AI isn’t the hard part. Adoption is.

Most organisations have moved past the first question: “Can we access AI?” The bigger question now is what happens after you switch it on.  

Deploying an AI tool is not the same as making it valuable. Once it lands in the business, people immediately start asking the real questions:

• How do I use this in my day-to-day work? 

• Can I trust the output? 

• Am I responsible if it’s wrong? 

• Is this optional, or is it expected? 

• What does “good use” look like here?  

If those questions aren’t answered, adoption doesn’t fail loudly. It fails quietly, through hesitation, low confidence, and people reverting to old habits. 

The result is familiar: licences are assigned, policies are published, training happens… and leaders still ask why nothing has really changed. 

Deployment makes AI available. Adoption makes it useful. 

A lot of organisations call deployment “adoption”. But they’re different things.  

Deployment looks like: tools enabled, licences assigned, platforms configured, basic policies and training rolled out.  

Adoption looks like: people knowing when to use AI, trusting it appropriately, integrating it into real workflows, having managers reinforce good practice, and measuring whether it’s improving work (not just generating activity).  

That’s the gap many organisations are sitting in today: AI is “there”, but it isn’t embedded.  

Why AI adoption is different to other tech change 

It’s tempting to treat AI like any other software rollout. Use standard change management, do a comms plan, deliver training, and move on. 

Some of that still matters. Leadership sponsorship, governance, capability building, workflow redesign, measurement, reinforcement, communications are all part of adoption. 

But AI behaves differently to a typical system rollout. There are six reasons which make adoption uniquely hard and uniquely human. 

1) AI is probabilistic 

AI doesn’t always give the same answer twice. People have to evaluate outputs, apply judgement, and learn how to ask better questions. That’s a different skill to “click here, then here”. 

2) It’s general purpose 

AI touches almost every part of the business: writing, summarising, analysing, coding, customer interaction. It doesn’t sit neatly inside one department or one process.  

3) It’s role-sensitive 

The same tool can feel helpful to one person and threatening to another. For someone, it saves time drafting emails; for someone else, it feels like it overlaps with their core value.  

4) It sits close to judgement 

AI isn’t only automating tasks, it can appear to replicate reasoning. That creates friction because people start to wonder what “good judgement” means when the tool can produce convincing answers quickly. 

5) It’s psychologically disruptive 

AI can change how people see themselves at work, their identity, their confidence, and their sense of contribution. 

6) It’s trust-dependent 

People need to know when to trust it, when not to trust it, when to disclose its use (especially with customers), and when AI simply isn’t the right approach. 

These are not “features and training” problems. They’re operating model problems, and leadership problems. 

The psychological contract 

A psychological contract is an unwritten, informal, and evolving set of mutual expectations and obligations between an employer and employee, which is different from the legal contract.  

AI adoption isn’t just behavioural change. It’s a shift in the relationship between employee and organisation.  

Every employment relationship includes an unspoken deal:
“I bring effort, judgement, and expertise, you give me security, purpose, and a sense that I’m valued.” 

When AI enters the workplace, it can quietly disrupt that deal. People start asking themselves: 

• “If a machine can draft what I draft, where do I fit?” 

• “If I use it, am I responsible for the output?” 

• “Will using AI make me look more capable, or less capable?” 

• “Am I relying on it too much… or too little?” 

• “Is it expected that I use this, or risky?” 

And crucially: people don’t always voice these concerns. They often go quiet. They disengage. They hesitate. 

So, if you’re a leader wondering why AI uptake is slow, don’t assume it’s stubbornness or lack of skill. It might be uncertainty about safety, accountability, and role impact, all things leaders can address. 

Adoption by design: treat it like an operating system 

Successful adoption doesn’t happen because people are told to “use AI”. It happens when the organisation builds the right operating environment around it.  

In practice, this environment contains six parts: 

1) Leadership clarity 

People need to understand why AI matters to this organisation, and how it connects to their role, not just headlines about AI in the news.  

2) Governance that enables (not blocks) 

The goal isn’t to stop experimentation, it’s to create clear boundaries: what’s approved, what’s sensitive, what needs review, and where accountability sits. 

3) Role-based capability building 

Different roles need different AI skills. A finance user, a service desk analyst, a project manager, and a software developer shouldn’t all be trained the same way. 

4) Workflow integration 

Many organisations use AI “in isolation”, as a clever side activity. Value comes when it’s integrated into the workflows people rely on every day.  

5) An innovation pathway 

Once people understand AI, they’ll have ideas. The question is: what happens next? You need a way to capture, test, evaluate, scale, or stop, ideas safely.  

6) Measurement and reinforcement 

If you can’t explain what AI is changing, you can’t prove value. Leaders need a practical way to measure adoption, quality, workflow change, and business impact. 

Measuring value: stop at “usage” and you’ll miss the point 

A common trap is measuring activity instead of outcomes.  

Using a simple “value ladder”: 

1. Use: are people using AI at all? (Many organisations stop here.)  
2. Quality: are they using it well and safely, in ways that fit their role?  
3. Workflow change: is it actually changing how work gets done? 
4. Improved speed, quality, or capability: are outcomes measurably better, faster cycle times, better quality, more capacity for higher-value work?  

You’ll sometimes hear a stat like “a third of the team is using it”, but usage alone doesn’t tell you whether it’s making work better.  

What “good” looks like when adoption is working 

When adoption is designed properly, you start to see a different workforce: 

• People who understand where AI helps, and where it doesn’t 

• People confident enough to challenge outputs, not just accept them 

• Teams forming “clusters” of improved productivity and quality as AI becomes embedded in the work 

• Ideas being tested and governed, not spreading through guesswork 

• Value becoming measurable, not just anecdotal  

This is also why the loudest “AI champions” aren’t automatically the best role models. The real champions demonstrate good judgement, responsible use, real value creation in their work, and the ability to bring others with them.  

A practical starting point for leaders: five readiness questions 

If you want a quick way to sanity-check whether your organisation is set up for safe, valuable AI adoption, start here: 

1. Do people know when they’re allowed to use AI, and when they’re not?  
2. Do managers know what “good use” looks like, and how to reinforce it?  
3. Are you building role-specific capability (not generic training)?  
4. Have you integrated AI into real workflows, not side experiments?  
5. Are you measuring outcomes (quality, speed, capability), not just usage?  

If any of those answers are unclear, the barrier usually isn’t technology, it’s operating conditions. 

Closing thought: strategy sets direction. Adoption creates movement. 

AI strategy matters. But strategy alone doesn’t change how work happens. Adoption is what turns ambition into day-to-day practice, and measurement is how you know it’s working.  

And one final point from the talk is worth making explicit: AI tools can move very quickly. But good adoption takes time. People need time to learn, space to practise, and clarity on what’s expected.  

That’s the difference between “we rolled out AI” and “we’re getting value from AI”.