From AI pilot to practice: the middle nobody plans for
The pilot goes well. It usually does. You pick a motivated team, give them access and a bit of hand-holding, help them work out a few good ways to use AI on their real tasks, and within a few weeks there are results worth showing. People are faster. The output is better. Someone demos it and the room nods.
Leadership sees the pilot and does the logical thing: greenlight the rollout. Take what worked for the pilot team and give it to everyone. And then, quietly, almost nothing happens the way it did in the pilot. The energy does not transfer. The good practices the pilot team developed do not appear elsewhere. A few months in, the rollout looks less like the pilot and more like the situation before the pilot, except now with more licenses.
This is the middle nobody plans for, the stretch between "it worked for one team" and "it is how we work," and it is where most AI programs stall. Understanding why is the difference between a program that scales and one that quietly reverts.
Why the pilot worked, and why that does not transfer
It is worth being honest about why pilots succeed, because the reasons are exactly the reasons they do not scale.
A pilot works because it is small, motivated, and supported. You chose people who wanted to be there. They had attention from you or a champion. When someone hit a wall, help was close. And critically, the good practices that emerged did not have to travel anywhere: the pilot team was small enough that a better way of doing something spread across it by sitting near each other and talking.
None of that survives contact with the whole organization. The next thousand people did not opt in. There is no champion at every desk. And the good practices the pilot developed now have to reach people who are not in the room, do not know the pilot team, and have no reason to go looking for what worked. The thing that made the pilot succeed, proximity, is exactly the thing you cannot scale.
The middle is a circulation problem, not a training problem
The instinct when the rollout stalls is to conclude that people need more training. So you run more sessions, produce more guides, hold more office hours. Some of it helps. Most of it does not move the thing that is actually stuck.
Because the problem in the middle is rarely that people cannot use AI. By the rollout stage, access and basic capability are widespread. The problem is that the good ways of working the pilot discovered are stranded. They exist, they are proven, and they are sitting with the twenty people who developed them while the other two thousand improvise from scratch. That is not a knowledge gap you close with training. It is a distribution gap you close by getting proven practice to the people who do that kind of work.
Put plainly: the pilot did not just prove that AI helps. It produced specific, tested ways of doing real tasks. The entire value of the middle is moving those from the pilot team to everyone who does the same work. If they do not move, the rollout is just a license distribution, and license distribution is not adoption. This is the same circulation gap that stalls programs at the "documented" stage, arriving one step earlier, at the moment you try to scale.
What the middle actually requires
Getting through the middle is less about pushing harder and more about building the thing the pilot never needed: a way for good practice to travel and be seen.
- Route by the kind of work, not by team. The pilot team's good approach to a task is useful to everyone who does that task, and those people are scattered across the org chart. Sending it to a department channel reaches the wrong rooms. It has to reach people by the job they do.
- Let use mark the standard. In the pilot, everyone knew which approach was good because they watched it work. At scale, nobody can watch. So the signal has to come from real, repeated use by people other than the author, which is what tells the next person this is the version to trust.
- Make contributing effortless. The pilot team shared naturally because they were close and motivated. At scale, if capturing what works takes effort, it will not happen. The good stuff surfaces only when saving it is a byproduct of doing the work.
- Measure the right thing, early. Not license activation, which will look fine and mean nothing. Watch whether practices are actually spreading beyond their origin, because that is the only signal that tells you the middle is being crossed rather than stalled.
The honest version of the plan
Most rollout plans have a pilot phase and a scale phase and a straight line between them. The straight line is the fiction. The real work is in the middle, and it is a different kind of work than the pilot: not proving AI helps, which the pilot already did, but moving proven practice to people who were not there to see it proven.
Teams that cross the middle are not the ones with the biggest training budget or the most licenses. They are the ones that treated the middle as its own problem, a circulation problem, and built the path that lets a good way of working reach everyone who does that kind of work.
If you are staring at that gap right now, the honest first step is to find out how far across it you already are. The maturity assessment places you on the path from ad hoc to evidenced in twelve questions, and names the one move most likely to get you unstuck. The pilot proved the idea. The middle is where you find out whether it becomes how the organization works.
Where does your program stand?
Most AI enablement stalls in the same place. The maturity assessment tells you where yours is in twelve questions, with no sign-up and no email to see your result.