Skip to content
allanninal.dev/build
Collaborate

Part 4 of 7 · Visitor check-in logger series ~5 min read

Why checking out is the hard part

Every visitor management system solves check-in, because the visitor is motivated. Check-out has no such force behind it and needs to be engineered around rather than requested.

Key takeaways

  • Nobody is motivated to check out. Asking harder does not work.
  • Use four mechanisms; each catches a different group.
  • The host is the most reliable single source and needs one tap.
  • An end-of-day sweep closes the rest, marked as assumed rather than confirmed.
  • Never auto-close during the day. That is the hour the alarm goes.

The asymmetry

A visitor checks in because a door is closed and a person is waiting to be told they have arrived. The incentive is immediate and personal. On the way out the door opens from the inside, the meeting is over, and the visitor is thinking about the train.

No amount of signage changes that, and a system whose check-out relies on the visitor remembering is a system with an inaccurate roll call by mid-morning. So the design uses several mechanisms and expects each of them to work partially.

Four mechanisms

Four mechanisms for checking visitors out and their approximate coverageThree boxes stacked on the left. Visitor taps out at a screen by the door, labelled catches about thirty per cent. Host checks them out with one tap from a message, labelled catches about forty-five per cent. Badge returned to a box or a reader, labelled catches about fifteen per cent. All three converge on The rest, unconfirmed, and that leads down to an End-of-day sweep, marked assumed rather than confirmed. A note says no single mechanism is close to sufficient and together they get most of the way.Visitor taps outa screen by the doorcatches ~30%Host checks them outone tap from a messagecatches ~45%Badge returnedto a box or a readercatches ~15%The restunconfirmedEnd-of-day sweepmarked assumed, not confirmedNo single mechanism is close to sufficient. Together they get most of the way.
Fig 1. The four check-out paths and roughly what each catches. The design assumption is that all of them are partial.
  • Compute
  • Security & identity
  • Management
  • Front-end & mobile
  • People

The host is the best source

The host knows the meeting ended, is at a desk, and has a phone. A message at a sensible interval — “has Sarah Chen left?” with two buttons — is answered a good proportion of the time, and it is the single highest-yield mechanism.

The timing matters. Sending it as the meeting was scheduled to end is better than sending it two hours later, and asking once is better than asking three times. A host who is nagged stops answering.

The badge

A physical badge returned to a box is a genuine signal and it is worth capturing even crudely: a member of reception clearing the box twice a day and tapping the names is enough, and a reader is better if one exists.

It is also the mechanism that produces the useful secondary metric: badges that never come back. A steady rate of unreturned badges is a small cost and a reminder that the roll call for those people was closed by assumption.

The end-of-day sweep

How remaining checked-in visitors are closed at the end of the dayA vertical chain of five steps entered by a box labelled End of the working day, at a stated hour. Step one lists who is still checked in after all mechanisms. Step two asks the host once with a final message; a reply exits to Confirmed out, with a time. Step three asks whether any are known to be here for an evening event or late meeting; if so it exits to Leave them on, genuinely present. Step four closes the rest as assumed, never as confirmed. Step five counts them as the accuracy metric. A note says the distinction in the fourth box is what keeps tomorrow's roll call honest.AWS ACCOUNTEnd of the working daya stated hourStill checked in?after all mechanismsAsk the host oncea final messageConfirmed outwith a timereplyAny known to be here?evening event, late meetingLeave them ongenuinely presentyesClose as assumednever as confirmedCount themthe accuracy metricThe distinction in the fourth box is what keeps tomorrow's roll call honest.
Fig 2. How the day is closed out. Marking the swept records as assumed rather than confirmed is what makes the accuracy measurable.
  • App integration
  • Machine learning
  • Management
  • Analytics
  • People

Never sweep during the day

The obvious optimisation is to auto-close visitors after three or four hours, and it is dangerous for exactly one reason: the alarm goes during the working day, and somebody auto-closed at two o’clock who is still in a meeting at half past two is invisible on the roll call.

Marking them as unconfirmed on the roll call achieves the same reduction in noise without removing them, which is the correct trade for a safety list.

Measuring the accuracy

Confirmed versus swept visitor check-outs over three quartersA stacked bar chart with three bars in per cent. Two series: confirmed check-outs in green, and closed by the sweep in orange. January: sixty-one per cent confirmed and thirty-nine swept. April: seventy-four confirmed and twenty-six swept. July: eighty-one confirmed and nineteen swept. A note says the green share is the only meaningful measure of whether the roll call works.050100150200~100Jan~100Apr~100JulConfirmed check-outs, %Closed by the sweep, %The green share is the only meaningful measure of whether the roll call works.
Fig 3. Confirmed against assumed check-outs over three quarters. Improving the green share is the whole goal, and it is only visible because the two are recorded differently.

Improvements come from small things: moving the check-out screen to where people actually leave, changing when the host message is sent, putting the badge box somewhere on the way out rather than behind reception.

None of them are clever and all of them are measurable, which is the point of distinguishing confirmed from assumed in the first place.

Next: the people who are not visitors.

All posts