Why licence usage is the wrong headline
A rollout report that says “80% of licences active” answers the question the vendor cares about. It does not tell you whether any piece of work is faster, better or cheaper. People can open a tool every week and use it for nothing that matters, or use it for a task where checking the output takes longer than doing the work.
The worksheet separates three questions that usually get blended together:
- Are people using it? Licence utilisation (weekly active users ÷ licences) and reach (weekly active users ÷ people in scope).
- Is it used for this work? Workflow adoption: the share of tasks of this type done with AI assistance.
- Did the work change? Time per task, rework and quality, before and after, on the same definitions.
How the figures are calculated
- Gross hours saved per week = AI-assisted tasks per week × (baseline minutes − minutes with AI) ÷ 60. “Minutes with AI” must include checking and editing.
- Rework-adjusted hours saved = AI-assisted tasks × (baseline minutes × (1 + rework rate before) − minutes with AI × (1 + rework rate after)) ÷ 60. This assumes a reworked task costs roughly its own time again.
- Quality change = quality check pass rate with AI − pass rate before, in percentage points.
- Capacity is shown as working days per week at 7.5 hours a day, with the reminder that capacity is not cash.
Blank fields are treated as missing, not zero, so a missing baseline produces a flag rather than a misleading figure.
Capacity is not cash
Hours saved become a financial result only when something changes: more work handled by the same team, less overtime or contractor spend, or a role not backfilled. Until then they are capacity, and the honest report says what that capacity was used for. This distinction is what lets a finance sponsor trust the rest of the numbers.
Illustrative example
Illustrative example using the worksheet’s example figures, not client data. A support team of 24 people has 24 licences and 15 weekly active users. Of 120 complaint replies a week, 70 are drafted with AI. Baseline time was 40 minutes per reply; with AI, including review, it is 28. Rework rose from 8% to 12%, and the quality check pass rate fell from 90% to 88%.
The worksheet reports licence utilisation and reach of 63%, workflow adoption of 58%, 14.0 gross hours saved a week and 13.8 after rework, which is about 1.8 working days of capacity. It raises one flag: time saved but quality fell. The right next step is not to celebrate the hours. It is to ask the reviewer whether a two-point fall in pass rate is acceptable for complaint replies, and what change to the prompt, template or review step would recover it.
Assumptions and limitations
- Every figure is yours. The worksheet does arithmetic and pattern checks; it cannot tell whether the figures are accurate or whether the before and after groups are comparable.
- It measures one workflow at a time. Adding unrelated tasks together hides the patterns the flags look for.
- The rework adjustment is a simplification. Where rework is a quick correction rather than a full redo, the adjusted figure understates the saving.
- It says nothing about what other organisations achieve. There are no benchmarks in it, deliberately.
What to do next
Use these figures as the evidence for the pilot scorecard when deciding whether to expand, revise or stop. If you have not yet chosen a workflow or recorded a baseline, start with the readiness self-assessment.
If the flags show a rollout where usage is high and the work has not changed, that is the situation AI adoption rescue is for. If you want help setting up measurement as part of a bounded pilot, with a named internal owner at the end, see AI enablement.