Skip to content

Predicting registration queue wait time from arrivals and service rate

Attendee analyticsUpdated 2026-08-188 min read

In short

Registration queue wait time can be estimated without a stopwatch using Little's result of 1961, which relates the average number in a system to the arrival rate multiplied by the average time in it. A queue of 210 people with 48 arrivals a minute implies a mean wait of about 4.4 minutes, and the estimate holds only while the queue is stable.

Somebody in the show office wants a number to put on the screen at the head of the queue, or on the radio to the marketing team who are fielding complaints. The obvious method is to send a steward to the back of the line with a phone timer and see how long it takes them to reach the front. That works, it takes as long as the wait itself, and by the time it produces an answer the answer is old.

Registration queue wait time can be estimated in under two minutes from two counts anyone standing in the hall can make. The arithmetic behind it has been settled since 1961.

The result that makes this possible

Little published a proof in Operations Research in 1961 for the formula written as L equals lambda times W. The average number of items in a system equals the average arrival rate multiplied by the average time each item spends in it.

What makes it useful here is what it does not require. Little's proof needs the means to be finite and the underlying processes to be stationary, and it makes no assumption at all about the shape of the arrival distribution or the service distribution. It holds for a queue where people arrive in coaches and for one where they trickle in. It holds whether service times are consistent or wildly variable. Almost nothing else in queueing theory is that undemanding.

Rearranged for what you want, W equals L divided by lambda. Time in the system equals the number in the system divided by the arrival rate.

So you need two counts. How many people are in the queue, and how many join it per minute.

Counting both numbers in the hall

The queue length is the easier of the two, and the usual mistake is to guess it. Guessing runs low, consistently, because a queue folded into a serpentine looks smaller than it is.

Count it properly instead. Take one full lane of the serpentine, count the people in it, then count the lanes. Eight people per lane across twelve lanes is 96, and a person standing on a chair can do that in forty seconds. If you have overhead cameras on the queue area, that count is already available, and if you have entry counters at the queue entrance and the desks, running entries minus completions gives it continuously.

The arrival rate needs one minute of attention. Stand where people join the back of the queue and count them for sixty seconds. Do it twice, at either end of a five minute window, and average the two, because a single minute during a coach drop is unrepresentative.

Neither count needs any system. Both are better if the system provides them, and the point of doing it by hand first is that you will find out whether the system's version agrees.

What does a 4.4 minute answer actually mean?

Work an example. At 09:20 the queue holds 210 people, and counting the back of the line gives 48 arrivals a minute.

Two hundred and ten divided by 48 is 4.375, so about 4.4 minutes. That is the mean time a person spends in the queue, provided the queue is in a steady state during the period.

That proviso does more work than it looks like. Steady state means the queue is neither growing nor shrinking, which means arrivals and completions are balanced. If your desk is clearing 47.6 people a minute while 48 arrive, the queue is drifting up slowly and the estimate is close enough to publish.

Now change one number. Suppose arrivals are running at 56.7 a minute, which is what a 3,400 person peak hour looks like spread evenly, while the desk still clears 47.6 a minute. Divide 210 by 56.7 and you get 3.7 minutes, which is lower than the previous answer even though the situation is considerably worse.

That result is wrong in the way that matters, and the reason is worth holding on to. A person joining the back of a growing queue has to wait for everyone ahead of them to be processed, and processing happens at the service rate. Divide by the service rate instead: 210 over 47.6 is 4.4 minutes. When the queue is growing, the arrival-based figure understates the wait, and the gap between the two numbers tells you how fast the situation is deteriorating.

The rule I would use on the day is simple enough to hand to a steward. If arrivals and completions are roughly equal, divide the queue by either. If arrivals exceed completions, divide by the completion rate and expect the answer to get worse over the next fifteen minutes. Getting a reliable completion rate is a separate measurement, and registration desk throughput per lane per minute is where it comes from.

Run the same two counts three times across a morning and the shape of the problem becomes obvious without any modelling. At 08:50, 60 people queueing against 34 arrivals a minute and 47.6 completions gives a wait of 1.3 minutes and a queue that is shrinking. At 09:20, 210 against 56.7 arrivals and the same 47.6 completions gives 4.4 minutes and a deficit of 9.1 people a minute. At 09:50, if that deficit has held for half an hour, the queue has grown by roughly 273 people to about 483, and 483 over 47.6 is 10.1 minutes.

