Where CFOs Should Actually Start With AI (a Function-by-Function Map)
Every CFO, controller, and finance team is now asking the same question. Not "should we use AI," that debate is over, but "where do we start." The honest answer is not a slogan about replacing people. It is about removing the repetitive work so the team's hours land on the decisions that actually move the business.
The trouble with most "AI in finance" advice is that it treats the function as one undifferentiated blob. It is not. AI pays off very differently in accounts payable than in the monthly close, and differently again in forecasting. Point it at the wrong stage first and you get a demo that impresses no one and a team that quietly goes back to Excel.
This is a map. Six areas of the finance function, what AI genuinely does in each today, where a human still has to sign, and the one starting point that compounds every month instead of paying off once.
The honest version of the automation list
You have seen the list: accounts payable, commissions, close, forecasting, reporting, controls. All real, all being automated somewhere right now. But a list ranks nothing. What a finance leader needs is the two axes that decide sequence: how much time the area consumes, and how much judgment it requires.
The sweet spot to start is high-volume, low-judgment work. The place to be careful is low-volume, high-judgment work, where a wrong AI output is expensive and hard to catch. Run every candidate through that filter and the sequence sorts itself out.
1. Accounts payable and receivable
This is where the packaged tools are strongest and where you probably should not build anything yourself. Invoice capture, coding, approval routing, matching payments to invoices: high volume, rule-based, low judgment. The market is mature (Ramp, Bill, Brex and the ERP vendors themselves all ship this), and buying beats building.
Start here if your pain is transactional throughput. But be clear-eyed: automating AP does not make you better at finance, it makes you faster at bookkeeping. It frees hours; it does not change the quality of your decisions. For most CFOs it is table stakes, not a differentiator.
Where to stop: payment authorization stays a human control. Let AI prepare and match; keep a person releasing money.
2. The accounting close (the ledger)
Booking accruals, prepaids, reclasses, and producing the trial balance happens inside your accounting system, under accounting rules, with an audit trail your auditors will inspect. This is not the place to let a generative model invent journal entries.
Where AI helps here is around the ledger, not inside it: drafting the narrative for a reconciliation, documenting why an accrual was booked, summarizing what changed since last month, flagging entries that look inconsistent before you post. Treat those as first drafts a human reviews, not postings a machine makes.
If a vendor tells you "AI closes your books," read carefully what it actually automates. Anomaly detection and reconciliation assistance are real and useful. Autonomous journal entries into your system of record are a governance decision, not a feature you switch on.
3. The monthly close as a model operation
Here is the area most "automate the close" pitches skip, and it is where the recurring pain actually lives. Once the books are closed, someone has to get those actuals into the model that drives reporting and forecasting. In a spreadsheet, that means pasting a fresh actuals column, re-pointing formulas, fixing the rows that shifted, and hoping nothing broke silently. Twelve times a year, the model quietly degrades.
This is a structural problem, not an effort problem. In a spreadsheet the logic and the data live in the same cells, so loading new data and preserving the logic are the same fragile operation. The fix is to keep the model's structure separate from its data, so a monthly actuals load is a data operation that leaves the logic untouched.
This is the single highest-leverage place for a CFO to start with AI, and it gets its own detailed treatment: how to use AI for your monthly financial close. It is the starting point that compounds, and the reason is in section 6 below.
4. Forecasting and reforecasting
Static annual budgets are being replaced by rolling, scenario-based forecasts, and AI is a real accelerant here, on one condition: it has to see the structure. Handed a flat grid of numbers, a model can tell you that a variance exists ("marketing is 18% over") but not why, because the driver relationships are invisible in cells. Given an explicit dependency graph, it can trace a variance to its cause and run a forward scenario against the real relationships instead of guessing.
That is the difference between AI narrating your numbers and AI actually reasoning about them. Ask "if this month's overspend continues, where does cash land in Q4," and a model connected to the structure answers against the dependencies; a model reading a static export makes something up.
Use AI to draft variance commentary, rank variances by materiality, and run the scenarios. Keep a human deciding which explanations are real and which are noise. Variance work deserves its own attention: budget vs. actuals, and why finance teams lose hours every month.
5. Reporting and board materials
Drafting the board narrative, generating summary slides, answering "what does this mean for runway" in plain language: AI is genuinely fast at the last mile of reporting. This is often the most visible win and a fine second or third step, because it is where leadership sees the output.
But the board pack is only as trustworthy as the model underneath it. AI that writes a beautiful narrative on top of a fragile spreadsheet is polishing a number nobody can trace. The commentary that goes to the board is something a person signs, not something a model asserts. Draft with AI, own the sign-off yourself.
6. Controls, audit, and the thing that makes all of it safe
The reason the close and the model layer are the right place to start is not that they are the flashiest. It is that they are the substrate everything else sits on. Get the model structured, versioned, and legible, and every other area above gets safer and faster. Leave the model as a pile of fused cells, and AI just adds speed to a fragile base.
This is where the profession is unambiguous. Among senior modellers surveyed, not one would rely on an AI-generated model for a high-stakes decision without independent human review, and the overwhelming majority say the sign-off must stay human. AI adoption in finance is now near-universal, but trust in unsupervised AI output is not, and the gap between those two facts is exactly where a CFO earns or loses credibility.
Three lines hold across the whole function: reconciliations get human confirmation against the system of record, journal entries stay in the accounting system under review, and anything that ships to the board carries a human name. None of that slows a well-structured team down. It is faster to review an AI draft than to write it, as long as you can actually read what the AI did. That readability is a property of your model, not your prompt.
The sequence, in one paragraph
If your pain is transactional volume, start with AP/AR and buy a tool. If your pain is the monthly grind and the quality of your forward view, and for most CFOs it is, start with the model: get the close, the actuals load, and the reforecast running on a structured layer an AI can operate without breaking the logic. Reporting and board automation come next, because they sit on top of that model. Controls are not a step, they are the discipline you apply at every step. Skip the model layer and everything above it is built on sand.
Why the model layer is the starting point that compounds
The AP tool you buy pays off the day you turn it on and never gets better. The model layer is different: it pays off more every month, because the structure persists. Last month's model, its conventions, its history, and its audit trail are still there next month. The AI starts inside your logic instead of reconstructing it from a fresh export every session. A close you can repeat is worth far more than a close you can generate once, and the same is true of the whole function built on it.
This is the split every finance leader is now navigating: the execution layer AI is absorbing, and the judgment layer that gets more valuable as it does. We wrote about that shift in will AI replace the CFO, and about why agents need a structured model to work on in why finance agents need a model layer. The short version: the finance teams pulling ahead are not the ones with the most AI tools. They are the ones whose model is structured enough for AI to work on safely.
The bottom line
Do not start with "AI for finance" as a slogan. Start with the filter: high volume and low judgment first, high judgment and low volume last, and a human on every sign-off. Buy the transactional automation, be conservative in the ledger, and put your real energy into the one place that compounds: a structured model layer where the close loads without breaking, the reforecast runs against real dependencies, and every change stays traceable.
That is the workspace Layerz is built for. The AI operates on a stable structure, so actuals load without rewriting the logic, variances trace to their drivers, and the model you ran last month is still there next month, ready to refresh instead of rebuild. Start where it compounds. See how it works.
