Over the past year I have seen dozens of AI pilots — our own, our clients’, and others I heard about from peers in the market — and I have gradually arrived at a conclusion that sounds almost heretical coming from the founder of a technology company: the success of a project barely depends on the model. The same technology, with the same prompts and the same architecture, transforms a process in one organization within a quarter and dies in the neighboring one a month later with the status “we played with it.” And when you start digging into what set those organizations apart, the answer every single time turns out to be not the technology but one specific person.
I call that person the AI champion, and today I want to unpack who this is, why it is almost never someone from IT — and how in our own projects we arrived at the hard rule “no champion — no agent launch.”
Who the champion is
Let me start with what a champion is not: not a job title, not a role in the project team, not a line in the implementation order. It is a person who owns the process and is personally, almost emotionally invested in making the agent work inside it. For our Registry Clerk the champion turned out to be not a developer but a records clerk together with the head of administration: they corrected assignments, the rule base grew out of their corrections, and they were the first to tell us where the agent was wrong. For the meetings loop, the champion is the facilitator of our Monday meeting. The assistant, however, had no single champion — and you can see it in the adoption curve: the audience grew organically, but autopilot configuration took off en masse only after several managers showed their morning briefings to colleagues.
Notice what all these people have in common: they come from the process, not from IT. IT people connect the agent; the champion makes it useful — and those are two entirely different crafts.
The global data says the same thing
My “heretical” observation, it turns out, sits comfortably on global statistics. McKinsey’s State of AI research shows that of all the factors influencing the financial impact of generative AI, the strongest is the redesign of workflows — not model choice, not budget size, but the willingness to rebuild the process itself around the agent. Yet only someone who owns the process and lives in it daily can rebuild it — that is, the champion, whatever the reports choose to call them.
The second telling example is Microsoft Copilot, the most massive enterprise AI rollout in the world. Hundreds of millions of employees received access through corporate licenses, yet roughly one in three uses it regularly. Access, it turns out, can be issued by decree — a habit cannot. The gap between “we issued the licenses” and “the work changed” is precisely the hole that a champion closes in living organizations: technology scales by copying, usefulness scales only through people.
Why it does not work without one
In its first weeks an AI agent resembles a new employee: formally it can do everything, in practice it does not know how things are done at your place. And it inevitably makes mistakes. The fate of the whole project is decided at the moment of the first visible mistake, because two scenarios are possible from there, and both are self-reinforcing. If the mistake is met by a champion, it turns into a correction, the correction into a rule, the rule into accuracy — and a month later the agent is noticeably smarter. If the mistake is met by an indifferent user, it turns into a screenshot in the common chat captioned “so much for your AI” — and a month later nobody uses the agent at all.
That is why in our transformation methodology every agent must have an owner at management level, and the working group meets weekly — not for reporting’s sake, but so that the agent’s mistakes have a keeper. And that is also why I am skeptical of the formula “implement AI for us turnkey, just don’t involve our people”: it does not happen. Technology can be delivered turnkey; usefulness can only be grown inside.
How to recognize a champion
Over the year I have settled on three signs by which I find these people almost unerringly.
First: their pain is the process, not the technology. They come not saying “we’d like to try AI” but saying “I spend three hours a day sorting incoming mail and it drives me mad.”
Second: they are ready to correct, not just to demand. A champion treats an agent’s mistake as a reason to fix a rule, while the indifferent treat it as long-awaited proof that “it doesn’t work.”
And third: colleagues trust them. Adoption spreads through an organization not by decree but by example — “it works for Aigerim, show me how you set it up” moves a project further than any regulation.
If no such person can be found at the start of a pilot, we now tell the client honestly: let’s start with a different process, one that has a champion. That is cheaper than resuscitating a dead pilot later.
What a leader should do
A champion should not be appointed by decree — they should be identified, because they already exist in the organization. It is the person who keeps their own spreadsheet better than the corporate system, who was first to bring ChatGPT into their work, who complains about the process loudest of all — not because they are a whiner but because they care. From there the leader’s job is to arm them: give them the agent, the right to make corrections and a direct channel to the implementation team, and then protect their time. An hour a day spent “raising” the agent in the first weeks is an investment that decides the outcome of the whole project more reliably than the choice of model.
Key takeaways
- The fate of an AI project is decided not by the model but by a person from the process whose pain is real.
- Agent mistakes are inevitable; the whole question is who meets them — a champion or a skeptic.
- Champions are not appointed — they are found and armed: the agent, the right to correct, protected time.
- No champion — don’t launch. Start with a process that has one.
We went through this on our own agents — from the registry office to meetings — and now we open every client project with the question “who is your champion?” If you want to try this on for your own process, come to the demo and I’ll show you what a champion’s work looks like from the inside.
