How do you measure whether champions spread useful practice?

Measure a champion network by following useful practices into other people's work. Ask colleagues what help they received, what they changed, whether the result met the task's quality standard and what happened at the next suitable opportunity. Keep that evidence attached to the support provided and the other changes that could explain the result.

A busy demonstration calendar shows effort. To decide whether that effort deserves more time, an adoption lead needs to know what colleagues can now do with it. This guide offers a small observation record for that decision, without assuming every improvement belongs to the champion network.

Use the people and change guide for the wider conditions that support adoption. If the network is still taking shape, the champion network launch guide covers responsibilities and support.

Define the practice you hope will spread

Start with one piece of work and the standard it must meet. “More people use AI” leaves too much open: someone can attend a session, try an irrelevant prompt or generate an output nobody can use. A specific practice gives you something to follow.

Suppose a 220-person property-maintenance company wants its facilities coordinators to prepare clearer customer emails after repair visits. A coordinator uses an approved AI tool to draft an email from the technician's visit notes. Before sending it, the coordinator checks what was repaired, what remains unresolved and whether another visit has actually been agreed. A polished email that wrongly says a repair is complete fails that check.

The champion's contribution might be showing how to separate completed work from outstanding work, helping a colleague adapt the prompt, or explaining when the notes are too incomplete to draft from. Record which help actually happened. The champion producing everyone's emails would be a different operating arrangement from colleagues learning the practice themselves.

OpenAI Academy's guidance on measuring a local champion network includes workflows tested by another team and examples of improvement. It calls these directional signals and notes that access, manager support and readiness also influence outcomes. That is vendor guidance, rather than a measured estimate of champion effectiveness.

Follow the example into a colleague's work

Ask the recipient to walk through a recent attempt. A prompt saved in a library is a useful starting point, but it cannot show what the colleague understood, changed or checked.

In the maintenance example, a colleague notices that some visit notes mention a return visit without a date. They adapt the shared prompt to flag the missing date for follow-up instead of suggesting one. The coordinator still checks the draft against the original notes and confirms arrangements before making a promise to the customer.

Useful adaptation preserves the important checks. Repeating the champion's exact wording is not the goal. A change that handles a real exception can make the practice more useful; a change that removes the completion check can make it worse. Examine both the adaptation and the resulting work.

Follow-up can reveal why a promising example failed to travel. In a June 2026 Reddit discussion, one Reddit user described creating workflows that proved less useful than expected. Asked what went wrong, they explained that colleagues preferred their own approaches, making the situation more complicated. The employer and type of work were unspecified. It is one person's account, but it points to a useful question: what did the intended recipients actually choose to use?

Agree a follow-up when the work is likely to recur. For the maintenance coordinators, that could be the next suitable completed visit. A finance analyst preparing monthly expense commentary, an explanation of spending against the budget, needs a different interval. If the opportunity has not occurred, record that rather than treating silence as failure. The retention guide explains how opportunities affect repeat-use interpretation.

Keep a small evidence record

Choose the practices and colleagues you will follow before asking champions for their best success stories. Include people who tried an example and stopped, as well as those who continued. If you only hear from volunteers who enjoyed the session, describe that limit when reporting the findings.

Use a short record like this one. It can be completed through a conversation and a permitted walkthrough of the work; you do not need to copy the underlying customer records into it.

Keep evidence strength visible. “The colleague said it helped” and “the colleague showed a checked output and explained a later reuse” answer different questions. Both can be useful, provided your report preserves the difference. Missing follow-up remains unknown.

If you summarize counts, define the group and period. Count distinct colleagues applying a practice separately from sessions attended or help requests answered. Explain how many people were invited to follow-up, how many responded and how many had an opportunity to use the practice. A handful of reviewed examples can guide the next support decision, but it does not establish a company-wide adoption rate.

Microsoft's feedback guide for champions recommends task-specific questions, several feedback channels and conversations with skeptics as well as power users. It also asks which scenarios people stopped using. Those prompts can help you understand the missing or disappointing entries in the record.

Consider what else changed

A colleague's account may show that a champion helped them overcome a specific difficulty. It rarely tells you how much adoption would have happened without that help.

