Queues in operation - open the counter before the queue tips over
A queue doesn't tip over all at once. It gives notice, ten or fifteen minutes ahead. React only once it has tipped, and you open the second counter for customers who have already left.
Managing queues covered what to measure at a queue: the five metrics, abandonment and balking, perceived against measured waiting time. This article picks up where measurement is in place and operations begin. How does a waiting time become a staffing decision - in time, at the right counter, with the right people?
Usually an allocation problem, rarely a staffing problem
The obvious answer to long queues is more staff. It's often the wrong one. Picture a business with several advice desks where waiting and service times are measured. Now you can see whether people are missing or simply in the wrong place. If the less experienced staff sit at the busiest desks, more headcount won't help - a different distribution will: the experienced staff where the demand is, and rotas set by the actual service intensity at peak times.
What matters is the order: first see where the time is lost, then decide whether you need more people or different positions. That takes three things, which this article works through in turn: a performance figure per counter, a signal before things tip, and an overview of every waiting area at once.
Throughput: the figure made at the counter
Every staffing decision starts with one question: how many people can a counter handle in an hour? In day-to-day practice the answer comes from the till system or from gut feeling. But the till only knows who bought something - not who was advised, returned an item or stood at the information desk. And gut feeling says one thing on a Friday evening and another on a Tuesday morning.
A sensor above the queue supplies the figure directly: whoever leaves the waiting zone has been through. That gives three values any shift lead understands:
- Throughput per hour, averaged over the day - the counter's performance.
- The peak hour and its value - what the counter can do when it matters.
- The last 15 minutes - what it is doing right now.
Combined with the number of people waiting, throughput gives the simplest and most robust calculation in queuing theory, Little's law: waiting time equals people waiting divided by throughput. If twelve people are queuing and three leave the queue per minute, the next person waits about four minutes. Anyone who sees these two figures live doesn't need a stopwatch to know whether a second counter is needed.
One distinction helps when reading it: throughput counts who leaves the zone. It's the performance of the queue, not the number of sales. Putting the two side by side is revealing - if throughput holds but sales drop, the problem isn't at the till.
Before it tips: a signal, not an alarm
A threshold that fires when the wait passes ten minutes fires too late. By the time the second counter is staffed, it's fifteen. Active queue management therefore works with levels that kick in earlier:
| Signal | What it says | What follows |
|---|---|---|
| Utilisation 80 per cent | The queue is filling up and the space in front is getting tight | get the second counter ready |
| Utilisation 95 per cent | The queue is practically full | open the second counter |
| Spike | Arrivals jump, after an announcement or a train arriving, say | move staff before the queue builds |
| Target time breached | The wait is longer than you promise | act and document it |
Two properties make these prompts useful in practice. First, thresholds are set per queue: the checkout has different ones from the fitting rooms, the information desk from the returns counter. Second, prompts can be acknowledged. That sounds like admin, but it's the actual tool: after a month, the list of acknowledged and ignored prompts shows how often the team reacted in time - and where a threshold sits in the wrong place.
How a system raises alerts without alerting too often or too late, and through which channels, is covered in Alerting: when a system may raise the alarm.
Ten waiting areas, one team
A large venue doesn't have one queue but many: entrance, security, tills, food and drink, toilets, cloakroom. There is one team. Without an overview it goes wherever someone last complained - to the loudest spot, not the worst one.
The fix comes with a rule that sounds simple and is often broken: a site's status is its worst queue, never the average. An average of nine quiet waiting areas and one overflowing one looks green and hides exactly the problem that matters.
Read over weeks, the same comparison shows something else: which waiting area is structurally too small. Moving staff around won't fix that; a layout change will. How zones are compared, and why that needs a minimum sample, is covered in Measuring dwell time.
The board that takes load off the team
Part of the steering can be handed to the people waiting. A public board shows each queue's waiting time, a colour and a trend - and marks the best queue with something like “Best choice”. Then guests spread themselves out, and the board does what a staff member with an outstretched arm would otherwise have to do.
For that to work, the board needs two rules. It only recommends between queues whose waiting time is known, and only when there are at least two comparable ones. And a queue without a reliable figure shows “unknown” rather than an invented minute. Why a visible waiting time lowers abandonment, and why every figure should carry its origin, is covered in Managing queues.
The same idea works in front of the door: people outside the pool, the theatre or the club want to know whether it's worth it. A display showing the count against capacity answers that before anyone joins the queue.
Staffing tomorrow properly
Steering in the moment is one half. The other is planning: staff are rostered for last week's rush. Tomorrow's is different - weather, holidays, a promotion.
Four tools help plan ahead instead of correcting after the fact:
- The visitor forecast for the coming days, with a band showing the uncertainty. When it can be relied on is covered in When a deviation is really a signal.
- The staffing recommendation per weekday and hour, derived from the 75th percentile rather than the mean. Why that value in particular is explained in The visitors who don't buy.
- The waiting-time pattern: the same hour on the same weekday over four weeks. It shows where waiting time builds up regularly.
- The quietest hours, which you can actively share with guests so demand spreads itself - described in Occupancy is a balance, not a count.
When Monday is no longer Monday
Thresholds and rosters rest on patterns, and patterns change. A new timetable, a new tenant in the building, a shift in home-working habits - and Monday morning is no longer the Monday morning it used to be. Carry on with old thresholds and you get alerted for normal days while missing the new peak on Tuesday.
What helps is anomaly detection that assesses weekdays and weekends separately and calculates robustly against individual outliers. It shows not only that a day was unusual but also when a new normal has set in - the moment to adjust the thresholds. How that works mathematically is in When a deviation is really a signal.
What belongs in the tender
- Is throughput measured per queue, and does the platform show the daily average, the peak hour and the last 15 minutes?
- Can thresholds be set per queue - waiting time and utilisation, in levels?
- Can prompts be acknowledged, and can you trace later how the team responded?
- Does the site status show the worst queue or an average?
- Does a public board only recommend between queues with a known waiting time, and how does it show a queue without a reliable figure?
- How do thresholds and alerts adapt when patterns change?
How we do it
In the ANALYSIT Counting System, exits from the queue zone count as people served: average throughput per hour, the peak hour with its value and the actual last 15 minutes, plus the headcount in the queue. Each queue has a target and a critical threshold in minutes, 5 and 10 by default, and a maximum capacity. Prompts at 80 and 95 per cent utilisation, on spikes and on SLA breaches are generated every 15 minutes, can be acknowledged or dismissed, and go by email, SMS, WhatsApp, Slack or Teams to the counter or store lead.
All queues at a site are on one page, and the site status is the worst measured queue. The public waiting-time board runs without login via a revocable key, refreshes every 5 seconds, shows colour status, trend and a live clock in local time, and awards “Best Choice” only between measurable queues, from two comparable ones upwards. Every waiting time visibly carries MEASURED or ESTIMATED; where there is no figure, it says “unknown”.
For planning there's the visitor forecast with a data-quality banner and uncertainty band, the staffing heatmap per weekday and hour from the 75th percentile, the four-week waiting-time pattern and the three quietest of the next twelve hours. Anomaly detection assesses weekdays and weekends separately, and the trend alert compares with the previous week, month or same week last year while guarding against days with a sensor outage.
If you want to know where time is lost at the queue in your operation, talk to us. The answer is often not more staff but a different position.