Back to Insights

Companies are bypassing big consulting firms for AI. Here's why.

By Edward Sharpless, D.Sc.

OpenAI just signed multi-year partnerships with McKinsey, BCG, Accenture, and Capgemini. Anthropic locked in Accenture and Deloitte. The two most important AI companies in the world looked at the enterprise market and concluded the same thing: they can’t get there alone. They need the consulting firms to carry their platforms into the Fortune 500.

That tells you something about how hard enterprise adoption actually is. It also tells you something about the model those consulting firms are going to use to do it. Because it’s the same one they’ve used for everything else.

They called it digital transformation in 2017. Cloud migration in 2020. AI implementation in 2024. The service offering changed. The engagement model didn’t. Armies of junior consultants staffed at $250 to $400 an hour. Twelve-to-eighteen-month timelines. Discovery phases that produce decks instead of systems. And a pricing structure that rewards duration over results, because the longer it takes, the more they bill.

A typical McKinsey strategy engagement costs more than a million dollars a month. An enterprise-wide implementation from Deloitte or Accenture can run into the tens of millions across multiple years. These aren’t outlier numbers. They’re the model. The consulting industry generates over $300 billion annually by staffing large teams against long timelines. AI is the newest thing being fed through that machine.

But here’s what should concern every executive writing those checks: the firms selling AI transformation haven’t even transformed themselves. McKinsey has 45,000 employees. Deloitte has 470,000. BCG has 33,000. If their internal AI tools worked as advertised, if their proprietary platforms actually delivered the productivity gains they promise clients, they would need a fraction of those people. They haven’t cut. The pyramid is intact. The billable hour model is intact. The productivity gains go to the firm’s margin, not the client’s bill.

And the early results bear this out. Capgemini found that 88% of enterprise AI pilots never reach production. S&P Global reported that 42% of companies abandoned their AI initiatives before delivery, up from 17% the prior year. The National CIO Review found that companies are increasingly bypassing consulting firms entirely because consultants “built compelling prototypes but lacked the depth or expertise to scale those ideas across an enterprise.” Deloitte itself had to refund part of a $290,000 Australian government contract after delivering a report filled with AI-generated fabrications, including non-existent academic citations and a fabricated legal quote. A researcher caught the errors because he personally knew the authors who had supposedly written the referenced works.

The pattern isn’t subtle. Companies pay premium rates for AI expertise and get generalists with fresh certifications who are learning on their client’s dime. The consulting model was built to scale human labor across long engagements. It was never designed to deliver the kind of fast, architecture-level work that AI transformation actually requires.

The model problem

This isn’t about whether individual consultants at these firms are talented. Many are. The issue is structural.

The traditional consulting pyramid works like this: a small number of senior partners sell the work, a mid-layer of managers oversee it, and a large base of junior analysts and associates do the bulk of the research, modeling, and deliverable production. Revenue comes from billing that base at rates far above their cost. The economics require volume. The more people staffed, the more revenue generated. The incentive structure rewards duration above everything else.

AI consulting doesn’t fit this model. The work requires senior-level architectural thinking from the first conversation. It requires people who understand both the technology and the business deeply enough to see where intelligence should replace process. No certification program produces that, regardless of how many thousands of employees you run through it.

When BCG says 90% of its 33,000 employees “use AI,” that tells you about tool adoption. It tells you nothing about whether those employees can architect an intelligence-native operating model for a healthcare system or a financial services firm. Using Copilot to draft emails is not the same as understanding how to restructure a claims adjudication workflow around autonomous agents.

The consulting firms know this, even if they can’t say it publicly. McKinsey built an internal tool called Lilli and deployed it to 72% of its workforce. The tool automates research and knowledge synthesis, tasks that junior consultants used to bill dozens of hours to complete. But instead of reducing headcount or lowering prices to reflect the efficiency gains, the firm absorbed the productivity and kept billing the same rates. The surplus goes to the firm. The client pays the same. That’s the incentive structure of the pyramid.

One industry analysis described it as “masking structural inertia with digital acceleration.” AI gets applied to the internal workflow to boost margin. It doesn’t fundamentally change what the client receives or what they pay.

