In most large companies the fourth AI project is the best one, and it is the one that gets starved. By then the energy, the enthusiasm and the budget have been spent on three earlier projects that were about visibility rather than results. The people who ran those three were rewarded regardless of the outcome, which is precisely why nobody stopped to ask what went wrong.
Being first pays better than being right
The standard account of pilot fatigue says the champions of failed AI projects suffer for it. Inside the room, that is not what happens.
The first person to push an AI project through a large organisation is seen as a pioneer. An innovator. The one who moved while everyone else hesitated. Their career goes up, and the outcome of the project does not change that. It is about brand and visibility, not delivery.
The larger and more political the company, the more true this is. Being first outweighs being right, because being first is legible to everyone immediately and being right takes eighteen months to establish.
Which means the incentive to run a second, third and fourth project is strong, and the incentive to examine why the first one did not work is close to zero. That first one was more likely to be abandoned than not: Gartner projected that at least 30% of generative AI projects would be abandoned after proof of concept by the end of 2025, citing poor data quality, inadequate risk controls, escalating costs, or unclear business value.
Why people join AI project teams
Ask honestly why the team assembled. Status. Visibility. A title with AI in it. A salary review that goes better than it otherwise would. In several companies, a two or three week trip to visit AI companies in Silicon Valley.
Many of the people on those teams did not know AI and were not especially invested in the deliverable. That is not a moral failing, it is what the organisation rewarded. And a company large enough to absorb the cost simply absorbs it and moves on.
The fourth project is the mature one
Here is the part nobody writes down.
Whoever builds the fourth project has studied the first three. They know what was attempted, who ran it, how it failed, and how leadership quietly stopped attending the reviews. They have seen which assumptions did not survive contact with the company’s actual data and politics. Very often they have ground the thing out on their own time, because no budget was left to give them.
So the fourth is the most realistic, the best organised and the most likely to produce something. It deserves wholehearted support.
It arrives exactly when the organisation has decided that AI does not work here. A Norstat survey across the Nordics found large AI investment producing little measurable effect in Norwegian businesses, which is the atmosphere the fourth project has to survive.
What a leader should say instead
Ask an executive about their AI record and you will often hear a version of this: I have approved three AI projects worth over a hundred million euros, and I am very proud of that.
That is not an outcome. Anyone with budget authority can approve spending, and approving it as an open account rather than a project is how the first three got funded. What a board should want to hear is what those projects produced, what was learned, and what was stopped as a result.
The same applies to the people running them. Not what was launched. What changed.
How to protect the fourth project
Three things, none of them expensive.
Find out whether it exists. In most companies with a history of failed pilots there is someone quietly building the version that would work, usually without a mandate. They are rarely senior and almost never invited to the board presentation.
Judge projects on outcome rather than on order of arrival. If being first is what gets rewarded, you will keep getting firsts. And ask whether the earlier ones were ever given an owner and an introduction to the business at all.
And before approving anything further, establish what the earlier attempts actually cost and what they left behind. Somewhere in a project written off as a failure there is often something valuable that nobody upstairs ever heard about.
I have sat on both sides of this: inside the company running the project, and across the table from the board being asked to approve the next one. The pattern is consistent enough that I now ask about project four before I ask about project one.
Questions this article answers
Does a failed AI project damage the career of the person who led it?
Usually not in a large organisation. The first mover is credited with initiative and visibility, and the outcome has less bearing on their standing than the fact that they moved first.
What is pilot fatigue?
The point at which repeated AI failures cause executives to disengage. By the third or fourth attempt, attention and budget have gone, and the organisation has concluded that AI does not work in its context.
Why is the fourth AI project usually the best one?
Because whoever built it studied the previous three: what was attempted, why it failed, and what leadership stopped supporting. It is the most realistic and best organised, and it arrives when the appetite has been exhausted.
How should a board judge an AI project?
On what it produced, what was learned and what was stopped as a result. The number of projects approved and the money committed are inputs, not outcomes.
How do we find the project worth backing?
Ask who is already building a version without a mandate. In companies with a history of failed pilots, that person usually exists and is rarely in the room.
If you are leading your organisation through this, this is the kind of work I do with a limited number of senior leaders each quarter. → https://anglero.com/the-reset/
Strategic AI Advisor, MerkabaPhi AS, Oslo. 450+ keynotes across 30+ countries.
Listen to the Clarity at the Top podcast on Spotify.
Enquiries: anglero.com