What an AI CFO can’t sign off on

An AI CFO is software that connects to your books and bank and produces financial work on demand: forecasts, scenarios, reports, and answers to questions about your numbers. Some products sold under the name are software alone. Others bundle the software with a human finance lead. Either way, the question worth asking is not only what the tool can produce, but who prepares, reviews and approves the decisions it informs.

The short version: an AI CFO can produce the forecast. It can’t be accountable for the decision.

That isn’t a knock on the software. Many of these tools are genuinely good at analysis, and getting better. But every product page tells you what an AI CFO can do. Very few tell you who owns what happens next. This guide does, and ends with the questions to ask before you sign up.

What is an AI CFO, really?

The job of a CFO contains two different kinds of work.

The first is analysis: gathering the numbers, reconciling them, forecasting cash, flagging what looks off, explaining what changed, and recommending what to do about it. Software can do more and more of this, often faster than a person and at a fraction of the cost. (We sorted that work by when you can actually check the output.)

The second is accountability: owning the call, answering for it when it goes wrong, and deciding which risk the business should take. When a CFO puts their name on a decision, this is what the name means. Someone will stand behind it, and someone will have to explain it if it doesn’t work out.

An AI CFO can take on a growing share of the first kind of work. The second needs a person who answers for the outcome. The name “AI CFO” quietly blurs the two, and that is where owners get into trouble. Not because the analysis is bad, but because it is easy to assume that buying the analysis means someone else is now carrying the decision. For the daily work underneath, see what an AI controller is.

Can an AI CFO tell me whether I can afford a hire?

It can tell you a great deal. Take a common question: “Can I afford to bring on two contractors for a new project?”

Assume a capable tool, connected to your books and bank. It pulls your cash position and recent profit. It models what the two contractors will cost each month. It notices that your largest customer has been paying about ten days late for most of the year, and it moves their next payment accordingly. Then it forecasts that if you hire now, your cash dips to its lowest point around week six, before that customer’s payment lands.

That is good analysis. It found the thing you might have missed, and it put a date on the risk.

Now look at what is left. The tool has made the situation clear, but it hasn’t made it go away:

• Hire now, and you protect the project, but your cash cushion runs thin for several weeks, and any other surprise in that window lands hard.
• Wait, and you protect your cash, but the client may not wait for you, and you could lose the work.

Both paths are reasonable. Which one is right depends on things only you can weigh: how much the project matters to the business, how you feel about running lean, whether that client would really walk, what else you have riding on the next two months. The tool can lay out both paths. It can even recommend one, and its recommendation may be sound. But it isn’t the one who lives with the result if the call turns out wrong.

That is the pattern to expect from any good AI CFO. Better analysis makes the choice clearer. It doesn’t make the choice for you. Before you ask the question, know whose call the answer is.

Should an AI CFO decide whether I take a loan?

The stakes rise with questions like this. Taking on debt is harder to reverse than hiring a contractor, and the cost of getting it wrong lasts longer. The principle is the same, but how the tool handles it tells you more.

It is tempting to think the safe tool is the one that refuses: ask whether to take the loan and it says “talk to a person.” That isn’t automatically a sign of quality. A tool that recommends (“here is what I would do, and why”) can be useful, and one that simply deflects every hard question gives you little.

The real test is three things:

1. It shows its assumptions. What interest rate, what repayment schedule, what revenue it expected, which customers it counted on paying. If you can only see the conclusion and not how it got there, you can’t tell a sound recommendation from a confident guess. The two look identical on the screen. (The same test, put to a vendor from the firm’s side.)
2. It respects the approval boundary you agreed on. Before you rely on it, you should know which decisions it will act on and which it will only advise on. A loan should sit firmly on the advise side, and the tool should behave that way every time, not just when the question is phrased carefully.
3. It makes clear who reviews a decision like this before it’s acted on. If a person comes with the subscription, what exactly do they look at? If no one does, the tool should be honest that the review has to come from somewhere else: you, or a finance professional you already work with.

Ask any AI CFO “should I take this loan?” before you buy, and watch for those three things. One answer will tell you more than the feature list.

Does buying an AI CFO mean someone is taking care of my books?

