The champion cascade

The situation

The first cohort went well. Thirty days later you name twelve champions at an all-hands, add a tag to their internal profiles, and open a channel. For three weeks the channel is loud. Then it becomes a queue: someone cannot get a licence, someone wants to know if customer records are allowed in, someone's export is broken. The champions answer, because they were named in front of everyone and cannot be seen to ignore it. None of this is their job and none of it is visible to their manager. By week nine the two best ones have stopped replying. One asks, politely, to be taken off the list.

You did not lose them to disinterest. You gave twelve people an unbounded obligation and called it recognition.

The principle

An unbounded obligation with a title on it is not recognition — a champion carries a rollout only when the ask has a named group, an end date, and a list of the questions they are allowed to refuse.

A cascade works because the second wave gets taught by someone who does their job, in their words, on their systems. That is a real advantage and it does not survive being turned into a support rota. The champion's credibility comes from being a practitioner; the moment they become a helpdesk they stop practising, and within two months they are a person who once used the tool.

So the charter is the whole intervention. Scope says which people. Time says how many hours, agreed in writing with the person who reviews them, and an end date. The refusal list says which questions leave their hands immediately, and to whom.

Worked example

Shown here in Gemini.

Pick on evidence, not on volume. You already have the evidence: every person in the first cohort built something, and thirty days later some of those things are still in use. Save this as a Gem so the same criteria apply each cohort.

Job: Shortlist 8 candidate champions from 34 people in the first cohort.

Inputs: One row each — role, team, site, the asset they built, and their
own one-line note on when they last used it. Ignore attendance,
feedback scores, and job title seniority.

Output: A ranked table of 8. Columns: person, team, the asset, the one
sentence of evidence I would read to their manager, and one risk of
picking them. Then a second short list headed "Coverage gaps": teams
where nobody built anything, named as gaps, not as champions.

Standard: Rank first on repeated use of their own build, second on
covering a team that has nobody else. Never rank on enthusiasm or
seniority. Where there is no evidence of reuse for a person, write
"no evidence" rather than inferring it.

The risk column is the one to read twice. It is what stops you naming the person whose team is about to lose two people, or the one who built something excellent that nobody else in their team could operate.

Then the charter, one page, one champion:

Champion: <name>          Scope: the 9 people in <team>, nobody else
Runs: 8 weeks, ending <date>

The ask
  1. One 30-minute clinic a fortnight, on the team's own live work
  2. Keep the team's shared instruction current as the work changes
  3. Send me one thing that broke, every Friday, in one line

Time: 2 hours a fortnight, agreed in writing with <manager> on <date>

Not their job — route to the named owner, same day
  licences and access ......... <name>
  what data is allowed in ..... <name>
  policy and legal ............ <name>
  the tool is down ............ <name>

Renewal: expires on the date above; renewed only if they ask.

The refusal list is why the good ones are still there in month three. The expiry date is why saying no later costs them nothing.

In your tool

The second wave should never retype what the first wave already got right. What you are looking for is how one person's working setup reaches the nine people around them.

Product surfaces checked 2026-08-04.

Try it

Take the list of people you would name as champions tomorrow. Pick one. Write only the two hardest lines of their charter: the scope — an exact number of named people — and the refusal list, with a real name beside every item.

You are done when no item on the refusal list has the champion's own name next to it, and when you would be comfortable sending the charter to their manager unedited. If you cannot fill a name in beside "what data is allowed in", you have found the thing to fix before you name anyone.

Common failures

• The volunteer badge. Champions self-select, so you get the people most excited by the tool rather than the people their colleagues actually copy. Enthusiasm is easy to spot and weakly correlated with influence; the quiet person whose spreadsheet everyone borrows is the one you wanted.
• The unpaid helpdesk. No scope, no hours, no refusal list — so every question in the company routes to the person whose name was on the slide. The work is invisible to their manager and competes directly with the job they are reviewed on, so it loses. They do not complain, they just answer more slowly and then not at all.
• The champion who cannot change anything. They can demonstrate a better way to produce the report, but the step, the template and the sign-off belong to someone else. The team watches a good demo, changes nothing, and concludes the tool does not fit their process — when what did not fit was the authority you gave the champion.