Railway ticket kiosk: remove a step rather than add one
A suburban railway station installs ten self-service kiosks for unreserved tickets. The flow has six screens, opening with language selection (English, Hindi, Marathi). At step 3 the passenger picks a train from a list of 40 services shown by four-character code and departure time. Abandonment is 34% at step 3; median completion for those who finish is 2 min 40 s; the staffed counter queue averages 12 minutes. Of observed users, 60% use a smartphone, 25% a feature phone and 15% neither; 40% are over 50. Staff report two recurring behaviours: people tap a language they cannot read to get past the first screen, and they scroll the train list until they give up. The kiosk allows one more physical control, and there is budget for exactly one software release.
1. Should the next release add a help button or remove a step? Argue from the numbers.
Read the worked answer
Remove, do not add.
The failure is localised: 34% abandon at one screen, while the median completion of 2 min 40 s for those who finish says the rest of the flow is fast enough. A help button adds a seventh decision and a new failure mode. With 40% of users over 50 and 15% carrying no phone at all, a help screen presumes a confidence this group has already demonstrated it lacks - they were abandoning, not asking.
The 12-minute counter queue is the real benchmark. Any change has to move people off that queue, and a help button mostly helps the people who were already going to succeed.
The one extra physical control should be a large, high-contrast cancel-and-start-over button. The observed failure is a trap - the wrong language, the wrong screen - not missing information.
2. Redesign step 3. State what it must show and what it must stop showing.
Read the worked answer
Show five or six departures from this station, each as a row at least 18 mm high containing the destination name in the local script, departure time, platform number and fare. If the destination is ambiguous, offer a list of destination names in local script that can be scrolled, not a list of numbers.
Stop showing four-character train codes as the primary label. A code is only useful to someone who already knows it, which excludes the first-time user by definition. Stop showing all 40 services; the list length is what produces the scroll-and-surrender behaviour.
Also stop forcing a language choice before anything useful appears. Open with the local language already selected and a persistent, large language control at the edge of the screen. The staff observation that people tap a language they cannot read is a design indictment, not a user error.
3. Give one metric that would show success and one that could mislead you.
Read the worked answer
Success: tickets completed per kiosk per hour during a defined peak window, paired with abandonment at the train-selection step, both measured before and after on the same weekday. Two numbers, because one without the other can be gamed by making the flow faster rather than better.
Misleading: average transaction time. It falls whenever more people abandon early, so it reports improvement exactly when things get worse. A second misleading metric is counter queue length - it moves with festival traffic, weather and staffing, and cannot be attributed to the kiosks within a six-week window. If you report queue length as the outcome, you will take credit for a quiet week and blame the staff for a busy one.