Why waiting hurts - utilisation, spread and the serpentine paradox
A counter running at 80 per cent utilisation looks relaxed. One running at 95 per cent looks a little busier. The wait in front of it isn't a little longer - it's almost five times as long.
This is where gut feeling fails most reliably with queues. Watching the counter, you see a small change. Standing in the queue, you live through a big one. This article looks at why queues don't grow evenly but tip over, why the spread matters more than the average, and what makes waiting unbearable beyond the minutes themselves.
It builds on Managing queues, which covers the definitions, the four movements inside a queue and the effect of a displayed waiting time, and on Queues in operation, which is about stepping in.
Why small changes make a big difference
With queues, even small changes can make a big difference. A standard result of queuing theory shows why. For a single counter with randomly arriving customers, waiting time doesn't grow evenly with utilisation - it grows ever more steeply:
| Counter utilisation | Relative waiting time | How it feels in operation |
|---|---|---|
| 50 % | 1.0 | comfortable |
| 70 % | 2.3 | unremarkable |
| 80 % | 4.0 | getting tight |
| 85 % | 5.7 | practical ceiling |
| 90 % | 9.0 | fragile: more than twice as long as at 80 % |
| 95 % | 19.0 | breaking: almost five times as long as at 80 % |
| 99 % | 99.0 | collapse |
Utilisation here means the share of time the counter is actually serving. The table comes from the simplest model, with one counter and random arrivals. Real counters deviate from it, but the shape of the curve holds: fairly flat up to around 80 per cent, then steep.
Two practical rules follow. First: a counter that is almost always busy isn't efficient, it's on the edge. A little idle time is the buffer that absorbs the peaks. Second: if you want to intervene, you have to do it while the curve is still flat. That's exactly why good systems alert early, in stages - not only once the waiting time has passed the target.
Of the quantities in this curve, one can be measured directly: throughput, meaning how many people a counter handles per hour. How it is measured and read is covered in Queues in operation.
The problem isn't the average, it's the spread
The curve above carries a second message that's often missed. It applies to random arrivals and random service times. The more irregular both are, the sooner things get tight. Two counters with the same average utilisation can have entirely different queues - the one with more spread has the longer one.
Four sources make a queue unpredictable:
| Source | What it means | Where you can act |
|---|---|---|
| Arrival times | Not how many come, but how unevenly - peaks after a train arrives, an announcement, the end of a show | know the peaks and staff ahead of them |
| Different needs | Not every transaction takes the same time; a single long one blocks everyone behind it | send long transactions to a counter of their own |
| Different urgency | Whoever's flight is about to leave pushes to the front | a visible, fair rule for exceptions |
| Limited space | The queue spills over - into the aisle, onto the escalator landing, onto the pavement | treat it as a safety issue, not a service issue |
It all comes down to one sentence that sums up the programme: the more uncertain and unpredictable a queue is, the more important it is to manage it actively. An even queue forgives a lot. An erratic one forgives nothing.
Six reasons waiting hurts - and none of them is a clock
The same queue can be pleasant or unbearable for the same length of wait. There are six reasons for this, and none of them is the duration itself. They shrink through information, design and something to offer the people waiting:
- No expectation, no control. Anyone who knows neither the length nor the duration loses the sense of being in charge of their own time. That's why a displayed waiting time does so much.
- Empty time. People with nothing to do and nothing to look at in a queue consistently overestimate how long it takes.
- It looks unfair. A disorganised queue raises the worry of being passed over - long before anyone actually jumps ahead.
- Nobody explains the standstill. A slow queue is tolerated. One that stops without explanation is not.
- Crowding. Beyond a certain density, boredom turns into physical discomfort because there's no freedom of movement.
- The others are competition. Where no clear order applies, pushing in is rewarded. That's a design flaw in the queue, not a manners problem of the people in it.
Three of the six can be fixed with information: the missing expectation, the unexplained standstill and the impression of being passed over. Two with design: the crowding and the competition. Only one - empty time - requires offering people something while they wait.
The serpentine paradox
When it comes to how a queue should be laid out, performance and perception pull in opposite directions. Several parallel queues each feel short. But a single long transaction blocks a whole lane, and whoever picks the wrong one watches the neighbours move past. A single serpentine that leads to the next free counter demonstrably moves people fastest - and looks the longest.
The contradiction can be resolved, with a number. The serpentine stays, because it's the faster layout. Next to it goes the current waiting time. The visible length loses its sting because the expectation no longer comes from the sight of the queue but from the display. The fastest actual wait and the best perceived wait stop being opposites.
The condition is that the number shown is right - or honestly says it's an estimate. A display that promises four minutes and delivers twelve destroys exactly the trust it depends on.
What follows for operations
- Step in early. The curve is fairly flat up to about 80 per cent and then steep. Reacting at the target time is reacting too late.
- Reduce the spread, don't just add capacity. A separate counter for long transactions can do more than an extra one for everyone.
- Explain the standstill. An announcement or a note on the display costs nothing and takes the sting out of one of the six reasons.
- Make the order visible. Clear, fair queue guidance is cheaper than any argument at the counter.
- Show the waiting time. It works on perception and on the system at once, because the people waiting spread themselves differently.
What belongs in the tender
- Is throughput measured per counter, and at what resolution?
- Does the system alert in stages before the target time is exceeded?
- Can the waiting time be displayed publicly, and does the display show whether it is measured or estimated?
- How often is the display updated?
- What does the display show when there is no reliable figure for a queue?
How we do it
In the ANALYSIT Counting System, exits from the queue zone count as people served. That gives measured throughput per queue - averaged per hour, in the peak hour and over the last 15 minutes - plus the headcount in the queue. Prompts at 80 and 95 per cent of queue capacity, on spikes and on SLA breaches are generated every 15 minutes and go to the people responsible by email, SMS, WhatsApp, Slack or Teams.
The public waiting-time board runs without login via a revocable key and refreshes every 5 seconds. It shows each queue's waiting time with colour status and trend, and awards “Best Choice” only between measurable queues. Every waiting time visibly carries whether it is MEASURED or ESTIMATED: measured as the median of observed waiting episodes once there are enough of them, otherwise estimated from the headcount and the configured service time. Where there is no figure, it says “unknown”.
If you want to know how close your counters are running to the steep part of the curve, talk to us. Throughput is the figure to start with.