The AI Consulting Services Market, Counted Not Forecast

Eight firms. That is how many of the 191 verified agencies in this directory tell a buyer they build AI chatbots, which makes chatbot work the smallest of the nine service categories on the site and leaves it behind research-heavy machine learning and data science, a category with 84. Anyone reading the published estimates of the AI consulting services market would expect the reverse, because the chatbot is the service that gets demonstrated at conferences and written into procurement decks. The supply does not agree, and the supply is the half of this market nobody counts.

Every figure below comes from the listings on this site, read out of the database on 7 October 2026 and restricted to the 191 agencies a researcher confirmed against the firm's own website. The 179 seeded placeholder rows that pad the directory are excluded throughout. The method section at the end says exactly what that leaves and where it is weakest, because a number without a method is a decoration.

Forecasts measure the cheque, not the firm

The research houses that size this market do serious work, and they all do the same work: they estimate what buyers will spend. Organisations including the Stanford Institute for Human-Centered AI, through its annual AI Index, and the OECD AI Policy Observatory publish on adoption and investment, and the US Census Bureau's Business Trends and Outlook Survey asks firms directly whether they use AI in production. Those are demand-side instruments. They tell a buyer how much company money is moving and how fast adoption is spreading.

They do not tell a buyer what the person on the other side of the table actually sells. A spend forecast for the AI consulting services market is compatible with a market of forty enormous integrators and with a market of four thousand three-person shops, and those two markets price differently, fail differently and need completely different contracts. Counting firms rather than budgets answers a narrower question with more practical use: when you open a shortlist, what is on it?

Nine categories, and the ranking runs backwards

This site divides the AI consulting services market into nine categories, and each listing carries one or more of them, assigned from what the agency itself publishes rather than from what it bids on. Across the 191 verified firms the distribution comes out like this.

The shape of the AI consulting services market is visible in that ranking. The top of the table is where the difficult, slow, data-dependent work sits, and the bottom is where the demonstrable work sits. That inversion has a plain commercial explanation. A chatbot has become something a competent in-house developer assembles from a hosted model and an afternoon of documentation, so the firms that used to sell it have either moved up into integration and custom builds or stopped advertising it as a headline service. The agencies describing themselves as chatbot specialists in 2026 are mostly the ones who have not noticed that the floor moved.

Read the other direction, the two leading categories are the two in which a buyer is least able to judge the work before paying for it. That is not an accusation of bad faith. It is a description of where the money concentrates, and it is the reason the rest of this analysis is worth doing.

Four firms out of a hundred and forty-eight do one thing

Of the 191 verified firms, 148 publish at least one service category. Among those, exactly four claim a single category. Fifty-two claim two, and 92 claim three or more, with a mean of 2.59 categories per firm.

The specialist agency, in other words, has very nearly stopped existing as a business model in this market. The generalist is the norm by an enormous margin. A buyer should treat a three-category claim as the default rather than as evidence of range, and should treat a genuine single-category firm as the unusual case that it is, worth a second look precisely because it has decided to be legible.

There is a reasonable defence of the generalists, and it is worth stating. AI work does not divide cleanly: a forecasting model needs integration to be useful, and an integration project usually needs modelling work before it starts. Claiming three adjacent categories can be an honest description of one coherent practice. The problem is that it is indistinguishable, from the outside, from claiming three categories to clear three different filters. Sorting the two apart is what the first-time buyer's checklist exists to do.

The advice half and the build half have come apart

Fifty-four verified firms offer strategy consulting. Seventy offer integration. Only 21 offer both, which leaves 33 firms selling a plan with no stated capacity to execute it.

That ratio matters more than any single category total, because a strategy engagement ends in a document and a document changes nothing by itself. Roughly three in five of the firms that will write you an AI roadmap have not told you they can build what the roadmap specifies. Some of them are deliberate about it and say so, and a clean separation between adviser and builder is a defensible structure with real advantages in procurement. The rest have simply never been asked the question. What the roadmap leaves out goes through the handover in detail, and the short version is that the receiving team is the thing nobody scopes.

Thirty-seven firms will tell you a price

Of the 191 verified listings, 37 publish an hourly rate and 154 publish nothing. Among those that do, the rates run from $50 to $500 an hour and the published minimums average $97.