The useful part of that sequence is the middle reading. At 09:20 the queue was only 210 and the wait was still under five minutes, which looks survivable on a radio call. The deficit of 9.1 people a minute was already telling you what 09:50 would look like. Watching the deficit rather than the queue length buys you thirty minutes of warning, and thirty minutes is enough to open two more lanes.

Why does the wait people report differ from the calculation?

You can hit 4.4 minutes and still take the complaints, and the reason has been documented since the mid-1980s.

Maister set out what he called the first law of service in 1985, in a chapter for a collection on the service encounter edited by Czepiel, Solomon and Surprenant. He wrote it as satisfaction equals perception minus expectation. Both terms on the right are psychological. Neither is the measured wait.

The operational consequence is that two levers exist and most organisers only pull one. You can reduce the wait, which is expensive, since it means more lanes, more staff and more printers. Or you can move the expectation, which is nearly free.

An attendee who was told at the door that the wait is about five minutes, and who waits four and a half, has had a good experience of a four and a half minute queue. An attendee who expected to walk straight in has had a bad experience of the same queue. The wait is identical. The satisfaction differs, and the difference cost you a sign and a steward.

This is why publishing the estimate is worth more than the estimate itself. A number at the head of the queue, even an imperfect one, sets the expectation that the perception gets measured against. It also stops the worst version of the experience, which is a person who has no idea whether they are five minutes or forty from the door and cannot decide whether to leave.

If you do publish it, round up. A queue signed at six minutes that takes four and a half produces a small pleasant surprise. A queue signed at four that takes four and a half produces a complaint about the sign as well as the wait.

Where this stops

Little's result is exact under its own assumptions, and a registration hall at 09:15 on day one violates the main one continuously. Arrivals are not stationary. They ramp up, peak and fall away inside two hours, and every estimate you compute is a snapshot of a system that is changing while you compute it.

Practically this means the number is good for the next few minutes and worthless for the next hour. Do not use it to plan. Planning against a varying arrival profile needs a bin-by-bin approach, and building a proper registration desk staffing model works from expected arrivals per quarter hour rather than from a live queue count.

There is a second limit that catches people out. The calculation assumes one queue feeding all lanes. Many registration halls run separate queues by badge type, by language, or by pre-registered against on-site, and each of those is its own system with its own arrival rate and its own service rate. Averaging across them produces a number that describes nobody's experience. Count and compute each queue separately, and be prepared for the answer that one queue is at ninety seconds while another is at eleven minutes, which is an argument for merging them rather than for a better sign.

Finally, the estimate says nothing about the tail. A mean wait of 4.4 minutes is consistent with most people waiting three minutes and a handful waiting twenty, and the handful are the ones who tweet. If you want the tail you need the actual time in system per person, which means timestamping entry to the queue as well as completion, and that needs hardware at the queue entrance rather than arithmetic.

The step for this week costs ten minutes at your next event. Count the queue, count arrivals for one minute, divide, and then send somebody to the back of the line with a timer and see what they get. If the two numbers land within a minute of each other, you have a live wait estimate you can put on a screen next edition, and it came out of your existing attendee analytics rather than out of a procurement.

Questions people ask about registration queue wait time

How do you calculate queue wait time at a registration desk?
Count the number of people in the queue and count arrivals joining it over one minute. Divide the queue length by the arrival rate. A queue of 210 with 48 people joining each minute gives 210 divided by 48, which is 4.4 minutes. The calculation assumes the queue is neither growing nor shrinking during the period you measured.
Why does the calculated wait time differ from what people experience?
Maister set out a first law of service in 1985 stating that satisfaction is perception minus expectation. Both terms are psychological, so a five minute wait against an expectation of two feels worse than a nine minute wait somebody was warned about. The measured wait and the felt wait move for different reasons and need managing separately.
What do you do when the registration queue is growing?
Divide the queue length by the service rate instead of the arrival rate, because a person joining now must wait for everyone ahead of them to be processed. If the desk clears 47.6 people a minute and 210 are waiting, the true wait is 4.4 minutes even though the arrival-based figure reads lower. The gap between the two numbers is the warning.

Related reading

All on-site analytics articles