branch hub
AI adoption at work: from access to sustainable practice
A practical model for separating access, activity, workflow change, capability and durable organizational adoption.
Sustainable AI adoption means more than giving employees access to a tool. It occurs when people can identify suitable work, use AI within clear boundaries, verify important outputs and integrate the practice into repeatable workflows. Measurement should therefore separate access, activity, applied capability, work quality and organizational support instead of collapsing them into one adoption rate.
Separate access, activity and applied use
Access answers who can use an AI tool. Activity answers whether and how often they used it. Applied use asks whether AI became part of a real workflow and whether the resulting work met its objective. Each stage needs different evidence and a different intervention.
A login or prompt count can reveal reach, but it cannot show whether the person selected an appropriate task, supplied relevant context or checked a material claim. Skills England's benchmark describes workplace AI skills as a combination of technical, non-technical, responsible and ethical practice, which is broader than access alone.
Programme teams should define the stage they are trying to improve. Low access may be a licensing problem. Low activity may reflect poor workflow fit or trust. Frequent use with weak verification is a capability and governance problem. Treating all three as engagement produces the wrong remedy.
Sources: [1] [2]
Organize adoption around workflows, not tool features
Adoption becomes durable when teams can name the work to be improved, the evidence the work requires, the role AI may play and the checks that remain human responsibilities. Workflow-level design gives training, policy and measurement a shared unit of analysis.
Start with recurring tasks such as synthesizing research, preparing a forecast narrative or drafting a customer response. For each task, identify sensitive inputs, decision-changing claims, required approvals and the final audience. This makes appropriate use concrete without assuming one tool or prompt pattern fits every role.
The OECD's work on changing skill demand emphasizes that AI adoption changes how work is organized, not just which technical skills are needed. Managers should expect role design, coordination and review practices to change alongside individual tool use.
Sources: [2] [1]
Measure a balanced adoption system
A balanced adoption view combines reach, repeat use, workflow coverage, demonstrated capability, output quality, incidents and user feedback. No single metric establishes success. The measures should be interpreted together and tied to the programme's stated objective.
A team may show high activity but low confidence in sensitive workflows. Another may use AI infrequently because it has automated one high-value task well. Both cases disappear inside a single monthly-active-user percentage. Add qualitative evidence about friction, usefulness and review burden.
Use staged reviews: first confirm access and awareness, then examine suitable workflow use, then sample work quality and verification, and finally assess whether the practice survives changes in tools or champions. This supports course correction without presenting adoption as a universal score.
- Track access and activity separately.
- Map use to named workflows.
- Sample the quality and verification of resulting work.
- Record organizational enablers and barriers.
- Review whether the practice persists over time.
Sources: [1] [2]