Four fifths of this market treats price as a conversation rather than a number, and the spread among the fifth that does publish is tenfold, which tells a buyer that the rate card is close to meaningless as a comparison tool. Two firms at $150 an hour can differ by a factor of three in what an hour produces. The useful comparison is not the rate but the scope of the first paid phase and what the buyer holds at the end of it, which is the argument of what you actually own when an engagement ends.

The directory does not publish an average rate for any city or category, and the reason is this distribution. An average of 37 self-selected numbers spanning an order of magnitude would be a figure with no defensible meaning, and buyers would quote it in negotiations.

What a buyer should do with these numbers

Does a three-category agency know less than a specialist?

Not necessarily, and the base rate says stop treating it as a signal either way. With 92 of 148 firms claiming three or more categories, breadth carries almost no information. Ask instead which single category produced the firm's last three engagements, and compare that answer to the list on its own website.

Why will so few agencies publish a rate?

Because scope dominates rate in this work and most firms know it. The honest version of the refusal is that an hourly figure invites a comparison that misleads the buyer. The less honest version is that the number moves with how well funded the buyer looks. A firm that declines to publish a rate and then gives you a clear fixed price for a scoped first phase is behaving well; one that will not price a small, well-defined piece of work at all is telling you something.

Is a local agency worth paying more for?

For the discovery phase and for staff rollout, physical presence earns its cost. For the modelling and engineering in between, it is close to irrelevant. The 191 verified firms sit in 105 distinct cities, so for most buyers the shortlist is national whether they intended it or not. The case for and against proximity splits the engagement into the parts that care and the parts that do not.

What does it mean when a firm claims strategy but not integration?

It means you need to know who builds, before you sign. Ask the firm to name the partners it hands work to and whether it has done so with the client in the room. Thirty-three firms in this count are in that position, and the good ones answer immediately.

Should the smallest categories be avoided?

The opposite, usually. Eleven firms in AI search and nine in voice work means a short list and a real chance of reaching a principal rather than an account manager. A thin category is a buying advantage as long as the work is genuinely the firm's main line, which in a nine-firm category is easy to verify by reading three pages of its site.

Method, and where this count is weak

This is a supply-side census of the AI consulting services market rather than a forecast of it. The population is the 191 listings on aiagencysearch.com that a researcher confirmed against the agency's own website, as of 7 October 2026. Every service category is assigned from what the firm publishes about itself. No rating, price, client name or headcount is ever inferred, and where a firm does not publish a figure the field stays empty rather than being filled with an estimate. The 179 seeded placeholder rows in the directory are excluded from every number above.

Three weaknesses are worth stating plainly, because they bound what these figures can support.

• The geography is a research artefact, not a market map. The leading cities in the verified set are Denver with 13 firms, Atlanta with 9, then Chicago and New York with 7 each, which reflects where this directory has spent its research time rather than where AI agencies concentrate. The verified set spans 105 distinct cities, 144 of them in the United States and 47 outside it, so treat the spread as real and the ranking within it as provisional.
• Self-description is the only source. A category count measures what firms say they sell. It is the right instrument for the question of what a buyer will be offered and the wrong one for the question of what gets delivered, and nothing here should be read as the second.
• Forty-three verified firms publish no category at all and are excluded from the specialisation figures. That is a gap in this directory's own data, not a finding about the market.

The count will be rerun monthly. The interesting number to watch is not the total but the strategy-to-integration ratio, because if the 33 advice-only firms start adding build capacity, that is the market responding to the complaint buyers have been making for two years.

Where to start

If you are sizing a shortlist rather than a market, the practical move is to describe the problem once and let the matching work against every listing instead of against the nine firms you have heard of. You can tell us what you are trying to build and get matched for free, browse the full directory by service, or start from a city on the locations index. Agencies reading this can list or claim a profile, which costs nothing and is how the count above gets more accurate.

Related reading on this site: when one person is enough and when you need a firm, the four trades sold as custom AI solutions, what you are buying in an ML consulting engagement, and the work an integration agency does after the demo.

Sources

• Stanford HAI, AI Index, on adoption and investment trends.
• OECD AI Policy Observatory, on national AI policy and uptake.
• US Census Bureau, Business Trends and Outlook Survey, on AI use reported by firms themselves.
• NIST AI Risk Management Framework, on the governance questions a strategy engagement should answer.