Adopt —
The Human Layer

Mandate is seductive as a deployment strategy. You have the licences, you have leadership sign-off, you have a deadline. You send the email, you add the tile to the M365 launcher, you declare the rollout complete. And then, six months later, your admin data shows that 60% of users have opened Copilot fewer than five times, and the 40% who use it regularly were the same people who would have found it on their own.

The CIPD "People Profession 2024" data is unambiguous: 45% of UK employees are worried that AI will significantly change their role. Only 18% say their employer has explained what that means for them. That 27-point gap is not a failure of communication — it is what mandate looks like from the receiving end. An announcement without explanation reads as a change being done to people, not with them. And people who feel something is being done to them do not adopt it; they comply minimally and wait for it to go away.

Bottom-up adoption movements look messy and slow at first. They depend on peer credibility rather than authority. They require patience with early adopters who go off-piste and try things that do not work. But they produce something that top-down mandate cannot: genuine behavioural change that persists after the launch campaign ends, because the people involved chose it rather than being told to do it.

The practical implication is that your deployment infrastructure needs a social layer, not just a technical one. You need places where people share what they are learning. You need visible leadership engagement that is authentic rather than scripted. You need a mechanism for capturing and amplifying the peer stories that build credibility with the unconvinced. In the Microsoft ecosystem, that infrastructure is Microsoft Viva Engage.

Microsoft Viva Engage — the enterprise social network embedded in Teams and the Microsoft 365 suite — is the most underused infrastructure asset in most organisations' AI adoption toolkit. It is already licensed for most M365 E3 and E5 tenants. It integrates with Teams, SharePoint, and Outlook. And it provides exactly the social architecture that adoption programmes need: communities, storylines, campaigns, and leadership publishing.

The first use case is the AI Champions Community — a Viva Engage community dedicated to AI adoption, run by and for your champions network. Not a formal announcement channel. A genuine peer space where people share prompts that worked, report tasks where AI fell short, ask for help with specific use cases, and celebrate wins. The community is anchored to the real work — it is not a discussion about AI in the abstract but a practical exchange of what is and is not working in this organisation, in these roles, on these tasks.

The second use case is the AI Win of the Week — a recurring post, published by a different champion each week, describing one specific way they used AI that saved time, improved quality, or revealed something they would not otherwise have found. Not a headline. A brief, specific story with the task, the prompt structure they used, and the outcome. These posts do more for adoption than any amount of top-down communication, because they come from peers and they are concrete.

The third use case is Leadership Storylines. Viva Engage's Storylines feature allows leaders to publish short posts to everyone in the organisation — a LinkedIn-style feed, but internal. Ask your CEO, CHRO, or CITO to commit to a monthly Storyline post about their own experience with AI: what they tried that week, what worked, what they got wrong. Not a speech. Not a press release. A short, honest, personal note. Leadership vulnerability about AI failures is one of the most powerful signals an organisation can send to employees who are anxious about the technology.

Viva Engage also supports Ask Me Anything (AMA) events — live, structured Q&A sessions in the platform. A monthly AI AMA with a rotating subject matter expert, champion, or external guest is a cost-effective way to sustain momentum and surface questions that would otherwise stay unasked and therefore unanswered. The questions people ask in an AMA are your real-time read on where anxiety sits and where understanding has broken down.

Elisabeth Kübler-Ross's change curve — originally a model of grief, adapted for organisational change — remains one of the most practically useful frameworks for understanding where individuals sit in a significant transition. Applied to AI adoption, it maps almost exactly onto what we see in workforce data and on the ground in deployments.

Most of your employees are not in the "acceptance" and "integration" phase. They are somewhere in the early to middle sections of the curve: shock, denial, frustration, or depression. The CIPD data confirms that anxiety is the dominant emotional state. Your change programme needs to meet people where they are, not where you would like them to be. A relentlessly positive communications programme delivered to a workforce in the frustration stage produces cynicism, not engagement.

The manager's role in AI adoption is under-discussed but critical. Line managers are the primary sense-making interface between organisational decisions and individual employees. A manager who privately doubts the value of AI but publicly promotes it produces the worst outcome: visible compliance, invisible resistance, and a team that learns to perform adoption rather than practise it. Invest in manager briefings, manager Q&A sessions, and manager-specific enablement — not just end-user training.

Every successful AI adoption programme has a champions network. Most of them get it wrong in the same way: they select the people who are already enthusiastic about technology, which means they select the people whose peers view them as atypical. The result is a champions network that talks to itself in the Viva Engage community while the rest of the organisation watches with polite indifference.

The selection criterion for a champion is not enthusiasm for AI. It is credibility with their peers. The person other people in the team ask for advice when something is uncertain. The person whose opinion carries weight in team meetings. The person who is regarded as competent, pragmatic, and honest — not the person who is always excited about the latest tool. When that person says "I tried Copilot on this task and it saved me 40 minutes," it lands differently than when the office tech enthusiast says the same thing.

Training should cover three areas: enough AI literacy to be a credible guide (not an expert — a guide), specific use cases relevant to their business area mapped to the four zones, and facilitation skills for running peer learning sessions and handling the difficult questions about job security and change. The last area is the most important and the most neglected. Champions are often the first person an anxious colleague speaks to about what AI means for their job. Without preparation, that conversation either goes badly or produces false reassurance, both of which damage trust.

Recognition matters, but it needs to be the right kind. Public recognition in Viva Engage, inclusion in steering group briefings, and early access to new features are more valuable to most champions than a certificate. The thing that sustains champions is feeling like insiders — people who genuinely shape the programme rather than simply execute it. Build feedback loops from the champions network to the steering group. Act on what they report. Show them that their intelligence from the front line changes decisions.

Amy Edmondson's research on psychological safety — developed over 30 years and documented in "The Fearless Organization" (Wiley, 2018) — establishes a clear finding: teams with high psychological safety perform better under uncertainty, learn faster from failures, and are more willing to try new approaches. Conversely, teams with low psychological safety suppress information, avoid experiments, and perform compliance theatre when faced with mandated change.

AI adoption is a high-uncertainty environment. The technology is genuinely unpredictable. The frontier is genuinely jagged. Employees are genuinely asked to experiment with tools that may make things worse before they make things better. All of this requires psychological safety to navigate well. An organisation where people fear looking incompetent will produce the worst possible outcome: employees who pretend to be using AI effectively while actually avoiding the harder learning curve, and who cover their tracks rather than reporting failures that could inform the programme.

The second cultural condition is perceived fairness. Employees will watch carefully to see whether AI deployment is distributed equitably — whether some teams get better support than others, whether senior people are exempted from the parts of adoption they find inconvenient, whether the benefits flow upward and the disruption flows down. Perceived unfairness is deeply corrosive to adoption. The antidote is transparency: publish the deployment plan, the governance framework, and the principles that guide decisions about who gets what and when.

The third cultural condition is job security narrative — or more precisely, the quality of the narrative leadership provides on AI and jobs. Most organisations are silent or vague on this question, which allows employees' worst fears to fill the space. A specific, honest narrative — this is what AI is likely to change in your role over the next two years, this is what we are committing to do to support you through that change, and this is what we are not willing to do — is infinitely preferable to reassuring generalities. The WEF projects a net positive of 78 million jobs globally by 2030. That is a genuine basis for a hopeful narrative. But it needs to be made specific for your organisation's context, not just cited as a statistic.