A punctuality figure is not a measurement of speed. It is the distance between an aeroplane and a number the operator wrote itself, timed at an event the operator also chose, judged against a tolerance somebody else set, over a sample from which the worst flights have already been removed. Every one of those four elements can be moved without touching a single departure. Which is why two carriers on the same city pair can publish the same headline figure and run twenty minutes apart, door to door.
The yardstick is written by the party being measured
Almost every operational metric in aviation is referenced to a physical fact: a runway length, a mass, a temperature, a fuel quantity. On-time performance is not. It is referenced to the published schedule, and the published schedule is a commercial document produced by the same company whose performance is being reported. Nothing prevents that document from being generous, and quite a lot encourages it.
This is not an accusation of dishonesty. Scheduled block time has to contain the real variability of a route, and that variability is large: taxi-out at a congested hub, departure sequencing, air traffic flow management regulations, wind, arrival holding, stand availability. A schedule that reflected the best day would fail on most days. The question a professional reader has to ask is not whether there is buffer, but how much, and whether the figure in front of them survives the fact that the buffer is unequal between operators.
Airlines set their own schedules, which include extra time to handle delays. This means punctuality figures aren’t just about how fast planes fly—they’re about how much extra time airlines build in. Different airlines might add different amounts of buffer, making their punctuality numbers hard to compare.
Two airlines flying the same route from Paris to London might both claim 90% on-time performance. But if one schedules 1 hour 45 minutes and the other 2 hours 15 minutes, the second airline has built in 30 minutes of extra buffer. Both might arrive on time, but the first airline’s passengers experience a tighter connection.
What “on time” actually names
The word covers at least four distinct measurements, and publications rarely say which one they are using.
The first choice is the event. Departure can be timed at the moment the door closes, at brake release for pushback, or at the OOOI Out event recorded automatically, typically on doors closed and parking brake released. These are not the same instant, and on a bad day they can be several minutes apart. Arrival is normally the In event — on-blocks — which means a delay caused by an occupied stand belongs to the airline’s number even though the aeroplane landed on time.
The second choice is the tolerance. The fifteen-minute convention is the most widely published: the United States Department of Transportation counts an arrival as on time if the aircraft reaches the gate less than fifteen minutes after the scheduled time, and European delay reporting has long used the same threshold alongside average delay per flight. Operators also track D0 and A0 internally, which is a completely different discipline, and some publish D15 departure figures where the reader assumes arrival figures.
The third choice is scope: which flights are in the sample. Cancellations are the important one. A cancelled flight has no arrival time, so it usually leaves the punctuality calculation entirely — meaning that cancelling the flight most likely to be late mechanically improves the published figure while making the passenger outcome worse. Completion factor is the metric that catches this, and it is the one that is quoted least often.
| Lever | Typical variants | Effect on the headline number |
|---|---|---|
| Reference time | Scheduled off-blocks, scheduled on-blocks, scheduled door close | Sets the entire yardstick; longer schedule, easier target |
| Captured event | Door close, brake release, OOOI Out / In |
Minutes, not seconds; stand congestion lands on the airline |
| Tolerance | D0/A0, D5, D15/A15 | Threshold metrics ignore how late the late flights were |
| Sample | Cancellations, diversions, franchise and feeder flights, season | Excluding the worst flights raises the average without improving anything |
Block time is not flight time
Scheduled block time runs from off-blocks to on-blocks. It contains taxi-out, any departure queue, the airborne segment, the approach, and taxi-in. On a short European sector, the ground portion can approach a third of the total. That matters because the ground portion is where the variability lives and where the buffer is cheapest to hide: an extra ten minutes of taxi allowance at a slot-constrained hub is invisible in a timetable and decisive in a punctuality report.
The airborne segment has its own asymmetry. Arrival punctuality is partly purchasable. A crew that leaves twelve minutes late can recover several of them by flying a higher cost index, burning fuel to buy schedule. Departure punctuality cannot be recovered at all — it is a photograph of one instant. Any comparison that mixes departure and arrival figures, or that quotes arrival figures without saying what the operator’s cost index policy is, is comparing two different willingnesses to spend money.
The trap. A rising arrival punctuality with a flat departure punctuality is not, by itself, evidence of an operational improvement. Rule out the two cheapest explanations first: buffer added to the schedule, and fuel burnt to catch up. Both raise the same headline figure, both cost money elsewhere, and neither says anything about the turnaround.
The comparison that does not work
Take a single city pair and two operators. The arithmetic below is a constructed illustration, not measured data — its only purpose is to show how far apart two aeroplanes can be while producing an identical published figure.
| Operator A | Operator B | |
|---|---|---|
| Published block time | 1 h 55 | 2 h 15 |
| Median actual gate to gate | 1 h 58 | 1 h 58 |
| Median arrival versus schedule | 3 min late | 17 min early |
| Share arriving within 15 min | High | High |
| Real elapsed time for the passenger | Identical | Identical |
| Connection built on the schedule | Tight but honest | Protected by 20 min of buffer |
Both operators can advertise the same punctuality. Operator B has bought it with twenty minutes of published time, which it pays for in aircraft utilisation, in crew cost under agreements that pay the greater of scheduled and actual block, and in one fewer rotation per aircraft per day on a tight network. Operator A has not padded, and is punished by the metric for telling the truth about the route. Neither aeroplane flew any faster than the other.
