What step timing data can tell you, and what it cannot

What the numbers are

Records and analytics gives you, for each step of each procedure, the median and the 90th percentile of how long it took, calculated from your own organisation’s recent runs of that step. Nothing is compared with an outside benchmark, because there is no outside benchmark for your bench, your equipment and your staffing.

The honest reading of the median is "this usually takes about this long here". The honest reading of the 90th percentile is "one run in ten takes at least this long", which is often the more interesting number: it is where the surprises live.

The question they answer well

Which step of this procedure loses the time. That is a question about the procedure, and it usually has a fixable answer: a step that needs something fetched from another room, a wait nobody knew was avoidable, an instruction ambiguous enough that people stop and think every time.

• A step whose median is high for everybody is a design problem, not a people problem.
• A step with a tight median and a distant 90th percentile has an occasional obstacle worth naming.
• A step that got slower after you republished the procedure is feedback on your edit.

The question they answer badly

Whether somebody is doing their job properly. A step can be entirely typical for the people doing it and still be too slow for the work; it can be fast because it was done well, or fast because something was skipped that the record cannot see. Timing data cannot tell those apart, and a number that cannot distinguish care from haste should not be used to judge either.

Earlier articles on this site went the other way. They described setting target times, benchmark bands, stretch thresholds, on-time completion rates against an SOP target, alerts when a step exceeded a threshold, and "top performers". None of that is in the product: there is no target time field, no threshold, no alert, and no leaderboard. It was invented for the article.

What is actually in the product

• Per-step median and 90th percentile from your own recent runs.
• A per-performer breakdown, which is not a ranking, and reads best as a question rather than a verdict.
• History you can filter, and CSV export of whatever the filters select.
• Weekly and monthly summaries, generated on a schedule and stored in the console. They are not e-mailed; earlier versions of this site said they were.

If you must use it in a conversation

Bring the step, not the person. "This step takes everybody four minutes and the procedure assumes one — what is in the way?" is a question that gets an answer. Its counterpart addressed to an individual gets an explanation instead, and after a while gets a faster button-press rather than faster work.