Nobody Is Short of Consultants
The AI services industry answered an execution problem by building delivery capacity: 300,000 consultants to certify, a 77,000-person AI bench, 8,900 engineers sent to sit inside client businesses. Capacity was never the constraint, and none of it tells a buyer which workflow to rebuild first.
- AI implementation
- agency selection
- systems integrators
- OpenAI
- Accenture
- TCS
- market structure
Put three numbers from the past year next to each other.
In June, OpenAI launched its Partner Network with $150 million behind it and a stated target of certifying 300,000 consultants by the end of 2026. Accenture closed FY25 with $5.9 billion in advanced AI bookings, around 6,000 projects deploying advanced AI, and an AI and data workforce that had grown from roughly 40,000 at the start of FY23 to nearly 77,000. TCS told shareholders in its FY26 annual report that more than 270,000 of its people now hold advanced AI skills, three times the year before, and said in July it intends to convert 1% to 1.5% of its associate base into forward-deployed engineers. Against a headcount of 593,798 at the end of June, that's somewhere between 5,900 and 8,900 people whose job is to sit inside a client's business and make the thing work.
Read separately, each of those is a growth story, and the trade press filed them that way. Read together they describe an industry that looked hard at a demand problem and answered it by manufacturing supply.
The shortage nobody could find
Gartner surveyed 1,303 organisations with at least $50 million in annual revenue between January and April this year. Only 22% had scaled AI across multiple business units or adopted an AI-first approach. Roughly 11% couldn't say what their own function had spent on AI in 2025.
Now ask what those organisations were short of. Not models: the frontier is a credit card away and has been for two years. Not engineers, evidently, given the numbers above. Not certified expertise, which is being minted at a rate of 300,000 a year by a single vendor.
What the 78% were missing was a decision. Which of the forty things this business does badly should change first, what does better look like when it's measured, and who's going to be able to tell in six months whether it worked. The 11% who can't reconstruct last year's spend aren't short of consultants. They're short of an account of their own operation, which is not a thing you can buy by the head.
OpenAI, to its credit, said as much in the announcement. Model capability is no longer the main barrier to enterprise value; the bottleneck has moved to implementation, workflow redesign and change management. That's the execution argument, published by the company with the most to gain from the access argument. Then it committed $150 million to certifying people in the platform.
What a bench can't tell you
A certification tells you someone understands a platform. A forward-deployed engineer tells you someone will sit close to your work. A 77,000-person AI and data workforce tells you the firm can staff whatever you sign.
None of that answers the only question that determines whether the programme pays. A Codex specialisation doesn't teach anyone your accounts-payable exception path, or why three of your five regions quietly stopped using the CRM in 2023, or which of your reporting cycles exists because a customer once complained and nobody has revisited it since. The difficulty in all of those is undocumented knowledge, held by people with day jobs.
So the capacity arrives, correctly certified, and then spends the first eight weeks doing discovery that nobody scoped, priced, or thought to ask about during procurement.
What this does to your shortlist
Three years ago, "we have serious AI capability" sorted a shortlist. It doesn't any more, because every firm you'd consider can now say it and be telling the truth. Accenture can point at 6,000 projects. TCS can point at 270,000 trained staff. A twelve-person specialist can point at the same OpenAI certification the Big Four consultants hold, because it's the same certification.
The claim has stopped carrying information. Every one of these firms built real capability at real expense, and each of them has earned the line. The cost lands on the buyer, who now has considerably less to choose on than a year ago.
The differentiator moved upstream, to method. Not how many people a firm can put on your project, but how it decides what those people should touch. That's a question with a real answer, and it's testable in a first conversation. Ask a prospective partner to describe how they chose the scope on their last engagement: what they looked at, what they measured before they started, what they recommended leaving alone, and what would have had to be true for them to recommend doing nothing. A firm with a method will walk you through it and it'll sound unglamorous. A firm without one will show you the bench.
The uncomfortable part
There's a reason the method question doesn't come up much. A good answer tends to shrink the engagement. If a partner can genuinely identify which three of your workflows are worth rebuilding, the honest recommendation is usually three, not thirty, and the invoice reflects that. Capacity has the opposite property: it's easiest to sell in volume and its value is hardest to falsify afterwards.
Accenture reports FY26 in a few weeks. The advanced AI number will almost certainly be larger, and it'll be written up as further proof that the market is maturing. It'll tell you nothing about whether any of it worked, because none of these firms report the figure that would: how many of those 6,000 projects are still running, and what changed in the businesses that bought them.
Until someone publishes that, the bench is the only thing being measured. It's also the one thing that was never in short supply.
If you want that question answered for your specific situation, the Forge Playbook does it. Answer a few questions about your business and we'll put together a tailored outline of which workflows are worth automating and what a realistic budget looks like for each. Free, no obligation, takes about three minutes.
