Key Takeaways
- Only 13% of CIOs say their enterprise has realised significant or transformational value from AI so far, according to Gartner's October 2026 survey.
- McKinsey's research finds change management and siloed ways of working block AI scale more than technology infrastructure does.
- Successful programmes follow a 1:3:5 ratio: for every dollar on agentic technology, three on process redesign and five on capability building and adoption.
- Use five stages, Awareness, Belief, Commit, Develop and Enforce, and design each one around the specific fears of a specific role.
- Report cycle time, quality and cost per task to the board, not log-ins or licence counts.
A regional insurer's claims team was given AI agents six months ago. The licences are paid, the dashboards are green, and the IT team reports thousands of prompts a week. Yet the claims backlog has not moved, and the COO is now asking a simple question: what did we actually buy?
This is the AI adoption gap. The technology is deployed, but the work has not changed. Closing it is a leadership problem, not a procurement one, and it needs a framework of its own.
Why do enterprise AI tools go unused after rollout?
AI tools go unused because rollouts are run as technology projects while the real barrier is human. Employees fear looking incompetent, being blamed for an agent's error, heading into an undefined future and losing their professional edge. Without a plan for those fears, teams imitate adoption without changing how they work.
The data points in the same direction. Gartner's October 2026 CIO survey found that only 13% of CIOs say their enterprise has realised significant or transformational value from AI. In McKinsey's State of Organizations research, thousands of business leaders ranked change management and siloed ways of working above technology infrastructure as obstacles to scaling AI.
McKinsey's August 2026 article How to close the agentic adoption gap names four fears that sit beneath most stalled rollouts:
- Fear of being found out: people do not know where to start and will not admit it, so learning happens in private or not at all.
- Fear of accountability without control: if an agent is wrong and I act on it, the mistake is still mine.
- Fear of the unknown: tools change every few weeks and nobody can describe the end-state operating model.
- Fear of losing professional edge: if the agent does the analysis I spent years mastering, what is my contribution now?
These fears are rational, which is why pushing harder rarely works. McKinsey also notes that middle managers are often the most anxious group, because they must translate the vision into daily work while running every other change programme at the same time.
Governance pressure adds to the gap. Gartner reported on 9 November 2026 that 71% of organisations expect to deploy agentic AI within 12 months, yet only 21% have agentic AI governance capabilities in place. When employees cannot see the rules, the fear of accountability gets worse.
How much of an AI budget should go to change management?
Far more than most enterprises assume. McKinsey describes a 1:3:5 pattern in successful AI transformations: for every dollar spent on agentic technology, organisations spend three on process redesign and five on capability building and adoption. Most companies invert this, funding the platform generously and treating people work as an implementation detail.
The ratio is a planning heuristic, not a law, but it changes the conversation with your CFO. If a business case only prices licences, integration and tokens, it is pricing roughly one ninth of what success requires.
Under-funding adoption is also expensive in its own right. McKinsey points out that ineffective adoption can generate significant token, compute, governance and oversight costs without matching gains in performance. In other words, a half-used AI programme is not a cheaper programme; it is a costlier one per unit of value.
What the people budget actually buys
- Protected time for teams to redesign their own workflows
- Embedded coaches and peer champions, not one-off classroom sessions
- Clear human review rules so staff know when they own the decision
- Changes to KPIs, incentives and promotion criteria that reward the new way of working
What are the five stages of an AI change leadership framework?
The five stages are Awareness, Belief, Commit, Develop and Enforce. People need direct experience before they believe, belief before they commit, commitment before they build skills, and skills before new behaviour is locked in by metrics and incentives. Skipping a stage is the most common reason adoption stalls or slides back.
The framework below adapts McKinsey's A to E playbook into a sequence a Hong Kong leadership team can run inside one business unit.
| Stage | Goal | What leaders do | Signal it is working |
|---|---|---|---|
| 1. Awareness | Make the change tangible | Hands-on sessions on real business problems; say plainly what will change and what humans keep | Staff can describe one concrete use in their own role |
| 2. Belief | Prove it is safe and worthwhile | Senior leaders use AI visibly in real reviews; credible peer champions demonstrate it | Voluntary trial use rises without reminders |
| 3. Commit | Let people choose to engage | Protected time to rethink each role; acknowledge what is being lost | Teams volunteer workflows for redesign |
| 4. Develop | Build skill in the flow of work | Embedded coaching, review practices, visible skill levels | Output quality holds while cycle time falls |
| 5. Enforce | Lock in the new operating system | Update KPIs, decision rights, incentives and promotion criteria | Process metrics improve quarter on quarter |
Two details make the difference. First, design each stage per role, not for "the workforce". A claims handler, a finance controller and a relationship manager fear different things and make different decisions. Map the specific moment where an agent's input arrives and a person must decide whether to act on it.
Second, expect backsliding. A poor agent decision or a policy reversal can push a team from Develop back to Belief. Treat a loss of trust as a signal to return to the right stage, not as resistance to override.
