Why incumbent AI is a different sport
If your AI programme keeps stalling, the problem is not your team. You are running a playbook designed for a company one-hundredth your size.
I see it every few weeks. A bank or a telco brings in a senior team. They have read the same books and articles as the rest of us. The first pilot lands. The second pilot lands. The third one meets the operating model and dies a slow administrative death. Everyone is smart. Everyone is funded. Nothing scales.
The instinct is to push harder. Hire more engineers. Move the AI team closer to the CEO. Run another pilot. Wrong instinct, right diagnosis. The playbook does not apply. It was never written for the company you actually run.
Most AI transformation advice in circulation was shaped inside companies under five hundred people. Clean data. Concentrated talent. A leadership team that can decide on Monday and ship by Friday. Lovely if you are running a 200-person fintech. Useless if you are running a bank. Incumbents are playing a different sport, with a different clock, different penalties, and a different definition of winning.
Your data is forty years old. Stop pretending it isn’t.
A startup’s data sits in two systems and can be made usable in a quarter. Yours sits in forty systems. Three of them run on mainframes. The people who designed the original schema retired in 2014. That is not an IT failure. It is what decades of “yes, ship it” look like in real life.
The work is not waiting for the data to be ready. You will retire first. The work is designing AI for the data you actually have, and accepting that the cleanup runs on a separate clock from the use cases.
Hiring twenty engineers will not save a sixty-thousand-person company.
A startup hires twenty AI engineers and sits them next to the product team. An incumbent has to build fluency across thousands of people who already have full-time jobs, in functions the AI team has never met, in countries with different labour laws. You cannot hire your way through that any more than you can fix a public health problem by hiring more doctors.
The unit of progress changes. It is no longer “a model that ships.” It is “a workforce that becomes fluent.” Your AI function stops looking like a product team and starts looking like a public health programme. The sooner you accept this, the less expensive the lesson.
Your risk team is not the problem. It is the moat.
Founders get praised for moving fast and breaking things. Banks, insurers, energy companies, telcos? Not so much. Their risk function exists because a single serious mistake does not cost them a sprint. It costs them a decade. Customers do not trust you with their savings, their health data, or their power supply because you move fast. They trust you because you do not.
Treating the risk function as a problem to overcome is reading the situation backwards. The risk appetite is part of the product. The work is to design AI transformation that respects it and still moves. Hard, yes. Impossible, no.
Governance is the work. Not the friction.
Inside a young company, you can change a workflow on Monday and see the dashboard by Friday. Inside a bank, the same change has to clear works councils, internal audit, legal, risk, possibly a regulator, and a culture of decision-making older than most of the people in the room. Every one of those checkpoints is doing its job. Skipping them is how careers end.
The transformation that works plans for the governance path from day one. It builds the evidence each function needs to say yes. It treats those conversations as part of the work, not as friction in it. The leaders who get this right stop describing governance as “the blocker” within six months. The ones who do not are usually halfway through their second restructure.
The part nobody puts in the deck
In the work we do at Zeroto100 with banks, telcos, and industrial groups, the leaders who make real progress are the ones who stop trying to behave like a startup with more resources. The ones who struggle are still measuring themselves against companies they have no business comparing themselves to. The wake-up call is rarely a strategy document. It is usually a board meeting where someone finally says, out loud, “this is taking longer than we said it would.”
The good news is that the discipline exists. It is being practised right now, inside organisations that look very much like yours.
On 19 May at 5pm GMT+1, I’m hosting Dr Max Schumm for the next Zero to 100 Live. Max leads the AI Hub for Transformation at RWE, an energy company of sixty thousand people. The conversation will sit in the territory most case studies skip:
What actually stalled, and why.
How RWE is building fluency at that scale, across functions that had never met the AI team.
What the governance path looked like up close, week by week.
Free, online, with a Q&A you can actually use.
If your AI programme keeps hitting the same wall, you are running the playbook your peers read about. The one your organisation actually needs is shorter, slower, and built differently. So the question worth sitting with: which one are you running tomorrow?