What self service check in throughput really depends on at a busy entrance
Self service check in throughput is the number of completed badge collections per kiosk per hour, set by the full cycle of approach, lookup, print and clear. A 9 second interaction plus a 7 second print gives 225 an hour, and adding 4 seconds of clear time drops it to 180, which changes a 3,400 person peak from 16 kiosks to 19.
A supplier quotes a kiosk that completes a check-in in nine seconds. Nine seconds sounds excellent, so somebody divides 3,600 by 9, gets 400 an hour, and orders nine kiosks for a peak of 3,400 people. On the morning it turns out that nine kiosks clear rather fewer than 3,600 people an hour and there is a queue.
Self service check in throughput is the number of completed badge collections per kiosk per hour, and the nine second figure describes one part of a cycle with several parts. Getting the cycle right is most of the work.
The cycle, and the part everyone leaves out
Watch somebody actually use one of these and count what happens.
They approach and orient themselves, which takes a second or two even for a confident user, longer if the screen is showing an idle animation and they have to work out where to start. They do the lookup, either scanning the QR code from their confirmation email or typing a surname and picking themselves from a list. The badge prints. Then they take the badge, check it is theirs, find a badge holder, and step away.
The supplier's nine seconds usually covers the middle: from first touch to the print command. It excludes the approach on one side and the collection on the other, and both of those occupy the kiosk just as completely as the lookup does. A kiosk with a person standing at it holding a badge and looking for a lanyard is a kiosk nobody else can use.
So the cycle is approach, lookup, print, clear. Four phases, and the last one is the one that goes missing from every capacity calculation I have seen.
How many kiosks does a 3,400 person peak need?
Work it both ways, because the difference is the point.
Take the supplier's figures at face value: 9 seconds of interaction plus a 7 second badge print gives a 16 second cycle. Three thousand six hundred divided by 16 is 225 completions per kiosk per hour.
Your morning peak is 3,400 people. Divide 3,400 by 225 and you get 15.1, so 16 kiosks. Fifteen would clear 3,375, which leaves 25 people short in the hour, and short compounds.
Now add the phase that was left out. Four seconds of clear time, which is generous to nobody who has ever watched somebody thread a badge onto a lanyard, takes the cycle to 20 seconds. Three thousand six hundred over 20 is 180 an hour. Three thousand four hundred over 180 is 18.9, so 19 kiosks.
Three extra kiosks, a 19 per cent increase in the estimate, entirely from four seconds nobody counted. At the unit cost of a kiosk plus its printer plus the floor space plus the power, that is a real number, and it is the difference between an entrance that works and one that does not.
There is a way to get those three kiosks back without buying them, which is to move the clear phase off the kiosk. Put a badge holder table two metres away with a member of staff at it, and design the flow so people take the badge and walk. The cycle drops back towards 17 or 18 seconds, and you have swapped capital for one person.
Why the printer is usually the constraint
Look at where the 20 seconds sits and the priority becomes obvious.
Approach is two or three seconds and is mostly a design problem you can improve with a clearer idle screen and better floor marking. Lookup, for somebody scanning a QR code, is under a second of machine time and a few seconds of human time getting the phone out. Clear is behavioural and you fix it with layout.
The print is machine time and it is the largest single block. A thermal printer producing a common exhibition badge at a legible density takes what it takes. No software change compresses it, no queueing improves it, and no amount of interface work makes it shorter.
This puts a specific question at the top of your list, and it belongs to procurement. What badge stock, at what size, at what print quality, on what printer model, and what does that combination cost in seconds? Halving 7 seconds to 3.5 takes the full cycle from 20 to 16.5 seconds, which is 218 an hour, which needs 16 kiosks for the same 3,400. The same trade appears on staffed desks, where splitting registration desk throughput into lookup time and print time turns out to be the highest-value thing you can measure.
The other route is to remove the print from the peak entirely. Badges posted in advance, or printed the day before for pre-registered attendees, convert a 20 second kiosk transaction into a 4 second collection. The economics of postage against kiosks needs working out properly, and the answer depends heavily on what share of your audience registers in the last fortnight.
Why does a seven second print feel so long?
Because seven seconds sits in a gap that human factors research identified decades before anybody built a badge kiosk.
Nielsen set out three response time limits in Usability Engineering in 1993. One tenth of a second is roughly the limit for a system to feel instantaneous. One second is roughly the limit for a user's flow of thought to stay uninterrupted, though they will notice the delay. Ten seconds is roughly the limit for keeping attention on the task at all, and beyond that people expect to be told when the machine will be finished.
