Measure what changed, then scale or stop

Track a path from eligible employee to first useful attempt, repeat use on the intended job, accepted result, and improved business outcome. A seat is only the entrance. A weekly active-user number says more, but it can still reward busy clicking. A person who produces one reliable renewal brief a week may matter more than someone who generates 50 drafts nobody uses.

Count how many people in each role had a legitimate reason and approved access to try. Then count who actually used it for the named task. If the rate is low, ask why before launching another training campaign. The data may be missing, the task may be rare, the approved tool may be awkward, or the employee may correctly judge that it does not help.

Use quality and cost beside activity. For each workflow, track accepted outputs, severe errors, correction time, repeat contacts, and the person-hours needed to teach and support it. A six-minute draft with 15 minutes of checking is not a six-minute process. A time saving that lets employees serve more customers is released capacity; a cash saving needs a corresponding budget change.

By day 30, finish one or two narrow pilots. Have a named owner, baseline, reviewer log, and a documented stop rule. If you cannot explain what changed in the work, do not call the pilot complete.

By day 60, run a second cohort only for workflows that passed. Give new participants the tested guide, peer clinic, source list, and escalation route. Compare their results with the first cohort. If performance depends on one unusually skilled champion, redesign the training or keep the workflow limited.

By day 90, review quality and economics with finance and the operating owner. Decide separately for each workflow: expand with the current controls, invest in fixing data or integration, retain as a limited assist, or stop. A “rollout percentage” across all jobs hides these different decisions.

Show one clear win, one corrected failure, and one use you rejected. That is more credible than a page of only success stories. Tell leaders which company-reported examples inspired the approach and which numbers came from your own measured queue.

Keep an owner for the next 90 days. The tool, source data, and staff will change. Recheck accepted-output rate after those changes, not only at launch. If quality falls, use narrows, or the cost rises, the team needs permission to pause and redesign rather than defend the original purchase.