Reply Windows: Setting a Response Boundary for a Nine-Person Team

The team’s median reply time was 11 minutes during working hours, which sounds excellent until the distribution is examined: 34 percent of replies arrived after 20:00 local time. Fast replies were not a sign of a healthy process. They were a sign that nobody had agreed when the day ended.

Nine people supported a payments integration used by 300 merchants. Messages arrived from three time zones and from automated systems that never slept.

The cost of being available

Because replies came quickly, nobody built a queue, and because nobody built a queue, nobody could plan. Work was interrupted an average of 6.4 times an hour during the morning, measured by one volunteer who logged switches for three days.

The interruptions were not the senders’ fault. The team had trained every counterpart to expect an answer within minutes.

The proposal

The team proposed two windows a day for non-urgent messages, at 11:00 and 16:00 local time, and a separate channel for genuinely urgent issues with a named on-call person. The proposal was rejected in its first form because it sounded like a refusal to help.

The second version renamed the concept. Instead of response windows the team published an expectation: a reply within one working day, an answer within two.

What actually changed

Almost nothing changed for three weeks, because the published expectation had no mechanism behind it. What changed it was turning off notifications outside the two windows for the whole team at once, which made the boundary real rather than aspirational.

The first day produced two complaints and one genuine escalation that had gone to the urgent channel correctly.

The urgent channel

The urgent channel needed a definition, because everything feels urgent to the person asking. The team settled on one test: a customer cannot complete a payment. Anything else waits for a window.

In the first month the channel carried 14 messages, of which 9 met the test. The five that did not were answered anyway, which the team decided was acceptable because punishing a wrong escalation teaches people to sit on a real one.

The measurement

After six weeks, replies after 20:00 fell from 34 percent to 6 percent. Median reply time rose from 11 minutes to 2 hours 10 minutes, which is the intended trade. Self-reported interruption counts fell from 6.4 to 3.1 an hour.

Merchant satisfaction, measured by a short survey on resolved tickets, did not move: 4.4 out of 5 before and 4.4 after.

What the boundary cost

Two people found the two-window system harder than constant availability, because batching replies meant holding several contexts at once. One of them returned to answering continuously and simply stopped logging it, which the team tolerated.

The boundary also needed an owner. Someone has to notice when the urgent channel starts carrying routine work, and that person has to be senior enough to say so.

Five months on

The published expectation is still one working day, and the median is two hours. The urgent channel runs at about a dozen messages a month with two thirds meeting the test.

The person who answers the urgent channel

The on-call rotation existed before the windows and had been informal. Once the urgent channel had a definition, the rotation needed a published schedule, because a message with no answer is worse than a message that was never sent. The schedule is now a shared calendar entry that rotates weekly.

Being on call changed how people worked that week. Two engineers said they batch their own deep work into the on-call week because they cannot rely on an uninterrupted afternoon, which is a sensible adaptation nobody had designed.

The quiet failure

Six weeks in, one merchant issue sat in the normal channel for two days because the person who received it assumed the sender had escalated elsewhere. The message met the urgent test and was not escalated, which is the failure mode the definition does not prevent.

The team added one sentence to the definition: if a message meets the test, moving it to the urgent channel is the recipient’s job as well as the sender’s. It has been used twice since, both times correctly.

Thirty-four percent of replies once landed after 20:00. It is now 6 percent, and the difference is a boundary that was made real by turning off a notification rather than by asking people to be disciplined.