Miller had reached compatible conclusions in 1968, in a paper for the AFIPS fall joint computer conference, observing that silences of more than about four seconds become awkward because they break the thread of the exchange, and that feedback effectiveness has a boundary at around two seconds.
Seven seconds falls between the one second limit and the ten second limit. It is far too long to feel responsive and comfortably short of the threshold where any designer thinks to add a progress indicator. So the common kiosk gives the user a blank pause of seven seconds after they press the button, at the exact moment they are least sure they did the right thing.
The behavioural consequences cost throughput. Some people press the button again, which either queues a second badge or does nothing while they wonder. Some step back to look for a member of staff. Some walk away, and an abandoned transaction is worse than a slow one, because the kiosk was occupied and produced nothing.
None of this is fixed by a faster printer, and all of it is fixed by telling the user what is happening. A progress bar with the words describing the print, and a clear indication of where the badge will emerge, costs nothing in seconds and removes most of the abandonment. It will not raise your theoretical throughput by a millisecond, and it will raise your actual completions.
One queue, or a queue at every kiosk?
The last decision that moves throughput has nothing to do with the machines.
A bank of kiosks with a separate queue at each one is slower than the same bank fed by one queue, and the reason is that separate queues cannot share slack. When kiosk four gets somebody with a problem, the four people behind them wait, while kiosk seven stands idle with nobody in front of it. A single queue feeding whichever kiosk becomes free next never has that state.
The effect is largest exactly where you need it, because it grows with the variability of service time, and kiosk service time is highly variable when a mix of QR scanners and name typers use the same machines. Set a single serpentine feeding all kiosks, mark it on the floor, and put one person at the head of it directing people to whichever is free.
That person is worth more than they look. They also catch the attendees who cannot use a kiosk at all before those people occupy one for ninety seconds. Where those staff should be, and in what numbers across the morning, comes out of the same bin-by-bin logic as a registration desk staffing model.
Where this stops
Every number above assumes the kiosk works and the person can use it. Both assumptions fail at rates that are hard to predict and easy to measure afterwards.
Failure to complete is the honest gap in this model. Somebody whose email address does not match, whose registration is under a colleague's name, whose company registered them as part of a group, or who cannot find themselves in a list of twelve identical surnames will occupy a kiosk and then need a human. If 8 per cent of attempts end that way, your effective throughput is 92 per cent of the calculated figure before anything else goes wrong, and 19 kiosks becomes 21. Pull that percentage from your own logs, because it varies enormously with how you collected the registration in the first place.
The model is also silent on the population that will not approach a kiosk at all. That group exists at every show, it skews older and skews towards first-time attendees, and it is not small enough to ignore. Planning a self service entrance without a staffed fallback lane produces a queue of exactly the people least equipped to deal with one.
Finally, the arithmetic treats the peak hour as a single flat rate, which it never is. A 3,400 person hour with 900 of them arriving in one ten minute window behaves quite differently from an even one, and the kiosk count that survives the flat calculation can still fail on the day.
The step for this week is a stopwatch and twenty minutes. At your next event, stand beside the kiosks and time the full cycle, from a person arriving in front of the machine to the next person being able to start, for twenty consecutive users. Take the median. If it is materially above the supplier's figure, and it will be, redo your kiosk count with your own number before the next contract, and record the result somewhere your attendee analytics reporting can find it next year.
Questions people ask about self service check in throughput
- How many self service kiosks does a registration hall need?
- Divide the peak hour arrivals eligible to use them by the completions one kiosk achieves in an hour. At a 20 second full cycle a kiosk completes 180 an hour, so a 3,400 person peak needs 19. Using a 16 second cycle that ignores the time people take to collect a badge and step away under-provisions by three kiosks.
- Is the printer really the bottleneck on a check-in kiosk?
- Usually, once the interface is competent. A QR scan resolves in under a second and a name lookup takes a few seconds, while a thermal badge print at a common exhibition badge size runs several seconds and cannot be compressed by software. Halving the print time raises throughput more than any interface change available to you.
- Why do people abandon a self service kiosk?
- Most abandonment happens during an unexplained pause. Nielsen set out in 1993 that one second is the limit for keeping a user's flow of thought uninterrupted and ten seconds the limit of attention. A seven second print with no feedback sits between those, long enough to feel broken and short enough that nobody thought to add a progress indicator.
Related reading
- Measuring registration desk throughput per lane and per minute on show morning
- A registration desk staffing model built on quarter hour arrival bins