By Rebecca Keenan
Solutions Director – AI & automation at Expleo
When an AI pilot stalls, the post-mortem always reaches for familiar explanations: the technology was not mature enough, the data was not ready, the vendor overpromised. Of course, this can happen. However, from what I’ve seen on the ground, we appear to have got into the habit of overlooking a more ordinary reason.
It’s a pattern we’re all familiar with: the technology worked, leadership wanted it, the business case existed, but somewhere between the executive team and the people doing the work, the organisation simply failed to change. So why is this the case? It comes down to structure, not people. Middle management sits at the crossroads between strategy and delivery, and that’s where change tends to get stuck.
Rarely does anyone stand up and oppose AI. Instead, it stalls through a hundred smaller decisions. The project meeting moves because the operational meeting is more urgent, training is postponed because service levels are under pressure, the existing process continues to run alongside the new one “for now” or the manager decides that this is not the right quarter to disrupt the team.
Six months down the line, the pilot still exists and nothing about how the work gets done has changed.
Calling this resistance to change or a lack of commitment neglects the nuance behind these decisions. In many cases, it is a rational response to the incentives we have created.
Picture this: a middle manager is asked to implement a new AI-enabled process. They’ve always been measured on service levels, output and quality, and those targets don’t move, even during the pilot. Disruption inevitably follows, as people need time to learn the new process and early outputs are temporarily inconsistent because training is still underway. As a result, productivity dips before it eventually rises again.
Yet, the incentives around this dip are lopsided. If the initiative succeeds, the credit travels upwards. It becomes the organisation’s AI strategy, or the executive sponsor’s success story. If it fails, accountability travels the other way. It was their team, their customer, their missed target.
This kind of imbalance does not just erode goodwill, it breeds fear, which in turn raises a more uncomfortable question for middle management: what happens to my role and the roles of my colleagues if this works? A manager who has spent years building a team of 30 will not react the same way as the boardroom pitch suggests when told that successful adoption means managing a team of 20. This doesn’t make managers anti-technology. It makes them human.
Many organisations respond by increasing their communications effort: more town halls, more newsletters, more AI champions. While communication matters, you cannot communicate somebody out of a target they are still being held to. If managers are measured on protecting today’s performance while the transformation programme asks them to disrupt it, the old operating model will always win.
Two things need to change.
First, adapt the measures at the same time as you change the work. If a manager is asked to implement a new process but their objectives still assume the old system, they have been set up to fail. Be explicit about what is allowed to move during the transition. If service levels can dip while a new way of working takes hold, specify the acceptable range and the timeframe. “Permission to experiment” sounds good in a town hall. It means very little when the dashboard turns red on a Monday morning.
Second, involve the people who run the process in designing its replacement. This is usually framed as a way of creating buy-in, which undersells it. The people closest to the work know where the real problems are and which parts of the documented process bear no resemblance to what actually happens. Their involvement does not just make adoption easier. It makes the solution better.
Both depend on a change of mindset: adoption is not the final phase of an AI project. It is the project. A flawless technical implementation produces little value if nobody changes how work gets done.
That is also where the competitive advantage now sits. Model capability is becoming widely accessible. The ability to change processes, move accountability and get thousands of people working differently while still running the business is much harder for a competitor to copy.
AI adoption rarely fails dramatically. More often it slows quietly somewhere in the middle of the organisation until everybody moves on to the next initiative. If I had one piece of advice for organisations adopting AI: don’t treat middle management as an obstacle to work around. Treat them as the place to start.