The buffer is not wasted, though. It absorbs reactionary delay, which European delay reporting consistently ranks among the largest delay families. The IATA standard delay codes reserve a block for exactly this — aircraft rotation, crew rotation, connecting load — and a first-wave departure carries none of it. This is why punctuality reliably decays through the operating day, and why comparing two operators whose aircraft fly different numbers of sectors per day compares different exposure to propagation rather than different competence.
Where the schedule is denser than the airport
No amount of buffer fixes a schedule written against more capacity than exists. At a slot-coordinated aerodrome, the declared capacity is a figure set for the aerodrome after consultation with the airport, the air navigation service provider and the users, covering runway, apron and terminal, and slots are allocated against it under the framework in the EU slot regulation and the industry Worldwide Airport Slot Guidelines. If the sum of scheduled movements sits at the edge of that capacity, queueing theory does the rest: delay stops being an event and becomes a property of the timetable.
Note also what a slot is and is not. Slot adherence is measured against the allocated slot with its own tolerance window, and the historic-rights usage rule — normally eighty per cent of a series, waived during exceptional periods — rewards operating the flight, not operating it on time. An operator can hold excellent slot compliance and poor punctuality, or the reverse. They are different instruments measuring different things, and they are routinely quoted as if interchangeable.
The tail matters more than the threshold
Any threshold metric is blind beyond its own edge. A flight that arrives sixteen minutes late and one that arrives four hours late count identically in an A15 figure. For a point-to-point leisure passenger the difference is an annoyance; for a connecting passenger, for a crew about to run out of flight duty period, or for an operator facing statutory compensation, it is the whole story.
That is why the legally significant thresholds are nowhere near fifteen minutes. Under Regulation (EC) No 261/2004, as interpreted by the Court of Justice of the European Union, the compensation exposure turns on an arrival delay at the final destination of three hours or more, with the amount set by distance band — a rule that lives entirely in the tail of the distribution that the published metric ignores. The equivalent US framework is built differently, on refunds and carrier commitments rather than on a general statutory delay compensation, which is one more reason not to read a European and an American punctuality figure as the same object.
The right reflex. Ask for three numbers together and never one alone: the on-time share, the completion factor, and the average delay of the delayed flights. The first is easy to manage, the second exposes the cancellations that flattered it, and the third exposes the tail.
Reading a figure before you compare it
The questions below take a few minutes and settle most arguments. If an operator, a bid document or a press release cannot answer them, the figure is not comparable to anything.
| Question | Why it moves the answer |
|---|---|
| Departure or arrival, and which recorded event? | Arrival figures can be bought with fuel; departure figures cannot |
| What tolerance, and is it the same on both sides? | A15 and D0 are different disciplines, not different presentations |
| Are cancellations excluded, and what is the completion factor? | Removing the worst flights raises the metric and lowers the service |
| What is the scheduled block time on this exact city pair? | The only direct measure of buffer available without internal data |
| Are feeder and franchise flights inside the sample? | Regional operations often carry the network’s worst punctuality |
| What period, and does it match season for season? | Summer convective weather and winter de-icing are not comparable months |
| Primary or reactionary delay, by code? | Separates a fragile schedule from a bad day at one airport |
Inside the operator, the discipline is D0
The published figure and the figure a station manager is judged on are usually different metrics, and the gap between them is instructive. Externally the tolerance is fifteen minutes; internally, serious operators run a D0 discipline, because everything downstream of the turnaround is a chain of milestones and a threshold destroys the information those milestones carry. If the target is “within fifteen”, a turnaround that finishes eleven minutes late is a success and nobody looks at it. If the target is zero, the same turnaround generates a question about which milestone slipped: aircraft on chocks, doors open, cleaning released, fuel complete, loading closed, final load sheet, doors closed, ready for pushback.
That is the practical reason a punctuality percentage is nearly useless as a management tool and quite useful as a public relations one. The percentage tells you how often you cleared a bar. The milestone data tells you which of eight sequential dependencies is fragile, whether the fragility is at one station or across the network, and whether the fix is a process change, a staffing change, or twenty minutes of schedule. Only the last of those three improves the published number without improving anything real.
There is a further wrinkle for anyone benchmarking across borders. Ground handling is contracted differently in different markets, and the milestone definitions inside those contracts are not standardised the way flight data is. A “ready” declaration at one station may require the load sheet to be signed; at another it may not. Comparing turnaround performance between two networks therefore inherits every local definitional difference at once, on top of the four already sitting inside the headline.
What the figure is actually for
Inside an operator, punctuality is rarely a vanity number. It sits in ground handling service level agreements with penalties attached, in capacity purchase agreements where a regional carrier is paid partly on performance, in airport service quality contracts, and in the internal case for adding a spare aircraft or a standby crew line. That contractual life is precisely why the definition is worth arguing about: a threshold moved by five minutes, or a decision on whether towed departures count, transfers real money between parties without changing a single flight.

The honest position is that scheduled block time has genuinely lengthened over decades, and that only part of that is buffer. Congestion, taxi distances at expanded aerodromes, flow management and denser terminal airspace all cost real minutes. You cannot separate the real minutes from the padded minutes using published data alone. What you can do is stop treating a single percentage as a performance statement, and start treating it as the output of four editorial choices — reference, event, tolerance and sample — that somebody made before the aeroplane ever moved.
Operational planning, rostering and schedule construction are discussed regularly by people who build these numbers on the FL330 forums, where the definitional arguments tend to be sharper than the published figures.

Leave a Reply
You must be logged in to post a comment.