What the alternative looks like

The enterprises that are actually moving fast on AI aren’t using 40-person consulting teams. They’re working with small groups of senior architects and engineers, people who combine deep business understanding with hands-on technical capability, who can go from strategy to functioning system in weeks instead of quarters.

This is possible because AI has changed what a small team can accomplish. A team of one to five experienced practitioners, using the same AI platforms these consulting firms claim expertise in, can deliver enterprise-grade systems in weeks that would take a consulting firm a year or more of discovery, planning, and phased implementation. They can do it because they don’t have layers. The person who designs the solution is the same person who builds it. There is no translation loss between the strategy deck and the implementation. There is no six-week discovery phase that produces a document nobody uses. There is no bench of juniors learning on your time.

By the time a national consulting firm delivers your AI strategy document, a high-performing FDE team will have already built and implemented the solution.

The economics reflect the structural difference. Where a McKinsey engagement might cost $3 million for a strategy assessment, a specialist firm can deliver equivalent strategic depth in a week for a fraction of that. Where a Deloitte implementation might take 18 months and $15 million, a lean team with the right architectural vision can build and deploy in a single quarter for less than a tenth of the cost. This isn’t a marginal improvement. It’s a category difference, the same kind of cost structure collapse that AI is causing across every other industry.

And the quality argument runs the same direction. When your project is staffed by the same senior architects from start to finish, you don’t get the telephone-game degradation that happens when a partner sells the vision, a manager interprets it, and an associate builds something that resembles neither. You get direct translation from strategy to system. Every decision reflects the original architectural intent.

The credibility question

The uncomfortable question for enterprises isn’t whether consulting firms are good at what they do. It’s whether the consulting model is the right tool for this moment.

These are the same firms that led digital transformation initiatives over the past decade. The track record is mixed at best. Long timelines, massive budgets, and outcomes that frequently fell short of the original business case. That experience is well documented and widely understood by the executives who lived through it.

For decades, the safest decision in enterprise technology was to hire the biggest name in the room. You can’t get fired for hiring IBM. That was true until it wasn’t. IBM went from defining enterprise computing to defending a shrinking consulting business built on legacy systems most companies are trying to escape. The biggest name became the safest way to guarantee you’d be the last to move.

Now those same firms are rebranding the same engagement model for AI. The pitch has changed. The structure hasn’t. And the results from the first wave of AI consulting engagements are tracking similarly: high investment, long timelines, and a troubling gap between what was promised and what was delivered.

The national consulting firms have genuine strengths. Scale. Global reach. Established relationships with procurement departments and boards. Regulatory expertise in heavily governed industries. These matter. For certain kinds of work, particularly compliance-heavy implementations that span dozens of countries, their infrastructure is hard to replicate.

But for the core strategic and architectural work of AI transformation, for the part that actually determines whether the investment creates value, the model is wrong. The work requires depth over breadth, speed over scale, and practitioners who build over analysts who present.

The real question for enterprise leaders

The frontier AI companies partnering with consulting firms are making a distribution play. They need these partnerships to get their platforms into enterprise accounts. That’s rational for OpenAI and Anthropic.

But enterprise leaders don’t have to follow the distribution channel. The platform is available to anyone. The question isn’t who can get you access to the technology. Everyone has access. The question is who can see your business clearly enough to know where the technology should go and what it should replace.

That requires someone who understands the structural truth of your business: your entities, relationships, decisions, constraints, and value flows. Someone who starts from first principles instead of vendor frameworks. Someone whose incentive is to finish fast and finish right, because their model doesn’t depend on keeping you engaged for years.

The enterprises that figure this out first will operate at entirely different economics. The ones that default to the familiar model, that hire the big name because it feels safe, will spend more, wait longer, and get less. They’ll look back in three years and realize they paid a premium for the same playbook that underdelivered last time, just with a new label on it.

Intelligence-native transformation requires architects, not armies. The companies that understand the difference will move first.

The national firms aren’t enemies. They’re designed for a different problem. The question is whether the problem you’re solving is theirs or something new entirely.

That’s the question we help answer.

Share
Explore Fractional CAIO