AI adoption is a community problem, not a training problem
As organisations race to roll out ambitious AI solutions, many fall into the same trap.
A new tool is procured. Licences are issued. Training sessions are booked. Guidance is published. A community space appears in Viva Engage, Teams or somewhere similar, usually described as a place for questions and sharing.
Then everyone waits.
The expectation is that adoption will emerge naturally. Colleagues will experiment, discover useful applications and gradually build new habits around the technology. But that rarely happens.
In Chapter 9 of Digital Communications at Work, we describe this as the "empty dance floor" problem. A community is launched, the space is technically available, but everyone waits to see who goes first.
The music is playing. The glitterball is spinning. But nobody wants to be the only person throwing shapes. The platform exists. The purpose doesn't.
The usual response is predictable: more training, more communication, more reminders that the tool exists. But that assumes the problem is awareness.
More often, people aren’t trying to understand the technology. They're trying to work out if it's genuinely useful, whether it's safe to use, and if people like them are successfully incorporating it into their work.
That decision is not made in isolation – it’s made socially. Which is why AI adoption is less of a training issue than many organisations expect, and more a community challenge.
As we argued in our earlier post on the five-layer model for internal communications platforms, communities are not an optional extra in the digital workplace ecosystem. They're the Discuss layer through which people learn, share knowledge and make sense of change.
Adoption is social, not individual
Most organisations approach AI adoption as though it’s an individual learning problem.
Give people access. Show them how it works. Provide some guidance. Trust that adoption will follow. But technology adoption rarely works like that.
Drawing on Jennie Carroll's Model of Technology Appropriation, people tend to move through three stages when adopting new technology:
encounter
adaptation
integration
In other words, people first discover the tool, then experiment with it, and eventually incorporate it into the way they work. Digital Communications at Work uses this model as part of its framework for understanding digital adoption.
The challenge is that most enterprise rollouts focus heavily on the first stage and assume the rest will happen naturally. But it doesn't.
What moves people from experimentation to adoption is not simply individual skill. It’s seeing people like themselves using the technology successfully. A finance colleague sharing a workflow that saves an hour a week. A project manager showing how they summarise meeting notes. A recruiter explaining how they improved a job description.
These examples matter because they make the technology feel legitimate, practical and relevant.
We can see the same principle in Microsoft's own Copilot rollout. Through its Copilot Champs community, more than 7,000 early adopters, AI enthusiasts and peer leaders shared prompts, workflows and practical advice through Viva Engage. Microsoft reports that community members recorded significantly more active Copilot usage than colleagues outside the community. The technology mattered, but so did examples of peers finding value in it.
This is also why risk-first messaging can be counterproductive. Organisations understandably want to explain the dangers of AI. But when the dominant message becomes "here's what you must not do", the safest response is not using it at all.
The result is familiar – plenty of awareness, little adoption.
Start with a question people actually care about
One of the most useful lessons we've picked up from recent client conversations is surprisingly simple.
Don't launch a generic AI community and hope people find a reason to participate. Start with a question. Specifically, a question that's both high-stakes and high-curiosity:
How are colleagues actually using AI in their work?
That question gives the community an immediate purpose. It also reflects the broader principle we explored in our earlier post on discovery versus audits: start with what people are trying to do, not the tools you've already bought.
People rarely join a community just because the space exists. They participate when they can see a reason to do so, and when the value of taking part is obvious.
A general-purpose AI community sounds sensible, but often becomes a home for policy documents and unanswered questions. Conversations drift. Participation stalls. The space gradually fills with announcements rather than useful discussion.
But a community built around real examples of work creates something different. It creates relevance.
People do not join because they’re interested in AI as an abstract topic. They join because they want to know whether somebody else has already solved a problem they’re facing.
The community generates the adoption. The platform simply hosts it.
A good example comes from Daiwa Institute of Research. The organisation initially rolled out 300 Copilot licences with mixed results. Adoption remained patchy until it shifted towards an ambassador-led community model, encouraging colleagues to share prompts, use cases and lessons learned through Viva Engage. The community quickly became one of the most active internal spaces in the organisation. Licence numbers grew to 750, a waiting list emerged for additional access, and the community itself became the engine of adoption.
A practical operating model: surface, synthesise, steward
Organisations making progress tend to follow a pattern. The details vary, but the underlying motions are remarkably consistent.
The first is surface.
Create opportunities for colleagues to share prompts, workflows, experiments and examples that genuinely helped them. These do not need to be polished success stories. In fact, the rougher examples are often more useful because they feel achievable.
The goal is not perfection. The goal is visibility.
As Etienne Wenger's work on communities of practice demonstrates, people learn through participation and observation as much as formal instruction. Seeing peers solve real problems creates a powerful signal that formal training rarely achieves.
The second motion is synthesise.
Not everyone will spend time scrolling through a community feed. Valuable ideas quickly disappear when left buried in conversations.
Someone needs to turn the emerging patterns into a regular editorial digest:
What are people doing with AI?
Which use cases are gaining traction?
What mistakes are proving instructive?
What seems genuinely useful?
This serves two purposes. It spreads ideas and reassures people that meaningful progress is happening. As we explored in our post on internal comms distribution, publishing information is only half the job. Valuable lessons only spread when they reach the people who can use them.
Variations of this pattern are increasingly appearing in large organisations. For example, Pfizer has invested heavily in building AI fluency through shared learning, curated resources and peer-to-peer knowledge exchange. Rather than treating AI as simply another technology rollout, the emphasis has been on helping colleagues understand, experiment with and learn from one another. The principle is the same: make successful use visible, help people learn socially, and turn individual experimentation into organisational capability.
The third motion is steward.
Governance matters. Risk matters. But they work best when they’re visible, human and conversational. That governance role becomes especially important with AI. As we discussed in our governance post about why intranets fail, the challenge is not creating control for its own sake, but creating enough clarity and confidence for people to participate safely.
Rather than publishing AI policies into the void, create opportunities for colleagues to ask questions openly. Bring together representatives from IT, risk, knowledge management and practice areas. Let people see the conversations happening.
Good governance should reduce uncertainty, not create it. Communities thrive when leaders are visible, moderation is light-touch and participation feels safe.
The community is part of the rollout
The most common mistake is treating the community as something that sits alongside the rollout. A support channel. A place for questions. A nice extra.
But if adoption is social, then the community is doing much more than answering queries. It’s creating the conditions that allow adoption to happen.
People discover possibilities through peers. They build confidence through examples. They learn through participation. They gain reassurance from seeing others navigate the same uncertainty.
This is why successful AI communities produce outcomes that look disproportionate to their size. What spreads is not information. What spreads is belief that the technology has a place in everyday work.
And that’s ultimately what adoption looks like. Not dashboard metrics. Not licence counts. It’s the moment somebody changes how they approach a task because they saw a colleague do it first. The moment experimentation becomes habit. The moment the technology stops feeling new and starts being useful.
If, as we argued in Chapter 9, communities are strategic infrastructure, then AI adoption may be their most important test yet.
Because the community is not there to support the rollout. It’s the rollout doing its job.
Digital Communications at Work (Kogan Page, July 2026) is available for pre-order now. In Chapter 9 we examine in depth communities as strategic infrastructure, the role of participation in organisational learning, and how communities help new behaviours spread through organisations.
