The Product SystemThe Vault
Why AI Projects Fail to Pay Off: Count the Sign-Offs, Not the Models, the infographic in this PDF

Product decision-making

Why AI Projects Fail to Pay Off: Count the Sign-Offs, Not the Models

MIT found 95% of 300 company AI rollouts returned nothing measurable. The block is rarely the model. It is the approval chain, and who gets to decide.

Want this as a PDF?

You'll also get The Product System, my weekly note on leading product in the AI era. Unsubscribe anytime, and I'll never sell your email. How I handle your details

1

The full infographic, as a PDF

The exact one from the post, sized to print or keep within reach.

2

A copy in your inbox

The download link lands in your email too, so it's there when you need it. No follow-up campaign.

3

The Product System, weekly

My note on leading product in the AI era. One idea a week that compounds. Unsubscribe anytime.

Only 5% of company AI projects pay off. To be one of them, start here.

Count the sign-offs between a finished idea and a live customer. That gap is what decides it.

What the MIT research found

MIT looked at 300 company AI rollouts. 95% returned nothing the business could measure. Their own read: it isn't the models. It's everything around them.

That finding changes where to look when an AI programme stalls. If the model isn't the problem, a better model isn't the fix. The answer sits in how the organisation around the model works, and one part of that matters more than most.

Building is the easy part now

Your team can build faster than it ever has. What took a month, AI now does in days. Building is the easy part now.

Which means the slow part moved. A good idea gets built fast, then dies in the approvals before a customer ever sees it.

The picture that went with this post makes the point as a cartoon. A burning building is labelled "AI transformation". A row of leaders, the CEO, CTO, Head of AI, CPO and a consultant, stand behind a barrier marked "Leadership safety zone (observation only)", each offering an opinion. One AI builder is doing all the work with the hose. The headline: if your company is slow with AI, this is probably why. Too many leaders.

A two-week build, a year in the approval chain

A year ago I worked with a team stuck right here. They had a good idea: a two-week build that would genuinely change the lives of the people it was for.

Then it hit the approval chain. It had to clear the app team. Then the safety team. Then the team above them both. Then a board.

A year on, it still hasn't shipped. Nobody ever said no.

It was everyone's priority. So it was everyone's approval. So it was no one's call.

The build was never the problem. The deciding was. Tool speed went up. Decision speed stayed flat. And the people that idea was for are still waiting.

Decide who gets to decide

The fix isn't another tool. It's deciding who gets to decide.

Take the calls that don't need you and hand them down with three things.

The outcome you want. Not the solution, the result it should produce, so the person deciding can judge options against it.

The context behind it. The history, constraints and reasons that make the outcome matter, so they don't have to come back to you for them.

Where their decision is final, not yours. This is the part that makes the other two work. Without it, every handed-down decision turns back into an approval request, and the chain grows again.

Do that, and the next good idea ships in weeks, not quarters. That's when your AI spend starts showing up as outcomes your CEO can see, and how you join the 5%.

For more on giving a team what it needs to decide without you, see /vault/every-question-only-you-can-answer.

Count the sign-offs again next month

Start with a count. Take one recent idea and list every sign-off between "finished" and "live with a customer". Then hand down the calls that don't need you, using the three things above.

Then count the sign-offs again next month. If the number hasn't moved, the speed of your tools still isn't reaching your customers.

One more test: what is one decision your team owns now that would have ended up on your desk a year ago? If you can't name one, that is where to begin.

Questions people ask

Why do most company AI projects fail to pay off?
MIT looked at 300 company AI rollouts and found 95% returned nothing the business could measure. Their read was that the models aren't the problem; everything around them is. One of the biggest blocks is the approval chain: ideas get built quickly, then stall in sign-offs before a customer sees them.
What did the MIT study on AI rollouts find?
MIT looked at 300 company AI rollouts and found that 95% returned nothing the business could measure, leaving only 5% that paid off. The researchers' own conclusion was that the failure sits in everything around the models rather than the models themselves.
How do approvals slow down AI projects?
AI has made building fast, so the slow part has moved to deciding. When an idea needs sign-off from several teams and then a board, it can wait months or longer even though nobody says no. When something is everyone's approval, it becomes no one's call.
How can leaders speed up decisions on AI work?
Decide who gets to decide. Hand down the calls that don't need you with three things: the outcome you want, the context behind it, and a clear statement of where the other person's decision is final. Then count the sign-offs between a finished idea and a live customer, and track that number each month.
What is decision speed and why does it matter for AI?
Decision speed is how quickly a finished idea gets the approvals it needs to reach customers. AI raised tool speed sharply, but if decision speed stays flat, faster building just produces a longer queue of unshipped work. AI spend only shows up as outcomes when decisions move as fast as the build.