Your AI pilot did not fail because the technology was not ready. It failed because it was treated like a new employee that nobody onboarded: no owner, no introduction to the data, no explanation of how the company actually works, and everyone piling their requests on top. A person in that position gets overwhelmed and leaves. The AI cannot leave, so it just produces a bad outcome.
The new employee nobody introduced to anyone
Think about what a new hire actually gets in a well-run company. An onboarding process. A manager who owns them. A working laptop. An introduction to the data they will need and to the people who hold it. Someone explains the methodology, the acronyms, the governance, the political structure, and who is allowed to authorise what.
Now list what your AI pilot got. In most companies the honest answer is a budget and a deadline, which is how the money gets spent without anyone owning the outcome.
It was given no structured data, or data so messy that nobody inside the company could explain it either. No clear statement of what it was supposed to do, by when, and to what standard. No owner with the authority to pick up the phone and get what was needed. And then, because it was the interesting new thing, every department pushed its own request at it.
The organisation failed. The AI did not.
It was treated as a fancy hobby
Every company on earth knows how to run a project. A project manager with real ownership. Supervision. A budget. The authority to escalate. Deadlines that mean something.
Almost none of that was applied to the first wave of AI pilots. They were run in isolation by a small group, deliberately walled off from the rest of the business, reporting to nobody in particular. I have watched pilots spend six months and a hundred thousand euros to arrive at the conclusion that the company would now use ChatGPT.
That is not a failure of ambition. It is a failure to treat the work as work. MIT’s own research on enterprise AI found the same pattern at scale: across more than 300 enterprise deployments, 95 percent of GenAI pilots produced no measurable return, and the researchers traced the gap to organisational design rather than the technology itself. The Norwegian evidence points the same way: research for NHO and Abelia finds that the gains rise with how broadly and how deeply AI is integrated into the business, not with which tool was chosen.
Why nobody stopped it
Boards approved these programmes, and the reason is not the one usually given.
They did not approve because they were convinced. They approved because saying no would have meant being the person who held the company back while competitors announced their own AI initiatives. Peer pressure and political pressure, not conviction. Most had no AI expertise on the board, were briefed on potential rather than substance, and understood that approval let the company describe itself as AI-first.
Which is how you end up with an expensive programme that nobody can honestly evaluate afterwards.
What onboarding an AI system actually requires
The same things a person needs, in the same order.
Someone who owns it by name, with authority rather than enthusiasm. Data that is structured and explained, not just available. A written statement of what it is doing, how, and over what timeframe. Introductions to the teams whose work it touches, before it touches them. And a clear answer to who decides when it is going wrong.
None of that is technical. All of it is management.
This is the part I find leaders genuinely relieved to hear, because it is work they already know how to do. If you can onboard a senior hire, you can onboard an AI system. It is the same discipline that decides whether a small start goes anywhere. What you cannot do is skip it and expect the thing to find its own way around your company.
The question to ask about the pilot you already wrote off
Before you write the next business case, go back to the failed one and ask what it was actually missing. Not which model it used. Whether it had an owner. Whether anyone explained the business to it. Whether the data was ready. Whether the objective was a real outcome or the ability to say the company was doing AI.
Answer those honestly and you will usually find the failure was structural and repeatable, which means the next one fails the same way unless something changes. You may also find, buried in the wreckage, something that worked and that nobody upstairs ever heard about.
Questions this article answers
Why do most AI pilots fail?
Because they are run without an owner, without prepared data and without the project discipline any other initiative would get. The failure is organisational rather than technological.
Was the technology oversold?
Rarely. In most cases the tools were capable of more than the pilot asked of them. What was missing was structure, ownership and a real objective.
What does onboarding an AI system mean?
Giving it what a new employee receives: a named owner, structured and explained data, a clear statement of the task and timeframe, and introductions to the people and processes it will touch.
Why did the board approve a pilot it did not understand?
Usually because refusing would have meant being seen to hold the company back while competitors announced their own programmes. Peer pressure rather than conviction, with no AI expertise in the room to test the case.
Should we run another pilot?
Only after establishing why the last one failed. If the cause was structural, and it usually is, a second pilot repeats it at greater cost.
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