Where Your Team Actually Is

Most adoption plans describe where a team should be by a given date. Leaders who bring people along describe where the team actually is, and then decide what help fits. The two are often far apart, and the gap is where money and goodwill get lost.

Four stages, and what each one looks like

It helps to use four plain stages. In first trials, people have tried the tool, often once or twice, out of curiosity or because they were asked. In repeat use, a meaningful share of the team comes back to the tool without being prompted, on real tasks, week after week. In a changed way of working, the written procedure, the handover or the review step now assumes the tool is part of the task. In a measurable business result, an outcome that matters to the organisation, such as turnaround time or rework, has moved against an agreed baseline.

The stages are evidence-based, not aspirational. A launch event, a completed training course or a high login count tells you almost nothing about which stage you are in. Ask instead: do people return on their own, does the procedure now assume the tool, and is there a before-and-after measure? A team can also sit in different stages for different tasks, so name the workflow before you name the stage.

Matching support to the stage

Support that suits one stage can be wasted or harmful at another. Teams in first trials need low-risk chances to use the tool on real work, a named person to ask, and clear boundaries on what it may be used for. Teams in repeat use need help making the habit reliable: guided practice, shared examples and agreement on what a proper check looks like. Teams moving to a changed way of working need process redesign, with owners, updated procedures and a decision about what the human still does. Only when the workflow has changed does it make sense to measure a business result, because before that there is nothing stable to measure.

A common mistake is to send awareness material to a team that already uses the tool, or to demand a business result from a team still in trials. Both feel like action. Neither fits the stage.

Do not expand before the original use is consistent

Sponsors understandably want to take a promising pilot to more teams. The discipline to hold is this: do not expand until the original use is consistent. Consistent means that different people, including on busy days, apply the same checking step in the same way, and exceptions go to a named owner. If the first team applies a review step unevenly, a second and third team will inherit the unevenness and add their own. Expansion also removes the people best placed to fix the problem, because the pilot team's attention moves to training newcomers.

This is not a case for permanent delay. It is a case for a bounded condition: make practice consistent in the original team, keep a short record showing that, and then take a deliberately small second step. The record becomes the evidence a sponsor needs.

Awareness is not practice

Town halls, intranet pages and newsletters build awareness. They tell people the tool exists and why leadership cares. Practice is different: it happens when someone uses the tool on their own current work, makes mistakes, sees a colleague handle the same situation and adjusts. Awareness messaging is a reasonable start, but a leader who measures success by how many people saw the message has measured the wrong thing.

Worked example: Calder and Finch Housing

Calder and Finch Housing, an invented housing association, gave its repairs-triage team an AI tool that drafts first replies to tenants. After two months the sponsor planned a rollout to the lettings team. The initiative lead, Mara, looked for evidence. Logins were high, but when she sat with the team she found that about a third of staff had used the tool once and returned to templates. Of those who used it regularly, some read each draft closely and some sent it on a skim. The procedure document still described the old process.

Mara placed the team between first trials and repeat use, not at the stage the plan claimed. She recommended three things: hold the lettings rollout, run two practice sessions where the team reviewed real drafts together and agreed what a proper check involves, and update the procedure once the check was agreed. She told the sponsor that a business result could not yet be claimed and proposed a before-and-after measure on reply turnaround. The sponsor lost a month and gained a stable basis for expansion.

Two duties to keep in view

When AI is used in the workplace, the EU AI Act asks deployers to inform workers' representatives and the affected workers before use (Art. 26(7)). Even where a leader is unsure whether the duty applies, telling people early is good practice and avoids the sense that the tool arrived behind their backs. Art. 4 asks organisations to support AI literacy, as an obligation of means rather than result. It mandates no certificate and is proportionate to role and context, and the Commission reads it to cover contractors who operate systems on your behalf.

As of 24 September 2026.