I keep hearing the same story. A mid-size company hires a major consulting firm for an AI transformation project. Six months and several hundred thousand dollars later, they’ve got a beautiful strategy deck, a proof of concept that sort of works in a demo environment, and no production deployment.

Then they bring in a four-person AI team. Within eight weeks, there’s a working system in production. Actual users. Actual results.

This isn’t an isolated anecdote. It’s becoming the dominant pattern in enterprise AI delivery, and it’s worth understanding why.

The consulting firm playbook doesn’t fit AI

Traditional management consulting follows a well-established pattern: assess, strategise, recommend, implement. It works for process optimisation, organisational restructuring, and technology migrations. The problems are well-understood, the solutions are proven, and the value comes from systematic execution.

AI projects don’t work like that.

You can’t fully plan an AI project upfront. You don’t know if your data is good enough until you try. You don’t know if the model architecture will perform until you test it with real inputs. You don’t know if users will adopt it until you put it in front of them.

This means the waterfall-style delivery that big consulting firms excel at is fundamentally mismatched with AI development. You need rapid iteration — build something small, test it, learn, adjust. That’s not how a 30-person consulting engagement is structured.

Where small teams win

Small AI teams have structural advantages that compound in this environment.

Decision speed. In a four-person team, the person writing the code sits next to the person talking to the business stakeholder. Questions get answered in minutes. Pivots happen in a day, not a quarter.

In a large engagement, the junior consultant who spotted a data quality issue writes a findings document, which goes to the project manager, who raises it in the weekly steering committee, which decides to form a workstream. By the time a decision is made, the small team has shipped their third iteration.

Technical depth. The best AI engineers don’t typically work at consulting firms. They work at AI labs, startups, or small specialist shops. A four-person team of senior AI engineers has more raw technical capability than a 20-person consulting team where half the members are generalists learning on the job.

Personal accountability. When there are four of you, there’s nowhere to hide. If the model doesn’t work, everyone knows who built it. This creates a fundamentally different incentive structure than a large engagement where responsibility is diffused across management layers.

Lower overhead, lower cost. A large consulting firm bills $250-$400 per hour per consultant, with significant overhead going to office space, sales teams, and management layers. A small AI team can charge $150-$250 per hour, deliver better results, and still earn more per person.

The data backs this up

McKinsey’s own research from late 2025 found that 74% of enterprise AI pilots fail to reach production. That’s not a technology failure — it’s a delivery model failure. Most of those pilots were run by large firms using traditional project methodologies.

Compare that with the outcomes from focused AI agent development teams. The numbers I’ve seen from smaller, specialist AI shops show production deployment rates above 80%. Not because they’re smarter, but because their delivery model is designed for the inherent uncertainty of AI projects.

Gartner’s 2026 AI implementation survey tells a similar story: organisations using small specialist teams reported 2.3x faster time-to-production and 40% lower total project costs compared to those using large system integrators.

When big firms still make sense

Large consulting firms aren’t useless here. But their role is narrower than they’d like.

Enterprise-wide strategy. If you need to figure out where AI fits across a 50,000-person organisation, a McKinsey or BCG engagement makes sense. They understand organisational complexity at a scale small teams can’t match.

Regulatory frameworks. For heavily regulated industries — banking, healthcare, government — the governance expertise big firms bring is genuinely valuable.

Scaling proven solutions. Once you’ve got a working AI system and want to deploy it across 200 locations, the programme management capabilities of a large firm become useful. That’s a logistics challenge, not an AI challenge.

But for the actual building? Small teams win. Consistently.

What this means going forward

The AI consulting market is bifurcating. Strategy firms handle transformation and governance at the top. Small, technically excellent teams do the actual building at the bottom. The middle — mid-size IT consultancies that used to win by being “big enough to be safe, small enough to be responsive” — is getting squeezed.

If you’re an enterprise buyer, the implication is straightforward: don’t hire the same firm for strategy and implementation. Get your strategy from whoever you trust for big-picture thinking, then bring in a small specialist team for the build. You’ll spend less, ship faster, and end up with a better system.

The era of the 50-person consulting engagement producing a proof of concept that never sees daylight is ending. Good riddance.