Key takeaways
- Usage spikes in week one and troughs around week five. This is normal and it is the point at which most rollouts are abandoned.
- Watch three numbers: reach, depth, and concentration. Concentration is the one that hides failure.
- Run refresher training at day 45, when people have real problems, not at day 5 when they have none.
- The most effective intervention is a colleague demonstrating something useful in a meeting that already exists.
Almost every rollout follows the same curve. A burst of activity in the first fortnight as people try things, a visible drop by week five, and a settling point somewhere in weeks eight to twelve that represents the actual level of useful adoption. Organisations that have not been warned about this interpret the drop as failure, conclude the technology was overrated, and stop investing attention at exactly the moment attention matters most.
The curve is not a problem. Not knowing it is coming is.
What the trough actually is
The first two weeks are curiosity. People try the tool on whatever they happen to be doing, including things it is bad at, and they generate a lot of activity that has nothing to do with their working habits.
By week five that has burned off, and what remains is the set of tasks where the tool genuinely fitted into someone's day. That number is always lower than the peak and it is the honest baseline. The work of the first ninety days is not preventing the drop. It is finding out which tasks failed to stick and whether they failed for a fixable reason.
There are usually four reasons, and they need different responses:
- The task was never in their workflow. Someone was shown a capability that does not correspond to anything they actually do. Nothing to fix, remove it from the training.
- The output needed too much editing. Fixable, and usually a matter of context. The person was giving it three lines of instruction and expecting firm-standard output.
- They hit a limit or a wall. A usage cap, a file type it would not accept, a model that was not available on their tier. Fixable, and invisible unless you ask.
- They forgot. The single most common reason, and the only one that responds to prompting rather than problem solving.
You cannot tell these apart from usage data. The dashboard shows you that someone stopped. It never shows you why, and the four reasons need four different responses. Fifteen minutes of conversation with six people tells you more than any report, and it is the step organisations skip.
The three numbers to watch
| Measure | What it tells you | Warning sign |
|---|---|---|
| Weekly active users as a share of licences | Reach | Below half by week eight |
| Average sessions per active user per week | Depth, whether it is habitual | Fewer than three, meaning it is occasional |
| Share of total usage from the top five users | Concentration | Above half, meaning enthusiasts are carrying the average |
| Licences with no sign-in for 30 days | Waste | Any, after week eight |
| Spend against ceiling | Cost control | Tracking above plan with flat usage |
The third row is the one that matters most and the one nobody looks at. A firm with eighty licences and healthy total activity can be a firm where six people use it constantly and seventy-four have not opened it since the training. The average looks fine. The rollout has failed.
Check concentration monthly. If your top five users account for more than half of everything, you do not have adoption, you have enthusiasts.
The ninety day calendar
Week two: the friction sweep. Ask six people from different teams what has been annoying. You are looking for the limits and blockers, which are cheap to fix and invisible from a dashboard. Fix whatever surfaces within the week, and tell people you did, because visible responsiveness now buys you attention later.
Week four: harvest the wins. Find three specific things someone did that saved real time, and get those people to describe them in team meetings that already exist. Not a new session, not an email. A colleague saying "here is what I did with the renewal summaries last Tuesday" in a meeting people already attend.
Week six: expect the trough and interpret it. Usage is down. Pull the concentration number. Talk to people who have stopped, without any suggestion of monitoring or judgement. Sort what you hear into the four categories above.
Week seven: the refresher that matters. One short session per function, built entirely around the problems raised in week six. This is the highest leverage hour in the whole ninety days, precisely because people now have specific questions rather than general curiosity.
Week ten: reclaim and reallocate. Pull licences that have not been used in a month. Do it neutrally as housekeeping, offer them to people on a waiting list, and note that the offer alone often revives usage among those about to lose access.
Week twelve: report honestly. What worked, what did not, what it cost, what changed. An honest report with a mediocre number earns more credibility than an optimistic one, and it is what secures the next decision.
The intervention with the best return, by a distance: a peer demonstrating something specific and small in a meeting that already exists. Not the champion, not the enthusiast, not leadership. Someone whose job is like yours, showing a real thing from last week. It outperforms formal training consistently and costs four minutes.
Do not add process to recover adoption
When usage sags, the instinctive response is administrative: a mandate, a target, a usage report circulated to managers, a requirement to log what you used it for.
This reliably makes things worse. It converts a tool people were choosing to use into an obligation they are being measured on, which produces compliance behaviour rather than adoption. People will open it, do something trivial, and close it, and your numbers will improve while your outcome does not.
The things that do work are all forms of reducing friction or increasing visible peer proof. Fix the blocker. Show the example. Rebuild the assistant someone wanted. Answer the question. None of them involve measurement as a lever.
What good looks like at ninety days
Realistic targets for a mid-size organisation that has done this properly: somewhere around sixty to seventy percent of licence holders using it in a given week, most active users in it several times a week rather than occasionally, no single group carrying the average, unused licences reclaimed, spend inside the ceiling, and two or three tasks per function that people would now complain about losing.
That last one is the real test. Adoption is not a usage percentage. It is whether people would be annoyed if you took it away.
Frequently asked questions
Is a drop in usage after a month normal?
Yes, and it is the point where most rollouts get abandoned. The first fortnight is curiosity, not habit. What remains after week five is the set of tasks that genuinely fitted people's work. The job is to find out which tasks failed and whether they failed for a fixable reason.
What should we measure?
Reach, depth and concentration. Weekly active users against licences, sessions per active user, and the share of total usage coming from your top five people. The third is the one that hides failure, because a handful of enthusiasts can make a dead rollout look healthy on average.
When should we run refresher training?
Around day 45. In the first fortnight nobody has questions because nobody has hit anything hard. By week six people have real friction and abandoned tasks, so a short session built around their actual problems changes behaviour in a way a feature tour never does.
Should we mandate usage?
No. Mandates and usage targets produce compliance behaviour: people open the tool, do something trivial and close it. Your numbers improve and nothing else does. Reduce friction and increase peer proof instead, both of which change what people actually do.
Get past the point where most rollouts stall
We set it up properly, train your people by role, and build in the checkpoints that catch the week five drop before it becomes permanent, then hand over the administrator view and the numbers to watch. Fixed price, live in 30 days or less.