What is the real ROI on AI consulting?

Most ROI arithmetic in this field is constructed to produce a good answer. Hours saved multiplied by an assumed wage, against the build fee, ignoring everything recurring. That calculation always looks excellent and means very little.

Here is the version that survives scrutiny.

Get both sides of the fraction right

Costs, all of them. The build fee, plus data reconciliation work, plus your own people's time during the project, plus everything recurring: model usage that scales with volume, platform subscriptions, whoever maintains and tunes it. Over two years, not one. The full cost breakdown.

Returns, conservatively. Only what you can trace to the work, measured against a baseline you took beforehand.

The two costs everyone leaves out are your own team's time and the ongoing run cost. The first is real even though it is uninvoiced. The second is the one that turns an impressive year-one number into an ordinary two-year number.

The three return types, in descending reliability

Work that now happens and did not before. The most reliable and the largest in most companies. Follow-up nobody had time for, renewals that used to lapse, inquiries arriving outside working hours, dormant customers contacted.

This is easy to value honestly because the comparison is against zero. Count what came back and what it was worth. No assumptions about wage rates required.

Throughput at the same headcount. More handled without hiring. Straightforward if you have a volume baseline, and it is the number to lead with when presenting internally.

Time released. The one everybody uses and the weakest, because hours saved are only worth something if the freed time went somewhere. If a person's afternoon opened up and they now do higher-value work, that is a real return and you should describe the higher-value work. If nothing visibly changed, the hours were absorbed and the saving was notional.

Be strict here. This is where inflated ROI claims are manufactured.

What the diagnosis specifically returns

Consulting is a decision, so its return is measured in decisions.

The largest one is usually avoided cost: not building the thing you were about to build. A company that was going to automate a low-frequency process and instead automates a high-frequency one has gained the difference, and that difference frequently exceeds the whole fee before anything is constructed.

Harder to feel than a build, because nothing visible happens. The way to capture it is to write down what you were going to do before the engagement, then compare. If the roadmap changed your plan, that delta is the return.

If it did not change your plan, you did not need the diagnosis, and that is worth knowing too. The readiness test.

The negative return nobody models

Building the wrong thing costs more than its price.

You lose the fee, and you lose internal credibility, and you lose the appetite to try again. That last one is the expensive part: the conclusion inside the company becomes "we tried AI and it did not work for us", and reopening that takes a year or two.

Which is the actual case for spending on judgment before construction. You are buying insurance against a cost that never appears in anybody's ROI model.

Returns that cannot be measured, and should not be faked

Name these in words rather than inventing figures.

The customer relationship retained for years because you responded once at the right moment. Referrals from being available when it mattered. Work happening consistently rather than depending on somebody remembering. Not carrying an unfinished list around in your head all evening.

All real, none of them quantifiable without fabrication. A supplier who puts a number on these is guessing, and the guess will be flattering.

Attribution, honestly

Keep the metric close to the work. If you automated inbound response, measure inbound response, not quarterly revenue.

And note what else changed in the period, because you probably also hired somebody, ran a promotion, or had a strange season. A supplier presenting a revenue increase next to their invoice is inviting a conclusion the data cannot support. Measurement design.

The honest summary

For most small and mid-sized companies, the reliable return is concentrated in work that was not happening at all, and it shows up in weeks rather than quarters. The time-savings story is real but softer than it is usually sold.

And the return on the diagnosis is mostly the mistake you did not make, which is genuine, unglamorous, and impossible to put on a chart.

We attach an estimate to each recommendation in the roadmap so there is something to check the outcome against, including what each item costs to run rather than only to build, because the run cost sometimes changes the recommendation to "buy something existing" or "leave this alone". How the engagement works.