How much should AI consulting cost?

Nobody can give you a number in an article, and anybody who does is describing their own rate card rather than your project. What is actually answerable is the shape of the deal, and the shape is where most of the money is won or lost.

Four models exist. Pick the wrong one and a fair rate produces a bad outcome.

Fixed fee for a defined piece of work

You agree a scope, a deliverable, and a price. Good when the work is genuinely knowable in advance, which for a diagnostic engagement it usually is: interviews, analysis, a written roadmap, a priority order.

What to watch: fixed fee makes a consultant want to finish, which is fine, and it also makes them want to avoid discovering extra work. If your scope is defined narrowly enough to be priced, ask what happens when the diagnosis reveals something outside it. The good answer is a named process for adding scope, not silence.

Best for the first engagement, and the model we use for the roadmap itself, because you should know your total exposure before you have any evidence about us.

A monthly retainer

You pay for continuing access, usually a set amount of attention per month. Sensible once something is running and needs tuning, sensible for ongoing advisory work, and useful when your situation changes faster than a scoped project can accommodate.

What to watch: retainers decay. The first two months are busy and by month six nobody remembers what it is for, and it renews anyway because cancelling requires someone to make a decision. If you sign one, put a review date in the agreement and honor it.

A day rate

You buy time. Honest, easy to compare, and the least aligned of the four, because the supplier's revenue rises with the hours and yours does not.

Day rates are appropriate for unpredictable work and for expert review. They are a poor fit for a project with a known deliverable, because you are absorbing all the estimation risk in exchange for flexibility you may not need.

What to watch: who is on the clock. A blended rate across a team can hide a lot of junior time.

Outcome-linked pricing

Some part of the fee depends on a result. Attractive in principle and difficult in practice, because it requires both sides to agree on a baseline, a measurement method, and attribution, before anybody has data.

It works when the metric is unambiguous and close to the work. It falls apart when the outcome depends on things the consultant does not control, which is most of the time, and the usual failure is an argument six months in about whether the number moved because of the project or because of the season.

Treat an offer of outcome pricing as a good sign about confidence and read the measurement clause very carefully.

What drives the cost up

More useful than any rate discussion, because these are the levers you control.

How messy your data is. The single biggest multiplier. If the information a system needs lives in three places with different customer identifiers, somebody is reconciling that before anything works.

How many systems have to talk. Cost scales with integrations, not with ambition. One process inside one system is cheap. The same process spanning your CRM, your finance tool, and a spreadsheet somebody maintains by hand is a different project.

Whether a decision-maker is available. Projects stall on unanswered questions, and stalled time gets billed one way or another. This is the cheapest thing on the list to fix and the most commonly ignored.

How much change management is involved. Automating something nobody objects to is straightforward. Automating something that changes how a team is measured is a political project wearing a technical costume.

Whether you need the thing run afterward. Building it and operating it are separate costs, and the second one is recurring. Ask early, because a build price with no run cost attached is an incomplete number.

The comparison that matters

Not the fee. The fee against the value of the decision it informs.

A diagnostic engagement is worth what it saves you from doing wrong. If it stops you spending a year automating the wrong process, it paid for itself before anybody wrote code. If it produces a document nobody acts on, it was expensive at any price.

It is also worth knowing that the rate is the least movable term in any proposal, while several others are negotiable, and that the honest return calculation has to include what recurs monthly rather than only the build.

So when you compare two proposals, compare what you own at the end and whether it is specific enough to act on. Cheap and expensive consulting differ in ways that are not obvious, and most overpaying happens through scope rather than rate.

Why we do not publish figures

Because a number without your context is marketing. Two companies asking for the same roadmap can be a fortnight apart in effort depending on how their data sits, and quoting the smaller one publicly would be dishonest to everybody else.

What we will tell you is the model before anything else: a fixed fee for the roadmap, so your exposure is known, and the roadmap is yours to execute with anyone. How the engagement runs, and if you want the actual figure for your situation, that is a conversation rather than a page.