How does this framework apply in a Hong Kong enterprise?
Start with one function and one workflow where the decision moment is clear, such as claims triage, shipment exception handling or first-draft client reports. Run all five stages inside that unit, measure process outcomes, then use the evidence and the trained champions to carry the model into the next function.
Financial services: claims or KYC review
Accountability fear dominates in regulated work. Write the human review rule before rollout: which agent outputs a reviewer must check, which can pass on sampling, and who signs off. The PCPD's AI guidance already expects human oversight in proportion to risk, so this step serves compliance as well as adoption.
Logistics: shipment exceptions
Here the fear of the unknown is strongest among supervisors who built their reputation on firefighting. Script the new week: the agent gathers tracking data and flags exceptions, the supervisor decides on reroutes and customer calls. When people can see their future role, they stop hedging.
Professional services: drafting and research
Loss of professional edge is the deepest fear among senior associates. Leaders who show their own agent-assisted work in partner meetings, and say openly when they overrode the AI and why, do more for belief than any training calendar.
Peer networks scale the effect. McKinsey cites a global bank that mobilised more than 2,000 AI champions who ran floor walks, office hours and short demonstrations. The lesson for a 300-person Hong Kong firm is proportional: a handful of credible users in each team, chosen for what they have actually done with AI rather than for their title.
What mistakes stall AI adoption in large organisations?
The most common mistakes are treating rollout as an IT project, giving executives days of immersion while frontline staff get a ten-minute briefing, relying on one-time training, measuring log-ins instead of outcomes, and leaving incentives unchanged. Each one lets the old way of working survive alongside the new tools.
- IT-owned rollout: McKinsey describes an enterprise software provider whose adoption stalled until it stopped treating agentic AI as engineering tooling and put leaders through immersive bootcamps.
- Asymmetric immersion: the top team spends days exploring the technology, while employees are told to start using it next week. That signals the change belongs to leadership.
- One-time training: tools that change every few weeks need continuous, embedded coaching. McKinsey cites a Fortune 500 technology company that placed one AI coach for every five to ten engineers.
- Activity metrics: prompt counts and licence utilisation reward imitation, not better work.
- Silence on jobs: avoiding the head-count question raises anxiety. Transparency, even with incomplete answers, builds more credibility.
Trust is the thread through all five. McKinsey's survey found that employees who report low trust in their organisation's support during an AI transformation are 1.5 times more likely to feel anxious about AI-related changes than those reporting high trust.
How should you report AI adoption to the board or CFO?
Report process outcomes, not usage. Show cycle time, error or rework rate, cost per task, human review burden and decision speed for each redesigned workflow, alongside where each team sits in the five stages. Pair this with the governance status of each agent so value and risk appear on the same page.
Gartner's November 2026 guidance makes the same point from the risk side: AI investments should be judged against the risks they introduce, and controls such as audit trails and human review can be "good drag" that lets valuable work scale with confidence.
A one-page board view
- Workflows redesigned this quarter, with before and after cycle time
- Quality: error, rework or escalation rate per workflow
- Economics: cost per task including tokens and human review time
- Adoption stage by team, with the main barrier named
- Governance: agents in production, named owner, review rule, incidents
This format answers the CFO's real question, whether the combined output of people plus AI is rising, and it gives the board a credible reason to fund the next function.
Lead the learning, not just the rollout
The AI adoption gap will not close with another licence or a better model. It closes when leaders fund the people side properly, design for the fears of each role, move teams through Awareness, Belief, Commit, Develop and Enforce, and report outcomes instead of activity.
The enterprises pulling ahead are not the ones with the most tools. They are the ones learning to redesign work faster than their competitors, one workflow at a time, with technology that serves the people using it.
Empowering AI. Centered on you.
Start with one workflow and prove it
Pick the function where adoption matters most and we'll walk you through every step, from role and workflow design to deployment, human review rules and the outcome metrics your board will ask for. AI Staff comes with $0 setup and a UD team that configures it with you, with data kept in Hong Kong.
Book a Free AI Staff Consultation
Frequently Asked Questions
How long does it take to see real AI adoption in one department?
There is no fixed timeline, but one workflow in one team is the right unit. Expect measurable process change only after the Develop stage, once people are using AI in daily work with coaching, rather than in the first weeks after licences arrive.
Is training enough to drive AI adoption?
No. McKinsey argues that one-time training creates awareness but rarely builds habits for tools that change every few weeks. Embedded coaching, peer champions and updated KPIs are what turn trial use into changed work.
Who should own AI adoption: IT, HR or the business?
The business unit should own outcomes, with IT owning the platform and HR owning skills and role design. McKinsey assigns distinct roles to the CEO, CFO and CHRO, and Gartner recommends a central governance function that coordinates value and risk across teams.
Should we tell employees whether AI will reduce head count?
Yes, as far as you honestly can. McKinsey finds that transparency about role changes, even with incomplete answers, builds more credibility than messaging that avoids the question, and low trust makes anxiety about AI 1.5 times more likely.