Where you make people stand changes the wait
Take the defaults: 96 customers in the peak hour, 75 seconds at the till, three lanes. Feed the three lanes from a single line and the average wait is about 33 seconds. Split it into three lines, one per till, and the average wait is 2.5 minutes — roughly four and a half times as long, with identical staffing, identical tills and identical customers.
The reason is that a single line never lets a lane sit idle while somebody is waiting. With separate lines, that happens constantly: a customer with one item finishes, that lane goes empty, and there is a person standing three feet away in the next line who cannot use it. Every second of that idle time is a second the system loses and never gets back. Pooling the queue is one of the very few changes in a shop that costs nothing, removes no staff and makes the service measurably better.
What it costs is floor. A single line needs its standing room in one place, and at the average length that is only a few feet, but averages are not what you build for. The switchback layout exists because the straight version reaches across an aisle whenever the peak runs a little hot.
The wait is not proportional to how busy you are
This is the part that catches people who plan lane counts from a daily average. Waiting time does not rise smoothly as the tills get busier. Between 50 and 70 percent utilization the wait creeps up. Between 80 and 90 percent it more than doubles. Past that it climbs without limit, because the queue no longer clears between arrivals — it just gets longer for as long as the peak lasts.
Run the lane table on your own numbers and you will see the knee. That shape is why "we are only at 85 percent capacity" is not the reassurance it sounds like, and why removing one lane from a busy bank of four is a completely different decision from removing one from a quiet bank of four. The same single lane can be worth twenty seconds or five minutes depending on where you are on the curve.
It is also why staffing from an average hour makes a shop feel slow even when the totals look right. If a seventh of the day arrives in one hour, that hour is running at two or three times the average arrival rate — and because waiting is not linear, it is not two or three times the wait. It is many times the wait.
Service time is a measurement, not an estimate
Everything above is driven by the seconds at the till, so it is worth getting that number honestly. Time twenty transactions with a phone, from the first item to the customer stepping away, and include the ones that go wrong — the card that declines, the price check, the customer who starts looking for a loyalty card after the total appears. Those are not outliers to be excluded; they are a large part of why the average is what it is.
The other reason to measure rather than estimate is that shaving the till time is often cheaper than adding a lane. Ten seconds off 75 is a 13 percent increase in what each lane can process, which at a busy utilization is worth more than it sounds — frequently more than the last lane you added.
What the model does and does not assume
The single line is modelled as M/M/c and the separate lines as independent M/M/1 queues. That means random arrivals at a constant average rate and exponentially distributed service times, and real checkouts violate both. Arrivals bunch. Till times in a shop selling two or three items are much more predictable than exponential, which means the model overstates the waiting; till times in a shop where basket sizes vary wildly are worse than exponential, which means it understates it. And customers jockey between lines, which claws back some of the gap between the two layouts without closing it.
So use it for the shape and for the size of the gap, not as a prediction. The conclusions that survive the assumptions are the ones about direction: pooling helps, the curve has a knee, and the knee is closer than it feels.
Related pages
The floor the queue stands on has to come out of the layout, which is the store fixture run layout calculator. The same queueing arithmetic applied to trucks and dock doors instead of customers and tills is the dock capacity and truck queue calculator, and it is worth a look if you want to see the same knee in a completely different setting.
For what the lane hours cost once loaded, the employee cost calculator builds the hourly figure this page asks you to supply, and the labor cost percentage calculator puts a whole schedule against a sales forecast. The takings side is sales per square foot.
Questions people ask
Is a single queue really faster than one line per till?
For the same staff and the same customers, yes, and the gap is larger than most people expect. At the defaults on this page it is about four and a half times the average wait. The mechanism is simple: with separate lines a till can go idle while somebody is waiting a few feet away, and that lost capacity never comes back. A single line makes that impossible. The cost is floor space and a layout that can hold the line without blocking an aisle, which is the real reason plenty of shops do not use one.
How many checkout lanes do I need?
Work it from your own peak hour, not from a daily average, because waiting time is not proportional to load and the average hour is not the hour anybody queues in. Count arrivals for an hour on your busiest day, time twenty transactions, and read off the lane count that meets whatever wait you are prepared to accept. This page will not tell you what that wait should be — that is a service standard you set — and it publishes no figure for what a shop your size typically runs.
Why does the wait blow up above about 85 percent utilization?
Because the queue only shrinks during the gaps between arrivals, and as utilization rises those gaps get shorter faster than the queue can use them. Below about 70 percent there is enough slack to recover; above 85 there is barely any, so a small increase in arrivals produces a large increase in waiting. Past 100 percent there is no recovery at all and the line grows for as long as the peak lasts. This is why capacity planned to sit just under full is planned to feel terrible.
Does self-checkout change the arithmetic?
It changes the inputs rather than the model. A bank of self-checkouts is a set of servers with their own service time — usually longer per transaction than a staffed till, sometimes much longer for large baskets — attended by one person who handles interventions. You can run this page twice, once for the staffed lanes and once for the self-service bank with its own measured time, but the interaction between them is not modelled here, and the attendant is a shared resource that becomes the bottleneck at exactly the moments the queue is worst.
Should I reduce till time or add a lane?
Compare them on the same table. Ten seconds off a 75-second transaction raises what each lane processes by about 13 percent, which at a busy utilization can be worth more than an entire additional lane and costs nothing per hour thereafter. Adding a lane is immediate and reliable but recurring. The honest answer is usually that the till-time work is worth doing first and the lane is what you add when the till time has stopped improving, and the way to tell is to run both changes through the lane table and read the two waits.