Widespread GenAI use is not workflow transformation; measure depth
A large UK workforce survey reports broad GenAI exposure, while other research shows uneven workplace adoption. Leaders need task-level measures of repeated, governed use before declaring a process transformed.
What happened
Deloitte published findings from a survey of about 25,000 UK workers on GenAI use, skills and expectations; a separate workplace survey reported substantial variation by occupation, age and organisational context.
Why it matters
User prevalence cannot reveal whether AI changes a core workflow, improves an outcome or merely assists occasional drafting. Investment decisions need measures of depth, task coverage, quality and exception handling.
Deloitte's UK workforce survey reports responses from about 25,000 workers on generative AI use, skills and expectations. A separate summary of University of Konstanz research describes workplace AI adoption as uneven rather than universal. These sources offer useful prevalence and perception signals, but neither establishes that a particular workflow has been transformed or that GenAI caused a productivity gain.
That distinction matters because “use” can cover very different behaviours: trying a public chatbot once, drafting occasional text, repeatedly completing a bounded task, or redesigning an end-to-end process around AI with controls and measurable outcomes. Combining those states into one adoption percentage makes a broad trend visible while hiding the operational depth that leaders need for investment, workforce and risk decisions.
Define depth at the task level
A depth measure should begin with a named workflow and stable denominator. For recruitment, that might be all vacancy briefs or all candidate communications in a month; for service operations, all eligible cases; for software delivery, all changes in a defined repository class. The organisation can then measure the share of eligible tasks where AI is used, how often the result reaches production, where human review changes it and which exceptions fall back to the original process.
Repeated use is more informative than a one-time trial, but repetition alone is not value. Pair it with outcome measures appropriate to the workflow: cycle time, defect or rework rate, customer outcome, escalation rate, policy compliance and distribution across employee groups. A faster draft that creates more downstream correction may shift effort rather than remove it. A tool used mainly by already advantaged roles may widen capability differences even when overall adoption rises.
Separate exposure, adoption and transformation
Exposure means that a worker or task could encounter GenAI. Adoption means that the tool is actually used with some regularity. Transformation requires a material, sustained change in how work is organised, including roles, handoffs, controls and outcomes. These categories should not be inferred from one another. A survey can estimate exposure or self-reported use; workflow telemetry and outcome data are needed to test the deeper claims.
Leaders should also inspect the governance context. Is the tool approved? Are data boundaries understood? Is there an accountable reviewer? Can the organisation trace which version and prompt pattern contributed to an output? Are employees free to report failures without being treated as resistant? Governance affects measured depth because hidden use and unsafe workarounds make both adoption and risk estimates unreliable.
A practical dashboard therefore has a small hierarchy: eligible tasks, active users, repeated use, production acceptance, human modifications, exceptions and outcomes. Segment it by workflow, role and business unit. Do not convert a correlation between frequent users and better outcomes into a causal claim without a design that addresses selection effects. Pilot comparisons, staged rollout or other credible evaluation can improve the evidence.
The surveys justify investigation and targeted support, not a declaration that the enterprise has transformed. Qualitative interviews can explain why measured depth differs: access, confidence, manager permission, task suitability and fear of monitoring may all shape use. Those explanations should guide experiments, but they should not replace outcome measures. A team may report enthusiastic adoption while keeping the same handoffs and bottlenecks; another may use AI quietly in one critical stage and achieve a material change. The unit of analysis must remain the workflow rather than the publicity surrounding the tool.
The Skills Atlas can map the capabilities required for evaluation, verification and process redesign; leaders should connect those capabilities to named workflows and evidence of sustained outcomes.