The big four are all selling "AI transformation." They're using the same model they use for everything else: large teams, long timelines, impressive slide decks, and a dependency that never ends. For AI, this model is fundamentally broken.
Here's why:
AI delivery is product work, not advisory work. Traditional delivery produces recommendations. AI delivery produces working systems. The skills are different: engineering, architecture, integration, data pipeline design. A team of MBAs who can write a brilliant strategy deck can't build a RAG pipeline or a document processing system. They'll subcontract it, and add 40% margin for project management.
Time-and-materials incentives are backwards. Consulting firms bill by the hour. AI foundations are designed to make things faster and cheaper. See the conflict? A delivery firm that builds you a genuine compound AI foundation, where capability #4 ships in 4 weeks instead of 14, is actively reducing their own revenue. The incentive is to build standalone projects that maximise billable hours, not shared infrastructure that minimises them.
The team that scopes isn't the team that builds. In big delivery, partners sell the work, managers scope it, and a rotating cast of analysts and associates deliver it. In AI, the people who understand the problem need to be the people who build the solution. Domain context doesn't transfer through a brief. It transfers through building.
Six-month assessments are six months too long. By the time a traditional delivery engagement has completed its "AI readiness assessment," a product-oriented team has built and deployed the first capability. The technology moves too fast for the deliberate pace of enterprise delivery. Discovery should take weeks, not quarters.
Dependency is the business model. Consulting firms grow by embedding deeply and becoming indispensable. The right AI partner does the opposite: builds your capability, transfers knowledge, and reduces their involvement over time. Ask any potential AI partner: "What does our relationship look like in 18 months?" If the answer is "we'll still be deeply embedded," they're building a dependency, not a capability.
What to look for instead:
- A team that builds and advises, not a team that advises and subcontracts the building
- Fixed-scope engagements with clear deliverables, not open-ended time-and-materials
- Foundation-first architecture that compounds value, with visible acceleration from capability #2
- Knowledge transfer baked into every phase. Your team should be more capable after every engagement
- A partner who's comfortable becoming less needed, not more
The right AI partner is more like a product studio than a delivery firm. They bring deep technical capability, domain experience, and a delivery model designed to build your independence, not your dependency.
The delivery model isn't evil. It's just wrong for this work. AI delivery needs builders, not advisors.