The Empty Column
A service can get faster for the people who make it through while serving fewer of the people who need it.
Before the clock starts
Imagine a library that lets neighborhood groups book a meeting room. Elena runs a sewing circle. Her group can meet on Tuesdays, and one of its members needs a room she can reach without using stairs. Elena used to call the branch and ask whether the Tuesday room was upstairs. Whoever answered could check the floor plan and, if needed, try the branch across town.
Now the library has an AI booking assistant, and the branch phone line directs callers to it. Ask for a room on Tuesday and it finds one in seconds. Ask whether that room is upstairs and it can only repeat the building’s general accessibility information. The assistant asks Elena to choose a room before it will open a reservation. She tries a second branch, gets the same answer, and closes the page.
On Monday, the library sees an improvement in booking speed. A patron who knows exactly what room to ask for has reason to celebrate. Elena’s group is still looking for somewhere to meet. In this example, the clock starts when a reservation request opens. Elena never makes it far enough to start the clock.
Change who counts
Suppose 100 groups need a room in each period. Each bar shows bookings that reached confirmation as a share of the group you select.
Illustrative numbers. Among confirmed bookings, median time falls from 12 minutes to 3. Ten fewer groups get a room through the library.
Look outside the queue
There are three questions here. How fast did confirmed bookings move? How many requests became bookings? How many of the groups that needed a room got one through the library? Each answer can move in a different direction. The GOV.UK Service Manual defines digital completion rate in terms of transactions people start. To understand the wider journey, it also recommends methods beyond digital analytics, including user research. A transaction rate is useful when the question begins where the transaction begins.
A Pennsylvania experiment shows what you can learn by starting somewhere else. Economists Amy Finkelstein and Matthew Notowidigdo began with roughly 30,000 older adults who appeared likely to qualify for federal food assistance but weren’t enrolled at the start of the study. Over nine months, about 6 percent enrolled in the control group, 11 percent in the group informed they were likely eligible, and 18 percent in the group offered both information and application assistance. The researchers used Medicaid records to build the study cohort before sending any letters, so they could measure who came forward and who stayed out of the queue.
The same study found a limit to that success. People who enrolled because of the interventions were, on average, less sick and had higher net income than the typical control-group enrollee. The researchers also found signs of greater need among people who never called in response to outreach. A service can bring more people through its door and still leave a harder group to reach on the other side.
Give someone the question
I think about this as a question of where human attention goes. The AI can keep handling ordinary bookings in seconds. A librarian can look at the cases the system struggles to hear and find out what they have in common. Even one afternoon listening at the branch might reveal that “Is the room upstairs?” is a booking question. The floor plan can then become part of the booking service, available to the person at the desk and to the assistant.
There are two different ways to look for missing requests. To see where people get stuck, give them an easy way to ask for help before they choose a room, and count those contacts alongside online and phone requests. To learn about people who never approach the library, talk to groups through the places where they already meet. Ask what they needed and what they ended up doing. A group that found a better room elsewhere has a different story from a group that called off its meeting. Sampling outside the service helps you hear both.
My test for a faster front door is whether it gives the people running the service more attention to spend on the people who have trouble getting in. The first sign of improvement at the library might be a rise in recorded problems. For the first time, someone would have a place to enter Elena’s question.