The four patterns that keep repeating
Before the names, the patterns. Every failed enterprise AI effort I’ve watched runs one of these plays.
The Pilot Hamster Wheel. Dozens of pilots launched, none reach production. Motion mistaken for progress.
Governance Theater. Principles published in a designed PDF. Nothing changes on a Tuesday.
The Ball Chasers. A new model drops, a new offering appears, and the strategy resets around whatever sells this quarter.
The Demo Trap. Boards get product demos instead of oversight. Demos are marketing. Oversight needs questions.
Now the receipts.
The industry has an AI problem
Not a technology problem. A discipline problem.
MIT’s Project NANDA found that 95% of generative AI pilots produced no measurable impact on the bottom line, despite tens of billions in enterprise spending. Gartner expects more than 40% of agentic AI projects to be canceled by the end of 2027, killed by rising costs, unclear value, and inadequate risk controls. Even Accenture’s CEO admitted on an earnings call that value realization has been “underwhelming for many.” Deloitte’s own survey found regulatory compliance is now the top roadblock to deploying GenAI, and most companies expect less than 30% of their AI experiments to reach full scale within six months.
Read that again. The firms selling the transformation are publishing the evidence that it isn’t working.
The people selling governance keep failing it
This is the part nobody in the industry wants to dwell on.
Deloitte refunded roughly A$440,000 to the Australian government after delivering a report containing AI-hallucinated content (dig.watch). KPMG withdrew an agentic AI report when analysis showed its citations were largely fabricated (TechCrunch). EY pulled a report over apparent AI hallucinations in the same wave.
Said plainly: the firms marketing trustworthy AI shipped untrustworthy work. The governance experts didn’t govern their own output.
Chasing the ball instead of minding the goal
Here’s the pattern. Every cycle, the big consultancies chase the ball: the new model, the new agent framework, the new offering to sell. Meanwhile nobody minds the goal. The unglamorous fundamentals: clean data, clear ownership, honest measurement, the willingness to say no.
It’s not malice. It’s incentives. You can’t bill for telling a client their problem doesn’t need AI. So the machine keeps selling, the pilots keep failing, and trust keeps leaking out of the whole enterprise AI project.
When the big firms are the right choice
Sometimes they are. If you need hundreds of consultants across six countries by Monday, a statutory audit, or a global rollout run on someone else’s bench, hire them. Scale is a real capability, and they have it.
What scale does not excuse is unverified work. Whoever you hire, a global firm or an independent, the standard does not change: check the output before it ships. That standard is the whole argument of this page.
What I’d do instead
Start with the goal, not the ball.
Govern the data before you scale the models. Prove value in small, honest experiments before committing real money. Build the operating model first: who approves, who monitors, who pulls the plug. And keep one uncomfortable question on the table at all times: should this be AI at all?
That’s the work I do. Here’s what I believe.