns no human reviewer would catch at scale.

The catch: it surfaces patterns. It doesn't tell you which patterns matter. A CFO still has to triage and decide what's a finding vs. noise.

Use it for: month-end review, pre-audit clean-up, expense audits, vendor concentration analysis, customer-level revenue review.

Where it doesn't belong

04Final-mile forecast accountability

A CFO cannot tell a board "the AI said." The forecast is a personal commitment. AI generates options. The CFO chooses and owns. Same logic applies to capital plans, covenant projections, and any number you put your name behind in a board meeting or a lender call.

This isn't a tooling limitation. It's a governance one. The credibility of the finance function is built on accountability for numbers, not on the speed of producing them.

05Covenant interpretation and compliance

Too much legal nuance, too high a cost of being wrong. AI is improving at reading contracts, but covenant compliance, debt indenture interpretation, and material breach analysis are not places to take the AI answer at face value. This is human plus lawyer territory.

What I do use it for: summarizing key covenant terms across a portfolio of agreements as a starting input. The actual interpretation work happens with counsel.

06Anything tied to your buyer's diligence narrative

In a transaction, the buyer is evaluating two things: the numbers and the management team's command of those numbers. "Our AI generated this" is a confidence killer in the data room.

Use AI to prepare. Don't use it to substitute for understanding. By the time you're in diligence, you should be able to answer any question about your numbers without referencing a tool. If you can't, the buyer will assume the underlying numbers are fragile.

The takeaway

AI is a leverage tool for senior finance, not a replacement for it. The CFOs who get it right use AI to compress execution time on work that doesn't require judgment, and reinvest that time in the work that does... capital strategy, board narrative, transaction prep, the conversations that actually move the business.

The CFOs who get it wrong let AI generate the deliverable, sign their name to it, and find out in diligence or audit that they don't actually understand their own numbers.