On-Time Arrival Is a Grace Window and a Check-In Habit

Why this matters

On-time arrival is the number a shop quotes to a commercial customer during a bid and prints on a sign in the office, and it is the easiest operations number in a shop to move without improving a single arrival. Two parameters sit behind it, neither of them printed beside it, and either one can carry the figure across twenty points or more. A shop that reports 86 percent and means it will find, the first time somebody audits the claim, that it was reporting on two thirds of its scheduled work against a window it chose itself.

Both of those uses change what the number is: a performance figure stated in a bid can be written into the contract as a service level with credits or termination rights behind it, and a performance figure advertised to the public is an objective claim you must be able to substantiate at the moment you make it under Section 5 of the FTC Act, 15 USC 45. So before this figure leaves the building, print the grace window and the arrival coverage beside it, keep the working that produced it, and have your own attorney read any bid language that turns it into a promise.

The two assumptions on the face of the number

Arrivals landing within a short grace window of the scheduled start, over jobs that have both a scheduled start and a recorded arrival.

The first assumption is a promise. The grace window is a policy choice about what counts as on time, and nobody outside the shop agreed to it. It does not appear in the figure, it is not usually written down, and changing it changes the number without changing anything that happened.

The second assumption is a record. The denominator holds only jobs where somebody checked in. A technician who never records an arrival is not late in this measurement. They are absent from it. That makes the figure a rate over an opt-in population, which is the same structural defect that governs recorded deadlines; the reference on SLA compliance owns that claim in general. What is specific here, and worse, is that the missing rows go missing through a behaviour of the person being measured, so they are not merely incomplete, they are selected.

One quarter, and what the two parameters do to it

A shop completes 640 jobs in a quarter. Of those, 612 carried a scheduled start. Of those 612, 431 also carried a recorded arrival.

So the reporting denominator is 431 jobs, which is 70.4 percent of the 612 scheduled jobs and 67.3 percent of the 640 completed ones. Of those 431 arrivals, 371 landed no more than 15 minutes after the scheduled start, with every early arrival counted as on time, which is how nearly every shop computes it. Reported on-time arrival: 371 / 431 = 86.1 percent.

Parameter one, with nothing else changed. Take the identical 431 arrivals and move only the definition of on time:

Not one arrival changed. The reported figure moved 32.5 points. If a shop reports this number without printing the window beside it, the number is not a measurement of anything, and two shops comparing figures are comparing two definitions.

Parameter two, the missing rows. 612 minus 431 leaves 181 scheduled jobs with no recorded arrival. Those are not missing at random. Of the crew of five, one technician accounts for 96 of the 181, which is 53.0 percent of the missing records on one of five people. On the 48 arrivals that same technician did record, 33 landed no more than 15 minutes late, a 68.8 percent on-time rate against the shop's reported 86.1 percent over the whole recorded set.

So the person least likely to record an arrival is also the person least likely to be on time on the arrivals they do record. That is the shape this defect always takes, because the circumstances that cause a missed check-in, running late, arriving flustered, starting work before touching the record, are the same circumstances that cause a late arrival.

Bounding it, and the identity that makes bounding cheap

Put the 181 unrecorded jobs back into the denominator and take both extremes:

• If every one of the 181 arrived on time: (371 + 181) / 612 = 552 / 612 = 90.2 percent.
• If every one arrived late: 371 / 612 = 60.6 percent.

The true on-time share of all 612 scheduled jobs sits between 60.6 and 90.2 percent, a band 29.6 points wide, and the reported 86.1 percent sits near the top of it.

That band width is not a coincidence and you never have to compute it twice. The denominator is fixed at 612 and each missing row can move the numerator by exactly one, so the band width in percentage points equals the unmeasured share of scheduled jobs: 181 / 612 is 29.6 percent, and the band is 29.6 points. Coverage alone tells you how much the figure could be wrong, before you look at a single arrival.

A defensible middle estimate: if the 181 unrecorded jobs ran at the 68.8 percent rate of the technician who accounts for most of them, that is 124.5 on-time arrivals, so 371 + 124.5 = 495.5 over 612, or 81.0 percent.

Now read those two figures carefully, because this is the sentence the whole article exists for. The 86.1 percent is over 431 recorded arrivals. The 81.0 percent is over 612 scheduled jobs. They are not two readings of one quantity. They are one quantity read over two populations, and only the second one describes the promise the shop actually made to customers.

The companion figure, and the floor below which the number is not reportable

Quote arrival coverage in the same sentence as on-time arrival, every time: recorded arrivals over scheduled jobs, here 431 / 612 = 70.4 percent.

