Glossary
AI adoption
Adoption is how far a tool has genuinely entered daily work, which is a different measurement from how many people were given a licence or logged in once.
In plain terms
Everyone has access and the invoice reflects it. The question adoption asks is what proportion of those people reach for the tool when the relevant work appears, which is usually a much smaller number and a much more interesting one. Access is something the organisation controls directly. Use is something people decide, individually, every time the situation comes up.
Why it matters
Because it is the gap where the return quietly disappears, and because it is routinely reported as though it were solved. A tool bought for a department and used by a handful of enthusiasts is producing a fraction of the benefit at the full price, and the licence count on the report says otherwise. Worse, the pattern is self-concealing: the people who report on it are usually the people using it, so the picture that reaches a decision is drawn from the part of the organisation where it went well.
How it works
Access, activation and habit are three different things and only the last one is adoption. Everybody has a licence, some proportion tried it, a smaller proportion now reaches for it without being reminded. Reporting the first number as though it were the third is the commonest overstatement in this subject, and it is usually not deliberate: the first number is the one that is easy to get.
Use per task is more informative than use per person. A tool nobody opens on Monday but everybody opens when a particular kind of work appears is well adopted for that work, and a monthly active count will make it look neglected. Deciding which task the tool was bought for, and then asking what share of that task it touches, produces a number that means something.
People stop for reasons that are rarely about capability. It was slightly faster to do it the old way, the output needed more checking than expected, the tool sat one click too far from where the work happens, nobody was sure whether it was allowed. Each of these is fixable and none of them shows up as a complaint, because the person simply stops rather than reports.
Where a tool sits matters more than most rollout plans assume. Something reachable inside the application people already have open gets used; something requiring a separate tab and a separate login gets used by people who are motivated. That is a statement about attention rather than about laziness, and it is the reason integration into an existing workspace often beats a better standalone product.
The ceiling is usually the work rather than the people. Not every task benefits, and a tool at thirty per cent adoption may have reached everybody whose work it actually suits. Chasing a higher number past that point produces mandated use, which shows up in the statistics and not in the results. Knowing which is the case requires asking the people who stopped, which is the step that gets skipped.
Two reports on the same tool
Seen in the wild
Compare how many people hold a licence for an assistant against how many use it for the specific drafting task it was bought for.
ChatGPTWatch whether internal search is used for the questions people previously gave up on, or only for the ones they would have found anyway.
GleanNotice that a tool living inside the workspace people already have open is reached for more readily than one in a separate tab.
Notion AI
Common misconceptions
People assume
Everyone has access, so it has been adopted.
In fact
Access is the input to adoption and is routinely reported as the output. The number worth having is what share of the relevant work the tool now touches, which is smaller, harder to obtain and the only one that connects to any return. Licence counts are easy to produce, which is precisely why they are the ones that get produced.
People assume
Low adoption means the tool is not good enough.
In fact
More often it is friction or permission. It sat behind a separate login, the output needed more checking than the old way, or nobody was certain it was allowed for customer material. Those are cheap to fix and invisible from usage figures, which is why asking the people who stopped is worth more than another round of training.
People assume
More adoption is always better.
In fact
Only up to the work the tool actually suits. Past that point the numbers rise because people are complying, which produces activity without benefit and makes the statistics less informative rather than more. A plateau at the boundary of the relevant work is a success being misread as a shortfall.
Telling them apart
Adoption vs Rollout
Adoption
What people actually do afterwards. Measured, and mostly outside your control.
The act of giving everyone access, with training and communication.
A rollout can be complete and adoption near zero. The reverse almost never happens.
Questions
- What should we measure?
- The share of the specific work the tool was bought for that it now touches, rather than monthly active users. A tool used heavily but only when one kind of task appears looks neglected on a per-person count and is well adopted on a per-task one, and the second is the number connected to any return.
- Why do people stop using a tool that works?
- Usually friction rather than quality: a separate login, output that needed more checking than the old method, or genuine uncertainty about whether it was permitted for that material. None of these arrives as a complaint, because people quietly revert instead of reporting, which is why the reasons have to be asked for directly.
- Is mandating use a reasonable answer?
- It moves the statistics and rarely moves the results. Where adoption has stalled because of friction, removing the friction works; where it has stalled because the work does not suit the tool, a mandate produces compliance activity that costs time and yields nothing. Either way the mandate treats the symptom.
- What adoption level should we expect?
- There is no useful general answer, and figures quoted as benchmarks usually measure access rather than habit. The honest version is local: decide what work the tool was bought for, measure what share of it the tool now touches, and compare that against the same measurement a quarter later rather than against somebody else's number.
Key takeaways
- Access, activation and habit are three numbers, and only the third is adoption.
- Measure the share of the intended work the tool touches, not people per month.
- People stop for friction and uncertainty far more often than for capability.
- A tool inside the application people already have open beats a better one in a separate tab.
- A plateau at the edge of the suitable work is a success, not a shortfall.
Last checked July 2026