A pattern we have seen at every SME we have worked with on a Copilot rollout beyond the pilot phase: month three is good, month four is good, month six is suspiciously quiet, and the month-nine review shows two camps. The original five pilot users are still saving meaningful time. The next twenty users are using Copilot intermittently or not at all. The licence cost is steady, the value is concentrated, and the question "should we have rolled this wider" gets asked at the worst possible moment, halfway through the annual commitment.
This pattern is the post-pilot stall, and it is not Copilot-specific, though Copilot rollouts surface it more visibly than other software because the value is so user-dependent. Here is the shape we see, and the pattern that pulls a stalled rollout back into productive use.
Why the stall happens
Three reasons that compound.
The pilot users self-selected. The five people who were briefed, made time, and gave the rollout an honest run were already the kind of people who try new tools. The next twenty are not necessarily the same population. Most are too busy to add a tool to their day without a forcing function.
The wider briefing was lighter. The pilot users got a 60-minute session and 15-minute check-ins. The wider rollout often got a 20-minute group session and an email with a link. The drop in onboarding density correlates almost perfectly with the drop in usage.
The workflows did not transfer cleanly. The pilot users discovered the workflows that worked for their specific jobs. The wider rollout assumes those workflows are universal. Some are; meeting summaries work for anyone in meetings. Some are not; the way a specific salesperson uses Copilot to draft outreach does not translate to the operations team without a translation.
What the 90-day review of a stalled rollout shows
When we get called in on a stalled rollout, the pattern is consistent. The usage data tells the story before any conversations need to happen.
Two distinct populations. Plotting active users per week against weeks since onboarding, two curves emerge. The pilot group sits flat at a high usage level. The post-pilot wave starts at a moderate level, drops by week three, and is at near-zero by week six. Same product, same prompts, different onboarding pattern.
No prompt library handover. The pilot users built a working set of prompts. Those prompts mostly live in the pilot users' heads. The wider rollout never inherited them.
No mid-team champion. Each team that adopted Copilot wider had no nominated person inside the team who owned it. The IT lead is too far from the day-to-day work; the team needs its own person.
The check-in cadence collapsed. The pilot had a structured cadence of conversations. The wider rollout had a launch and then silence. Without a forum for "I tried this and it did not work", the team's quiet frustration becomes quiet disuse.
How to pull a stalled rollout back
The pattern that works, in the order we run it.
1. Segment the population
Pull the usage data and identify three groups: still-using, drifted, and never-started. The numbers are different per business but the shape is consistent. Plan the recovery for each group separately, because the intervention is different.
For the still-using group, the work is mostly to keep them productive and surface their prompts to the rest. For the drifted group, the work is a targeted re-onboarding with the workflows that fit their specific job. For the never-started group, the work is a short pilot of two or three people from the group to find out whether the product fits the work at all.
2. Build the prompt library properly
The library is the artefact that translates pilot-user knowledge into wider-rollout fuel. Two prompts per workflow, per role, with the source workflow described in one sentence. Lives in a single location, searchable, owned by a named person, reviewed quarterly.
This is unglamorous work, and it is also the highest-leverage piece of a recovery. A prompt library someone can copy from at the moment they have a task to do beats any amount of generic training.
3. Nominate team champions
One person per team of five to eight, whose role is to be the "ask me about Copilot" person inside that team. Not a heavy commitment; an hour a week of office hours plus visible use. The team needs a peer, not an IT lead two layers away.
The champions get a 90-minute briefing on the prompt library, the policy, and the most common workflows. They do not need to be technical; they need to be the person on the team who tries things and tells the truth about whether they work.
4. Reinstate the cadence
A monthly 30-minute meeting per team where the champion runs through what worked that month, what did not, and what the team is asking about. The meeting produces three things: a quick-wins list for the team, a problems list for the rollout lead, and a sense that this thing is being managed rather than left to fend for itself.
Three months of this cadence usually pulls the never-started group up to the drifted group's level, and the drifted group back into productive use. The still-using group keeps going.
What not to do
Three patterns we see attempted that mostly do not work.
More training. The team has had training. More training is not the problem and is not the answer. What they lack is the moment-of-task workflow that fits their specific job; a generic refresher does not produce that.
Cancel the licences. Tempting if the usage data is bad, but it punishes the still-using group for the rollout's failures. Try the recovery first; cancel at the year-end if the recovery does not land.
Switch products. The conviction that "Copilot is bad, we should try ChatGPT instead" is sometimes right but more often wrong. If the usage data shows the still-using group is getting real value, the product is fine and the rollout is the problem. Switching products restarts the problem.
Where this lands with us
Adoption recovery is one of the engagements our AI Enablement practice runs. We come in at the post-pilot stall point, look at the usage data, build the prompt library properly, nominate the champions, and reinstate the cadence for three months. The engagement is shorter than the original rollout because the core work, the discovery and policy, is already done.
The rollout that gets second-stage attention is the rollout that produces real value. The rollout that does not gets cancelled at renewal and goes down as an expensive lesson.
Sitting on a Copilot rollout that has stalled and not sure how to restart it? Drop us a note at info@jmopartners.co.uk.
JMO|Partners · Enterprise IT, sized for SMEs.