How we get teams using AI
AI adoption for communications teams
Most AI rollouts start by training a power user and routing everything through them. We used to recommend that, and changed our position. Get every person competent at the one workflow they use daily, then let small groups form around what people actually need to do. This is the method we run with agency and in-house comms teams, and the reasoning behind it.
The shift
We changed our own advice on this, and the reason is worth borrowing.
What we used to recommend
Train a power user, route everything through them
Pick the most capable or most enthusiastic person, make them the expert, and let the rest of the team ask them. It concentrates the learning curve, which feels efficient, and it concentrates the capability, which is the problem.
What we recommend now
Baseline for everyone, then sub-groups on top
Everyone gets competent at the daily workflow. Depth then forms in small groups around real requirements, wherever the genuine interest turns out to be. Several partial experts beat one complete one, because none of them is a single point of failure.
The method
Four things that decide whether it takes
Baseline competence first, for everyone
Every person, without exception, gets comfortable with the core workflow end to end - not a tour of features, one job done start to finish. Most of the value comes from everyone being competent at the thing they do daily, not from one person being expert at the things they do rarely.
Sub-groups form around a real requirement
On top of the baseline, small groups form around what people actually need: watching the news cycle and reporting on it, keeping announcement flows running cleanly, reviewing performance and measurement. Membership follows the requirement, not the org chart, and the groups do not need naming up front.
Pair every enthusiast with a sceptic
Pair whoever is moving fastest with a colleague who does not reach for AI tools first. The sceptic forces the enthusiast to reconstruct how they got from A to Z - and people moving fast with AI assistance often cannot. Unarticulated reasoning is lost judgement.
Name a Decider
Exploring everything is a waste of time. Somebody close enough to the work to see all of it decides which lines of exploration are worth pursuing, so effort compounds instead of scattering. The goal is a deeper understanding of what you already have, not more tools.
Why not power users
Two reasons we stopped recommending the internal champion.
A single trained generalist is a single point of failure. When they are on leave, or they leave, the team's capability goes with them. Several partial experts give you redundancy, and the cost of getting there is lower than it looks because the baseline is the same session for everybody.
People arrive with different use cases anyway. Genuine interest predicts who will go deep far better than a job title does. And exploration is transferable: what one person works out for their own problem gets reused by somebody else for a different one, which is the compounding you lose by routing everything through one desk.
The corollary is the part teams find hardest to accept. Your less engaged people are not a remedial cohort to throw extra training at - they are the judgement cohort. Treat them that way and you get better output from both sides of the pair.
What it asks of you
Four roles, and only one of them is everybody
| Who | How many | Chosen by |
|---|---|---|
| Baseline competence | All users | Non-negotiable. Everyone who will send. |
| Sub-groups | As they emerge | Requirement first, genuine interest second, never seniority. |
| A paired sceptic | One per enthusiast | Someone who does not go to AI tools first. |
| The Decider | Your account and hub leads | Close enough to the work to see all of it. |
Whoever ends up in the performance and measurement group owns the data between review check-ins, so make sure that group has a name against it.
FAQ
Common questions
What is AI adoption?
AI adoption is the process of getting a team using AI tooling as part of its normal work, rather than having the licences sit unused. It is mostly a people and process question rather than a technology one: the constraint is rarely the tool's capability, it is whether each person is confident doing their own job with it and whether what one person learns reaches anyone else.
Should we train a power user or an internal champion?
We used to recommend it and changed our position, for two reasons. A single trained generalist is a single point of failure - when they are on leave or they leave, the team's capability goes with them, and several partial experts give you redundancy. And people arrive with different use cases anyway: genuine interest predicts who goes deep far better than a job title does. Spread the baseline, then let depth form where the real requirements are.
How do you train a communications team on AI tools?
Start with the single workflow the team runs most often and take everyone through it end to end - for a comms team that is usually building a media list, drafting, personalising, sending, and reading what came back. One session per stage of the rollout, in the order things actually go wrong, followed by refreshers. Only once that baseline exists do you layer sub-groups on top for news monitoring, announcement flow, and measurement.
Who should own AI adoption internally?
Split it in two, because the halves want opposite things. Who may hold administrative access is a permissions question and wants one accountable owner. Capability is a knowledge question and wants spreading as widely as possible. Concentrating the second is what creates the single point of failure; fragmenting the first is what makes access impossible to manage.
What about the people who are sceptical about AI?
They are the most useful people in the room, and treating them as a remedial cohort to throw extra training at wastes them. Paired with an enthusiast, a sceptic forces the reasoning to be made explicit, which is what turns one person's shortcut into a process the rest of the team can follow. Remove all the friction and you remove some of the thinking with it.
How do you know AI adoption has actually worked?
Not by licences issued or logins created. A team is only ready when somebody in it has done the real job with the tool end to end and got a real result back - for a comms team, sent a real pitch and received a real reply. Access, training attendance and dashboard activity all look like progress and none of them prove the workflow runs.
The worked example
This method, applied end to end, in public
A method is easy to agree with and hard to run. So the whole thing is written up as a working guidebook: twelve stages across three phases, who owns each one, what has to be true before each starts, and the eight sessions the enablement is actually built from - ordered by what caused the most trouble on real rollouts rather than by feature.
It is the document we hand to an agency team on day one. It is unusually specific about the parts that go wrong, because those are the parts nobody writes down.