The failure is scheduled, and you can see it coming
A champions network dies at around the three-month mark. It is remarkably consistent, and it has nothing to do with the enthusiasm of the people you picked. They were handed a badge and a Teams channel, and no remit, no time and no way to escalate anything they found. Under those conditions going quiet is the rational response.
The arc runs like this. Week one, a kick-off call with forty people on it and genuine energy in the chat. Weeks two to four, they answer questions from their teams, which is what you hoped for. Somewhere in week five a colleague asks something they cannot answer: why can Copilot see that document, why can it not see this one, are we allowed to put client material in it, when is the finance system going to be connected. The champion posts the question in the channel. Nobody with authority answers it, because nobody with authority is reading the channel. Two more of those and the champion has learned that raising things produces nothing. By month three the channel is programme updates and a monthly reminder to share success stories, posted by you.
A champions network does not fail from lack of commitment. It fails because the loop was never closed.
A champion is a sensor, not an unpaid trainer
The role gets designed backwards. Organisations recruit champions to push adoption outwards, when the valuable direction of travel is inwards. These are the only people who can tell you what is actually happening in the work, in language that has not been through three layers of reporting. That is worth more than any number of lunchtime sessions.
A good champion tells you that the bid team stopped using it because the drafts read wrong for a regulator. That the ops managers keep asking it about a system it cannot see, and have concluded the product is broken. That two directorates now have completely different rules about what goes into a prompt, because neither could find the policy. None of that appears in a usage report. All of it is fixable, and only while somebody knows.
The second thing they are for is being a visible local example. Not a demo. An example, in the actual work, from someone whose job the team recognises. That is the part nobody can outsource to a supplier, including me.
Stop asking them to do these things
Every one of these is a request I have seen made of volunteers with no time allocated:
- Deliver the training you decided not to fund.
- Answer licensing questions. They cannot, and a confident wrong answer here is expensive.
- Fix access and permissions problems. That is an administrator’s job with an administrator’s rights.
- Report on adoption numbers for a team they have no authority over.
- Act as first-line support, indefinitely, with no route into the service desk.
- Produce success stories on a monthly cadence, regardless of whether anything succeeded.
That last one does the most quiet damage. Asking for a positive story every month teaches the network that only good news is welcome, which removes the reason you built it.
What I would put in place instead
Four things, and the first is the one people skip.
Named time, in the calendar, agreed by their line manager, with the manager knowing what it is for. An hour a fortnight is enough and it has to be real. A champion whose day job is unchanged is a champion who is doing this in their own evenings, and the goodwill runs out at around, as it happens, three months.
A route to someone who can decide. One named person, a stated response time, and a visible record of what happened to each question. The record matters more than the speed. People will accept “we looked at this and we are not doing it” far longer than they will accept silence.
Something they receive before everyone else. A short note on what changed in the product, what the programme decided, what is coming. It gives them a reason to open the channel and it gives their teams a reason to ask them things.
Explicit permission to say the tool cannot do it. If champions are only allowed to be positive, their colleagues stop trusting them, and you have converted your most credible people into marketing. The honest ones will refuse the role rather than do that, which is how you lose exactly the people you needed.
Why individual enthusiasm does not compound on its own
Microsoft’s 2026 Work Trend Index is a first-party survey and should be read as one: 20,000 knowledge workers across 10 markets, fieldwork run by Edelman Data x Intelligence between 18 February and 7 April 2026. With that label attached, one finding is worth the attention. Only 19% of AI users are what the report calls “in the Frontier”, where organisational capability and individual readiness reinforce each other.
That is the champions problem stated in survey form. You can recruit the readiest individuals in the organisation, and if the capability around them does not move, their readiness has nowhere to go. The network is not a substitute for the operating model. It is an instrument for finding out where the operating model is broken, and it stops working the moment people learn that what they report goes nowhere.
If you have a network that has gone quiet, the recovery is not a relaunch. It is going back to the last five things they raised and answering them, publicly, including the ones where the answer is no.