Not necessarily, and this is the trap most owners don’t see until later.

Every answer an AI CFO gives sits on top of your books. If the books are current and clean, the analysis has something solid to stand on. If they are three months behind, with a pile of uncategorized transactions and an account nobody has reconciled, any answer that relies on those books is behind too. The danger is when the tool doesn’t tell you, and the gap arrives looking just as confident as everything else. As a practising CPA put it to us, your judgement is only as good as your books.

Some AI CFO products include bookkeeping and reconciliation. Many assume it is already being done by someone else. Find out which before you sign:

• Is bookkeeping included, or assumed? If it’s assumed, who on your side is doing it, and how current is it today?
• Who resolves the transactions the tool can’t classify? A good tool will flag a payment it doesn’t recognise or two records that don’t match. Someone still has to decide what they are.
• How does the tool tell you when its data is stale? An answer built on last month’s books should say so. If you can’t tell how fresh the numbers behind an answer are, you can’t tell how much to trust it.

There is a fair point on the other side. Bank activity is often more current than the books, and a tool that reads it can spot a problem, like a large payment that didn’t arrive, before the books catch up. That’s useful early warning. It doesn’t settle what the money was for, how it should be recorded, or whether last month’s numbers were right.

If you buy one: what you still own

Buying an AI CFO can be a good decision. The analysis is real, it’s available when you need it, and for many businesses it is far more than they had before.

Just be clear about which parts stay with you. You own the state of the books the tool reads. You own the boundary for which decisions get reviewed by a person before anyone acts. And unless you have arranged otherwise, you own the decision itself. The questions below make each of those visible before you sign.

Frequently asked questions

What is an AI CFO?

An AI CFO is software that connects to a business’s books and bank and produces financial work on demand: cash forecasts, scenarios, reports and answers to questions about the numbers. Some products are software only; others bundle the software with a human finance lead. In both cases, the software handles analysis. Accountability for decisions has to sit with a person.

Does a human finance professional come with the subscription, or is it software only?

Ask this first, because products sold as an AI CFO vary widely. If a person is included, find out how much of their time you get and what they actually look at. If none is included, the subscription hasn’t supplied that review, so you need to arrange who reviews its recommendations and who approves decisions, whether that’s you, your CPA or a fractional CFO you already work with.

When I ask an AI CFO a decision question, does it show its assumptions?

It should. A useful answer to “can I afford this” or “should I take this loan” shows the figures it used, what it assumed about revenue and timing, and which customers it expected to pay. Without that, a sound recommendation and a confident guess look exactly the same.

Which decisions will it act on, and which go to a person for review?

Agree this before you rely on it. Routine work like categorizing transactions or producing a report can run on its own. Decisions that are costly or hard to reverse, like taking on debt or making a major hire, should be reviewed by a named person before anyone acts on them.

Who is accountable when a decision based on its analysis goes wrong?

The software isn’t. Find out whether anyone on the provider’s side stands behind the analysis, and in what way. Unless that is clearly agreed, the accountability rests with you, the owner who made the call.

Does an AI CFO include bookkeeping?

Not necessarily. Some products include bookkeeping and reconciliation; many assume someone else is already doing it. Ask who keeps the books current and who resolves the transactions the tool can’t classify, because every answer the tool gives depends on them.

How does it tell me when the data behind an answer is out of date?

A good tool says how current the numbers are and flags when an answer relies on books that are behind. If it doesn’t disclose that, you have no way to verify how current any answer is, which is reason enough to ask before you rely on it.

Is an AI CFO the same as having a CFO?

It can do much of the analysis a CFO does, and do it quickly. But a human being included is the start of the question, not the answer. Ask who prepares and stands behind the analysis, who reviews the recommendation, and who approves the decision.

About the author

Nilanjan Raychaudhuri is the founder of AgentLink (Tublian LLC, Columbus, Ohio), which builds controller-layer software sold through CPA and fractional CFO firms. He has spent the past year interviewing practitioners at fractional CPA and CFO practices about how advisory work is actually delivered. Team page

The practitioner observations in this guide come from design-partner conversations. Corrections to nilanjan@agentlink.finance.