Do not report an on-time figure below about 90 percent arrival coverage. The reasoning is the identity above. At 90 percent coverage the unmeasured share is 10 percent, so the widest the truth can sit from the reported figure is 10 points, which is comparable to a full period of genuine movement and is a tolerable uncertainty to carry. At this shop's 70.4 percent coverage the band is 29.6 points wide, which swamps any change a shop would ever act on. Tune the floor if you like, but tune it against that arithmetic rather than against a preference.

This shop is at 70.4 percent, well under the floor, so the honest quarterly report is the coverage figure plus a plain statement that the on-time figure is not yet readable. That is not a failure to report. It is the only accurate thing available, and it points at the fix, which is a check-in habit rather than a dispatch change.

Two things raise coverage fast and neither is a policy memo. Make the check-in the action that opens the job record rather than a separate step performed after it, so skipping it costs the technician the thing they need. And report coverage by name every period, because a rate that is 53 percent one person's behaviour is fixed one conversation at a time.

What the window is measured from, and the arrival nobody counts

A threshold's reference point is part of the threshold, and this one has trouble at both ends of its.

The datum. The scheduled start in your records is usually the dispatcher's plan for the day, not the time a customer was given. Where those differ, and they differ whenever a window gets narrowed in the office after the call, the figure measures adherence to an internal plan. It will look fine while customers are unhappy, and no amount of grace-window tuning will reconcile the two. Measure against the time the customer was told, and if your records do not hold that time separately from the planned start, adding that one field is worth more than anything else in this article.

The early end. The rule above is one-sided: anything at or before the scheduled start counts as on time, however far before. Of the 431 recorded arrivals, 27 landed more than 15 minutes early, and all 27 scored as on time. A two-sided rule, meaning inside the window at both ends, gives 371 minus 27 = 344 over 431, or 79.8 percent.

The 6.3 points that come off are the same 6.3 percent those 27 arrivals are of the 431, not a coincidence: the denominator does not change, so every arrival reclassified is one point of the same base. The reason to reclassify them is not scoring, it is that an early arrival on a residential call routinely finds nobody home, which produces a second trip, a rescheduled job, or a technician sitting in a truck. If your work is commercial and someone is always on site, the one-sided rule is defensible and you should say that is why you chose it.

What the promise was worth

The measurement is only ever worth what the promise was, and a grace window belongs to a named time, not to a window.

A named arrival time. The customer was told an hour. A 15-minute grace is the common starting point, it is short enough to mean something and long enough to survive traffic, and it should be stated to the customer rather than kept internal.

An arrival window. The customer was told a two-hour or four-hour span. There is no grace, because the window already contains it. The measurement is inside or outside, full stop, and a shop that adds 15 minutes of grace to a four-hour window has quietly promised four hours and fifteen minutes.

That distinction matters because of how shops improve. Narrowing a promised window is the single most valuable thing a service business can do for a residential customer, and it makes the number harder at the same time. A shop that narrows the window and leaves the grace alone has made the promise better and the score worse, which is exactly the right direction and exactly what will get the change reversed if nobody says so beforehand. Write down which promise each figure is measured against, and never compare a quarter under a four-hour window with a quarter under a two-hour one.

And the rule that does more for customers than either parameter: notify the customer when the arrival will slip past the promised time by more than 15 minutes, and notify before the promised time rather than after it - the same 15-minute slip threshold The Tenant Notification SOP sets, and distinct from How to Set Response Standards a Manager Can Hold You To, whose separate rule is that you notify within 15 minutes of KNOWING. One is the size of the slip, the other is your latency, and a shop needs both. A late arrival that was flagged ahead is a different event to a customer than a late arrival that was not, and neither the on-time figure nor the grace window can see the difference. Track notified slips as their own count beside the two figures above, because it is the one of the three that a technician can improve on a day that has already gone wrong.

The whole quarter, in one table

The same 431 recorded arrivals and the same 181 missing rows, on the one-sided rule, read every way the two parameters allow:

One quarter, one set of arrivals, none of which moved, reported anywhere from 43.8 to 96.2 percent depending on two choices nobody wrote down. That is a 52.4-point spread, and every cell in it is arithmetically correct.

The number is worth having. It is worth having with its window and its coverage printed in the same sentence, and it is worth nothing at all without them.

References

• See related: SLA Compliance Only Counts Jobs That Carry a Deadline - the opt-in denominator, derived in full
• See related: The Tenant Notification SOP - the 15-minute slip threshold
• See related: How to Set Response Standards a Manager Can Hold You To - the notify-within-15-minutes-of-knowing latency rule
• See related: Average Travel Time and the Route Density It Implies
• See related: Technician Utilization and What the Denominator Assumes
• See related: The Satisfaction Score and Who Actually Answers a Survey - the same selection problem in a response set