AI can do the valuation. It can’t sign it.
We are entering a genuinely exciting era. Artificial intelligence has become remarkably good at many of the things a business valuation involves — reading financial statements, running calculations, modelling scenarios, researching comparable transactions, drafting. Used well, it is a powerful assistant, and I use modern tools myself to do the work faster. But a business valuation is not, in the end, a calculation. It is a judgement. And that distinction is exactly where AI is both incredibly useful and quietly dangerous. The honest advice is the same one I would give about any powerful tool: move carefully.
Where AI genuinely helps
It would be dishonest to dismiss what these tools can do. AI can crunch years of financial data in seconds, surface patterns, run a valuation under a dozen different scenarios faster than any person, draft the routine parts of a report, and take much of the mechanical load out of the work. That is real, and it is good. It means the slow, repetitive parts of a valuation — the data entry, the recalculation, the first drafts — can be done quickly, leaving more time for the part that actually matters. I have no romantic attachment to doing by hand what a machine can do faster.
The danger is not that AI does these things. It is in mistaking them for the whole job.
Where it goes wrong
It does not know your business. A model works from what it is given, and it fills the gaps with averages. It does not know that 60% of your revenue comes from one customer, that the business only runs because you run it, that last year’s profit was distorted by a one-off contract, or that the salary line hides a family arrangement. These qualitative realities — owner dependency, customer concentration, the story behind the numbers — often move a valuation more than anything in the accounts, and they are precisely what a generic tool cannot see.
Garbage in, garbage out. A valuation is only as good as the inputs and the adjustments behind it. Owner-managed accounts almost always need normalising — adjusting the owner’s salary to a market rate, stripping out personal costs, correcting related-party distortions — before they mean anything. An AI handed the raw figures will produce a confident number built on unadjusted inputs. It will look authoritative. It will be wrong.
It cannot choose the right basis of value. As I have written elsewhere, the same business can have several defensible values depending on the purpose — a sale, a divorce, a SARS matter, a shareholder buy-out each calls for a different basis. Choosing the correct one is a judgement about your situation, not a calculation. A tool that does not understand why you need the value cannot select the basis that makes the answer meaningful.
It is confidently wrong. This is the trap that catches people. AI produces fluent, certain-sounding output even when it is mistaken, and there is no way for a non-specialist to tell a sound number from a plausible-looking error. A free, instant “AI valuation” gives you a figure with no documented basis behind it — a guess with formatting. In a low-stakes context, fine. In a sale, a dispute or a SARS query, relying on it is how owners walk into the most important financial conversation of their lives holding a number they cannot defend.
The one thing AI cannot do: take responsibility
Here is the part no model solves. A valuation that has to stand up — before SARS, in a divorce court, across the table from a buyer’s advisers — needs a named, independent person who produced it, who can explain every assumption, who can be questioned under cross-examination, and who signs their name to the conclusion and stands behind it.
An AI cannot do any of that. It cannot be cross-examined. It cannot exercise independence, because it has no stake to be independent of and no professional reputation to risk. It cannot take responsibility for being wrong. A court, a tax authority or a serious buyer is not asking for a number — they are asking for a defensible opinion from someone accountable for it. That accountability is not a feature you can add to a model. It is the whole point.
My honest position
I am not against these tools — I am transparent about using them. They do the mechanical work faster, which is to your benefit: the fee you pay buys judgement, not data entry. What does not get delegated to a machine is the judgement itself — the normalisation, the choice of basis and method, the reading of the qualitative risks, the independence, and the willingness to sign the conclusion and defend it. That is what you are actually paying a valuer for, and it is exactly the part AI cannot replace.
For business owners — move carefully
By all means use AI to orient yourself. Ask it to explain valuation concepts, to help you understand what drives value, to get a rough feel for the territory before a conversation. It is genuinely useful for that. What it is not is a substitute for a defensible valuation when something important depends on the number. Do not mistake a free, instant figure for an opinion you can rely on in a sale, a dispute, a divorce or a SARS matter — because the moment that number is challenged, the question will be “who produced this, and on what basis?”, and “an AI gave it to me” is not an answer that holds.
If you are weighing up a real decision and want a number you can actually stand behind, that is what an independent valuation is for. Send me the details of your situation — the first conversation is free, and there is no obligation either way.