Every ground handling operation produces a roster the night before. It’s a real piece of work — someone spent hours on it, it balances the banks, it respects the working-time rules, and on paper it covers the day.
By nine in the morning it describes a day that isn’t happening.
This isn’t a criticism of whoever built it. The roster was correct given what was known at 05:00. The issue is what a roster actually is, and almost nobody says it out loud: it’s a prediction about a system that changes faster than the prediction can be revised.
What breaks it, specifically
Take a normal morning. Nothing dramatic — no snow, no strike, no aircraft on ground.
An inbound slips twenty minutes. The crew allocated to that turnaround is now idle for twenty minutes, and whatever they were doing next has moved.
A stand changes. The walk from the crew room is four minutes longer, or the equipment is at the wrong end. Small, unless it’s the third one that morning.
Someone doesn’t come in. Not a crisis. A cold, a car, a child. On a shift of forty people this happens most days, and the person who’s missing is never the one whose skills you have spare.
A special service request appears that nobody rostered. Wheelchairs, unaccompanied minors, a late cargo change. Individually trivial. They arrive in ones.
A piece of equipment is down. The belt loader is being fixed, so that turnaround needs a different approach and probably an extra person.
Something takes longer than the standard time says. It usually does, and the plan assumed it wouldn’t.
Six things. None of them unusual. And here’s the part that makes it structural rather than annoying: they don’t add up, they multiply.
The reassignment cascade
Move one person and you haven’t solved a problem, you’ve moved it.
The crew you pull off the delayed inbound to cover the sick colleague were going to do the 08:15. So somebody else does the 08:15, and they were on the 08:40. By the third hop you’re four turnarounds away from the original delay, and nobody in the operation can see that the 09:20 is now covered by someone who’s been on shift since 04:30 and is about to hit a break they’re legally owed.
The supervisor solving this in real time is doing something genuinely difficult and doing it well. They’re also doing it with a radio, a printed sheet, and whatever they can hold in their head — which means they’re solving the nearest problem, not the whole problem.
A locally sensible fix and a globally sensible fix are different plans, and the difference shows up as cost you never attribute to the decision that caused it.
That’s not a criticism either. It’s the only thing a human can do at that speed. But the overtime at the end of the shift, the turnaround that ran long at 11:00, the person who ended up doing four heavy loads in a row because each individual reassignment made sense on its own — by the time anyone measures any of this, the shift is over and the causal chain is gone.
Three properties that make ground handling worse than most
Plenty of industries schedule people. A few things make this one particularly unforgiving.
Demand swings by the hour, not by the season. A departure bank at 06:00 and a quiet 10:00 need completely different staffing, and the gap between them is larger than what many industries see across an entire year. There’s no smoothing to hide behind.
Staff are not interchangeable, and the constraints are legal ones. Who can push back an aircraft. Who holds a current airside driving permit. Who’s certified on that type. Who’s already worked six hours. A plan that treats people as units produces assignments that aren’t merely inefficient — they’re not permitted.
The failures are visible. In most operations a service failure is a number in a monthly report. Here it’s an aircraft that doesn’t leave and a passenger standing at a gate, and your customer is an airline with a contract that specifies exactly what you owe them.
Put those together and you get an operation where the plan degrades fast, the recovery is constrained in ways a spreadsheet doesn’t know about, and the cost of getting it wrong lands the same day and in public.
The trap: making the morning plan better
The instinctive response is to invest in the plan. Better forecasting, more detail, a longer planning session, a more sophisticated overnight optimizer.
Some of that helps. But there’s a ceiling, and it’s lower than people expect, because you’re improving a prediction about a system whose whole character is that it deviates. Past a certain point, effort spent making the 05:00 plan more precise is effort spent on a document that will be wrong in the same way regardless.
The thing worth improving isn’t the quality of the morning plan. It’s the time between reality changing and the plan changing.
If that gap is a shift, the plan is a document and your supervisors are the real scheduling system. If it’s minutes, the plan is live and the supervisors are doing the thing they’re actually good at — judgement, people, the call that no model should make.
Which means the useful question isn’t “how good is our roster.” It’s “how quickly can we produce the next one, and does anyone trust it when we do.”
How to find out if this is costing you
You can measure this without buying anything. Two exercises, one week each.
Two exercises, one week each
- Count the reassignments. Every time someone is moved from what the roster said, mark it. Just a tally.
- Then take ten of them and trace the chain. What caused it, what it caused, how many downstream moves followed, and whether it ended in overtime, a late turnaround, or nothing at all.
Most operations have never counted, and the number is usually several times higher than the management estimate. If the roster is being overwritten forty times a shift, the roster is not what’s running the operation.
That second exercise is the one that changes minds, because it makes the cascade visible for the first time. People know reassignments happen. Almost nobody has seen what a single one costs three hops later.
You’ll also learn something more useful than a number: which of the six triggers above actually dominates in your operation. Most people assume it’s delays. Frequently it’s absence, or equipment, and those have completely different fixes.
What a system should be doing instead
Not producing a better morning plan. Doing four things continuously:
Knowing the real state of the operation now, not what was planned at 05:00. Holding the constraints that determine feasibility — qualifications, hours, locations, equipment — as data rather than as supervisor knowledge. Re-solving fast enough that the answer is still relevant when it arrives. And presenting the result in a form a supervisor can act on in thirty seconds without opening a configuration screen.
That last one kills more of these systems than the maths does. A re-plan that arrives late, or that a supervisor doesn’t trust, gets ignored — and then you own an expensive tool and a radio.
The plan was never the product. The loop is. That argument is worked through in more detail in your optimizer was right and the shift still went wrong, and the qualification side of it in the new EU ground handling rules.
Optimum Intelligence builds optimization systems for ground handling and other operations where the plan and reality drift apart before lunch. If you want to know what your reassignment count actually is, we’re happy to talk — it’s a conversation worth having before anyone talks about software.