What single thing most predicts adoption?

Whether the old way is still available. A tool introduced alongside an existing method that still works will be used by the people who like new things and by few others, because the alternative is known, adequate and requires nothing to be learnt.

Why is this harder for AI tools than for other software?

Because there is frequently nothing to retire. A tool that helps slightly with many tasks rather than replacing one completely offers no obvious switch to flip, so adoption depends on individuals changing dozens of small habits that nobody can identify in advance or measure afterwards.

When should training happen?

When somebody has the problem rather than at launch. A session before anybody has hit the situation the tool addresses arrives at the moment people are least able to absorb it, and the useful version is help available at the point of being stuck.

Glossary

Change management

The work of getting people to actually change how they work, which is usually where a tool succeeds or quietly does not.

In plain terms

Everything between buying the tool and people using it. It gets treated as communication, and the part that decides the outcome is whether anybody's existing way of doing the job was actually taken away or merely joined by a second option.

Why it matters

Because in this category the technology usually works and the adoption usually does not, so the money is at risk in a place nobody assessed during procurement. A tool that does what it claims can sit unused for a year while the licence renews, and nothing in the purchase process was designed to notice that.

How it works

The default is that the old way survives alongside the new one, and that decides most outcomes on its own. If the previous method still works, most people keep using it under any pressure, because it is known and the new one is not, and no amount of encouragement competes with a working alternative that requires no learning.

The word rollout suggests announcing rather than removing, which is why the announcement is what usually happens. Communication is the visible part and the effective part is a decision about what is being retired, which is a harder conversation and belongs to somebody with the authority to make it.

The people who adopt first are the least informative about whether it will work. Early enthusiasm comes from people who would have found a way to use the tool regardless, and the question that predicts the outcome is what happens with the person who was managing perfectly well before.

Training is treated as an event and behaves like a habit. A session at launch reaches people before they have a problem the tool solves, which is the moment they are least able to absorb it; the useful version is available at the moment somebody is stuck, which is a different shape of provision entirely.

AI tools have a specific difficulty here that older software did not. There is often no obvious task to switch over, because the tool helps with many things slightly rather than replacing one thing completely, so there is nothing to retire and adoption depends entirely on individuals changing habits nobody can point at.

Two rollouts of the same tool

What makes this worth stating plainly is that the left-hand column is not a failure of effort. It usually represents a genuine, well-run launch with material prepared and sessions delivered, and it produces a tool that a minority use enthusiastically and the majority remember being told about. The right-hand column frequently involves less activity and one uncomfortable decision. That decision is uncomfortable because retiring a working method imposes a real cost on people who were managing fine, and the person making it has to accept that cost knowingly rather than hope adoption happens without it. The AI category makes this harder in a way worth naming rather than glossing: for a great many of these tools there is genuinely nothing to retire, because the tool improves many tasks slightly instead of replacing one completely. Where that is true, the honest expectation is slower and patchier adoption, and the useful response is to stop measuring rollout as though a switch had been flipped and start looking at whether the people who never volunteered have found a reason of their own.

Seen in the wild

An assistant rolled out alongside every existing habit, so nothing was retired and use stayed with volunteers.
ChatGPT
A search deployment that took hold because the old intranet search was switched off.
Glean
An automation that stuck because the manual process it replaced was genuinely removed.
Make

Common misconceptions

People assume

It means communicating the change well.

In fact

Communication is the visible part and rarely the deciding one. What decides adoption is whether the previous way of working is still available, because a working alternative that requires no learning defeats any amount of explanation.

People assume

Early enthusiasm shows it is working.

In fact

Early adopters would have found a use for the tool without any help, so their behaviour predicts little. The informative case is the person who was managing perfectly well before and has no particular reason to change.

Questions

What single thing most predicts adoption?: Whether the old way is still available. A tool introduced alongside an existing method that still works will be used by the people who like new things and by few others, because the alternative is known, adequate and requires nothing to be learnt.
Why is this harder for AI tools than for other software?: Because there is frequently nothing to retire. A tool that helps slightly with many tasks rather than replacing one completely offers no obvious switch to flip, so adoption depends on individuals changing dozens of small habits that nobody can identify in advance or measure afterwards.
When should training happen?: When somebody has the problem rather than at launch. A session before anybody has hit the situation the tool addresses arrives at the moment people are least able to absorb it, and the useful version is help available at the point of being stuck.