Check two common alternative explanations:

• Other support arrived at the same time. Suppose managers gave the maintenance coordinators time to practise and IT enabled the approved tool in the week the champion shared the prompt. Later use could depend on all three. Record the sequence and ask which obstacle each change removed.
• The teams started in different positions. Volunteers may already be more interested or experienced. Managers may assign champions to especially ready teams or those facing the hardest problems. A higher or lower usage rate can reflect those differences.

A 2022 systematic review of 35 healthcare studies found mixed or limited evidence for several champion outcomes and highlighted weak descriptions of champion activities and exposure. Its setting does not establish what works for enterprise AI. Its methodological lesson is relevant: define what champions did and examine the surrounding conditions before attributing results to them.

For a practical internal review, report the contribution you can support: the recipient identified the help, showed how it changed their work and explained what else mattered. HM Treasury's guidance on contribution analysis describes testing an expected chain of results alongside other explanations. It provides evidence for a reasoned contribution claim, rather than definitive proof of a causal effect.

Protect the trust that makes peer help possible

Explain what you are collecting, who will see it and what decision it will inform. For this observation record, collect only the information needed to improve support. Agree how long to keep it and who can access identifiable notes.

A coordinator can demonstrate the missing-date check using an approved, redacted example. Customer names, addresses and private messages do not belong in a network progress report. Where a small team's aggregate would still identify someone, combine reporting or leave the detail out.

Give colleagues a way to describe problems to someone other than the champion who helped them. They may otherwise soften criticism out of politeness. Avoid public league tables of champions: departments differ in access, work opportunities and support needs, and a count can reward easy cases over difficult but worthwhile help.

The practical question is where support needs improving. It should remain possible to say that an example did not help, that another method worked better or that AI was unsuitable for the task.

Use the findings to change support

Bring a few contrasting observations to the network review. Pair evidence of transfer with the effort needed to support it, using the champion workload diary and time agreement. More questions resolved can also mean more work for a champion whose ordinary responsibilities have not changed.

Choose an action that matches what you found:

• Colleagues can adapt and reuse the practice. Preserve the useful checks and let another relevant team test the example. Keep checking quality as the setting changes.
• People attend but cannot apply it. Ask recipients to show where they got stuck. Improve the example or coaching around that difficulty.
• The approach depends on the champion doing the work. Clarify whether the team needs a service, more learning support or a simpler practice.
• Access or manager capacity prevents attempts. Give the issue to the responsible owner and revisit after the condition changes.
• Checked attempts are still unhelpful. Revise or retire the example. A sensible decision to stop can be useful learning.

Keep a short conclusion naming the evidence, its limits and the next action. For the maintenance team, a justified next step could be testing the missing-date check with another coordinator. A claim that the network caused a percentage increase in productivity would require different evidence. The AI value guide covers the broader assessment of quality, effort and business outcomes.

Questions about measuring champion networks

Which metrics show whether an AI champion network is working?

Choose measures that connect the network's help to useful work by other people. For a shared maintenance-email practice, that means knowing whether coordinators applied the approach, checked the draft against visit notes and returned to it at another suitable visit. Attendance and help requests describe activity; they cannot establish those results alone. Keep the group, period, response coverage and evidence strength beside any count. Use the small evidence record to start with a practice you can actually follow.

How can we tell whether a champion caused higher AI adoption?

Higher use after champion support is evidence of a change, but it does not establish the cause. The recipients may already have been motivated, or new access, training and manager support may have arrived at the same time. Document what the champion helped with and examine those other explanations. If you need a numerical estimate of the champion network's effect, plan an appropriate evaluation with specialist help. The section on other changes explains why a simple comparison between teams is insufficient.

Does changing a shared prompt mean the practice failed to spread?

Changing a prompt can be evidence that a colleague understood and adapted the practice. In the maintenance example, adding a flag for an unconfirmed return-visit date makes the draft better suited to the team's records. The important test is whether the adaptation preserves the quality checks and produces useful work. Removing the check that distinguishes completed repairs from unresolved issues would be a different result. Review the change with its recipient using the follow-up sequence.

How can we measure progress without monitoring everyone's AI use?

Follow a defined set of relevant practices through brief recipient conversations and permitted, redacted examples. Explain the purpose, limit identifiable notes and record whose experience is missing. This can show where support helps or breaks down, although a small selected sample cannot establish an organization-wide adoption rate. Collecting private prompts or tracking every employee is unnecessary for that narrower decision. The trust and collection guidance explains what to keep out of the progress report.