Most of the AI proposals that reach a leader’s desk look reasonable. They have a plausible use case, a confident vendor, and a slide that ends with a big number. The hard part is not saying yes or no to AI in general. It is telling the investments that will actually move your business from the ones that will look busy, cost money every month, and change nothing. The difference tends to follow a few recognizable patterns.
Where AI reliably pays off
AI earns its keep when the work is high in volume and moderate in stakes, and when a fast, good answer beats a slow, perfect one.
The clearest wins share a shape. There is a lot of the task. It repeats. It involves judgment that is real but not rare, the kind an experienced person makes many times a day without much drama. Think of sorting and routing large volumes of requests, reading through more text or data than any team could get to by hand, or pulling the three relevant facts out of a long document. In these cases the machine does not need to be flawless. It needs to be quick, consistent, and roughly as good as a tired human on a busy afternoon.
Drafting is another honest fit, as long as a person stays in the loop. AI is good at producing a first version that someone then checks, corrects, and signs off. The draft is not the deliverable; it is a head start. The value comes from the time saved on the blank page, not from removing the human who is accountable for the result.
The pattern underneath all of these is the same. The cost of a small error is low, the volume is high, and speed has real worth. When those three line up, AI tends to pay for itself and keep paying.
Where it quietly burns money
The failures are quieter, which is what makes them expensive. Nobody announces that a tool is being ignored.
The first trap is solving a problem you do not have. A capability looks impressive, so a use is invented to justify buying it. The work gets done, the tool exists, and no number in the business moves because nothing was broken to begin with.
The second is asking AI for precision and accountability it cannot promise. Some tasks require the answer to be right, every time, with someone who can stand behind it: regulated decisions, final financial figures, anything where a confident wrong answer causes real harm. AI can assist here, but if you need a guarantee, you are buying the wrong thing.
The third is the thin problem. Often the honest answer is that a simple rule, a report, or a better spreadsheet would do the job, cost almost nothing, and never surprise you. Reaching for AI on these is not ambition. It is expensive over-engineering.
Then there is the tool nobody adopts. It works, it demos well, and it sits idle because it did not fit how people actually do their jobs. And finally the slow leak: the ongoing cost of running and maintaining the thing quietly outgrows the value it delivers, and no one is watching the two lines cross.
Two questions that separate them
Before backing any AI investment, a leader can get a long way with two plain questions.
First: what specifically gets better, and how would we notice? If the answer is a capability rather than an outcome, or if no one can say which number should move, the project has no way to prove its worth. Clarity before code.
Second: what is the cheapest thing that would solve this, and why isn’t it enough? If a rule, a report, or a small process change would do, that is usually the right answer. AI should have to earn its place against the simpler option, not win by default.
It is worth being honest that a good deal of proposed AI work does not survive these questions, and that is a feature, not a failure. Saying no to the cases where it does not belong is how you free up the budget and attention for the few where it genuinely pays.
That sorting is exactly what a short, fixed-scope Business Diagnostic is for: a clear, prioritized view of where AI is worth your money, and where it would quietly drain it.