Rolling KPI
One and 100 average 50.5. Nobody waited 50.5 minutes.
Ask a hospital how its porters did last month and the answer is one number: the average response time. Managers ask for it, and executives ask for it. Take two patients, one answered in 1 minute and one in 100. Their average is 50.5, a wait neither of them had.
An average also forgets what the hour was like. Here are two hours in one hospital, every dot a job. Both averaged 10 minutes. On a monthly report they are the same hour.
Tuesday 30 June 2026, 10:00 73 jobs, 13 people
Average 10 minutes
Saturday 8 August 2026, 11:00 58 jobs, 13 people
Average 10 minutes
0Minutes waited30 What the average hid
26% more work
Tuesday carried 73 jobs and Saturday 58, with 13 people each time. The average did not move.
A bigger share waited long in the quieter hour
Saturday had fewer jobs, yet 28% of them waited over 15 minutes (16 of 58 jobs), while Tuesday, with more jobs, had only 19% waiting that long.
A wait nobody had
On Saturday no job took 10 to 11 minutes. Jobs were either quicker or slower, and the average landed in the gap between them.
A fixed target is no better
So people set a target: answer every job within so many minutes. But 11 minutes at three in the morning, with two porters and two jobs waiting, is fine. 11 minutes at 10:00, with a full team and 40 jobs in the queue, is a service failing. One number cannot be fair to both.
So we handed the problem to our interns.
Piyorot and SatapornPiyorot set the problem: find a better way to measure service than the average. Sataporn prepared the raw data. 
The interns of Class 26Together they thought through, designed and built a prototype of the rolling KPI, along with measures of how well the service and the people in the team perform. 
Find the hour that was like this one
They asked a different question, and we helped shape it: what has this team done before, in the same situation? Take Saturday 8 August at 11:00: 58 jobs among 13 people. The search goes back through the past year to every Saturday at 11:00 and keeps the one with the closest job load, the jobs each person carried. If hours tie, the number of jobs decides, then the pressure, the minutes of work asked for against the minutes of people available, and last, how recent it was.
Jobs per person, every Saturday at 11:00
4/4 11/4 4.50 18/4 25/4 2/5 9/5 16/5 23/5 30/5 6/6 13/6 20/6 27/6 4/7 11/7 18/7 25/7 This hour: 58 jobs ÷ 13 people = 4.46
That hour is the benchmark, and every hour finds its own, which is why we call the KPI rolling. It takes a division, a search and a percentile: simple mathematically, and very practical. And it is new. We believe it is a better way to deeply understand how your operations run.
Judge the jobs, not the mean
The benchmark’s 60th percentile, or P60, the time 60% of its jobs were answered within, becomes the bar. Each job now either met it or did not, and the share that met it grades the hour. Here are the same two hours again.
Tuesday 30 June 2026, 10:00 Under Controlled
Bar 16:07
63 / 73 met it
Saturday 8 August 2026, 11:00 Critical
Bar 6:01
17 / 58 met it
0Minutes waited30 The same 10-minute average, and opposite truths. Saturday’s benchmark had answered within 6:01, and only 17 of 58 jobs did. That Tuesday was measured against a hard hour, 16:07, and 63 of 73 made it.
An hour that does exactly what its benchmark did lands near 60%, in Elevated. Under Controlled means beating your own past. Here is one week, hour by hour, and no single number could have told you this.
Mon Tue Wed Thu Fri Sat Sun 00:0006:0012:0018:00
Hospitals are places of habit. Every day of this week rose and fell in almost the same shape, so the shape tells you little. What matters is the colour: the hours that did worse than this team has done before, in the same situation. The coloured line is the Monday above. At 11:00, 58 jobs were Critical. At 14:00, 59 jobs were Stressed. Almost the same traffic, and a different hour.
Jobs in each hour, Monday 3 August 2026: 638 jobsAverage response time, Monday 3 August 2026: 16:33
Under Controlled 10Elevated 7Stressed 3Critical 4 The hour’s averageTarget 10 minutes Now press Average only. Monday becomes one tall spike, and the stressed afternoon disappears under a 10-minute target. No fixed target solves this: six minutes flags half the week, and 10 misses half the trouble.
Three questions for your own number
If you take one thing from this page, take these to your next report.
Who is hidden in it?
An average hides the person who waited longest. Ask for the shape, not the middle.
What is it compared with?
A number means nothing alone. Ask what it is being held against, and who chose that.
Was that a fair comparison?
Same hour of the week, same load? If not, you are comparing a quiet night with a full morning.
Where it stands
Every figure on this page comes from Mu Hospital, the invented hospital Class 26 build against: 23 porters, 50 places, and nobody real. Rolling KPI is what Mu, our assistant, explains. We could not be prouder of Krittinee and Preawnapat.
Meet Mu →
From knowing to planning
To know when your operations are likely to break gives you a better chance to prepare a better plan. That is why this experiment complements Roster.
See Roster →