What does AI consulting cost, and what are you actually paying for?

The market in 2026, by shape of engagement

Prices below are ranges published by firms and pricing guides during 2026, not ours. We scope every engagement on a call, and we publish this so you can check whatever anyone quotes you, including us.

Readiness assessment. A fixed-scope diagnostic that produces a ranked list of opportunities, a data audit, and a plan. Pricing guides put the mid-market band at $10,000 to $25,000 over three to six weeks, with larger firms quoting $15,000 to $60,000 for the same shape. Below $10,000 you are usually buying a slide deck assembled from a questionnaire.

Implementation. Hands-on work inside one area, four to twelve weeks. Fixed fees start around $40,000 and reach $120,000 or more when the scope covers several teams or a data platform that has to be repaired first. Mid-market companies typically spend $90,000 to $300,000 all-in to get a first production AI system live, and most of that is not the model.

Fractional AI leadership. A senior person on retainer who owns strategy, governance, and adoption. Published tiers run from about $5,000 a month for an advisory seat to $10,000 or more for an embedded role; one UK provider lists £3,000 to £8,500 a month by tier, and enterprise-grade offers reach $1,000 an hour with a 100-hour minimum.

Hourly. $150 to $1,000 an hour depending on the tier of firm. Hourly is the worst way to buy this work, because the incentive runs the wrong direction. Buy outcomes with a scope.

What moves the price

Four things, in order of how much they matter:

1. The condition of your data. Nobody’s AI works on bad data. If the assessment finds that customer records live in four systems that disagree, the first engagement is data repair, and it is priced as such.
2. How many teams are in scope. One engineering team is a sprint. Engineering, IT, and finance together is a programme.
3. Whether a harness exists. A team with review gates, tests, and a policy already in place costs a fraction of one starting from “some people use Copilot.”
4. Governance requirements. Regulated industries, or any client who sells to enterprises and has to answer security questionnaires, add policy and documentation work that has to be done properly.

What you should get for it

An assessment should leave you with a written report, a ranked list with the hours each item saves, a refusal list of the things not to do, a 90-day plan with named owners, and an audit of your data and tooling. If any of those is missing, ask why.

An implementation should leave you with a working system, the runbooks to operate it, baselines measured before and after, and a team that can change it without calling the consultant. If the consultant has to stay for it to keep working, you bought a dependency, not a capability.

Four signs you are overpaying

• The proposal does not say what will be measured, or when.
• The first deliverable is a strategy document rather than a look at your actual systems.
• Nobody on the delivery team has run the thing they are advising on inside a real organization.
• The engagement cannot stop early. Good work is structured so you can end it after any phase and still be better off.

Why we do not put prices on this site

We tried it and decided against it. Every engagement we take is scoped to the specific systems and teams involved, and a number on a page would either be wrong for you or padded to be safe. The market ranges above are the honest substitute: check any quote, including ours, against them.