On 28 October 2026, Etsy reported two things that might sound difficult to reconcile. The value of merchandise sold per active buyer had risen 3.9% compared with a year earlier, measured over the preceding twelve months. Yet buyers were still shopping slightly less often. Larger orders helped explain how both things could be true. A shopkeeper might recognise the situation immediately. The till has had a good month. The person who used to come in every Tuesday is now coming every other Tuesday. 1

A new business rarely has twelve months to work with. In the weeks after launch, the owner has an opening party, some encouraging sales and a fairly pressing question about what to spend next month. The temptation is to compare this week’s receipts with last week’s. The answer the owner needs is harder to see. Of the people who tried the place when it opened, how many have found a reason to use it again?

Which customers could have come back by now?

Suppose a neighbourhood lunch counter has been trading for eight weeks. The owner can recognise some customers through a voluntary digital receipt or an account they already use to order. She groups those identifiable buyers by the fortnight of their first paid lunch. Each group is a cohort, a technical name for people who started at roughly the same time. Each buyer gets one chance to enter the table.

She counts a return when someone buys lunch again on a different day, within 28 days of their own first purchase. Two sandwiches on the opening visit count as one buying occasion. Friends eating on the house sit outside the calculation. Anonymous till sales still belong in the accounts. They stay out of this table because the owner cannot reliably tell who made them.

Here are illustrative figures after eight weeks. A group earns a 28 day result only when every buyer in it has had the full 28 days.

Fortnight of first purchase Identifiable first buyers Buyers who returned within 28 days Share who returned
Weeks 1 and 2 80 20 25%
Weeks 3 and 4 60 21 35%
Weeks 5 and 6 72 Still observing Pending
Weeks 7 and 8 66 Still observing Pending

In the two mature groups together, 41 of 140 buyers came back, about 29%. The second group brought in fewer new customers and produced one more returning buyer. Its better percentage sounds more dramatic than the difference in the number of people. Writing down both helps the owner decide how much to read into a swing of ten percentage points. A zero in either of the bottom rows would send the opposite signal for an even worse reason. Time simply hasn’t passed.

Money can come back without many people

When DoorDash filed to go public in 2020, it grouped customers by the calendar year in which they placed their first order. It reported that the people who joined in 2018 collectively spent 65% more in their second calendar year than in their first. It was a statement about the value of orders from that group. It didn’t say how many individual customers placed another order. Someone who joined in December also had much less of the first calendar year available to spend than someone who joined in January. 2

A lunch counter can see the same distinction on a much smaller scale. Suppose that, of those 20 returning customers in its opening group, one office manager places a substantial order for training days. The kitchen might be delighted to win that business. Hiring another cook on the assumption that the neighbourhood has developed a weekly lunch habit would be a different bet. Look at how many people returned, how many visits they made and how much they spent. Each number answers a question that the others leave open.

The calendar should follow the product. A month gives a local worker several chances to buy lunch. A couple who hire a wedding photographer may never need the same service again. For that studio, an introduction that leads to a new booking may matter more than a second wedding. A seasonal hotel may sensibly wait until guests have had a chance to plan the following year’s holiday. Etsy chose its own definition for a habitual buyer in its 2025 annual report. The buyer had to shop on at least six different days and spend at least $200 over twelve months. That is a decision about the business Etsy operates. It would be an odd way to judge a lunch counter. 3

Go back to the counter

The owner’s next move is to ask who was actually in the first two groups. Perhaps the party drew visitors who lived across town. Perhaps the later customers came from the office above the shop. She can check the receipt totals to see whether the identifiable purchases cover a similar share of trade in both periods. A new email receipt offer in week three, for instance, could have changed which people she can see. She can also check when the kitchen ran out of the lunch that brought a regular back. A missing transaction records neither a customer’s intention nor a sold out dish.

A conversation can explain something a percentage leaves open. Ask a returning buyer what they did for lunch yesterday. Ask a willing occasional customer when they last tried to order and what happened. A request for encouragement tends to produce encouragement. A description of last Thursday can reveal that the menu went online after the office had already placed its order.

If that turns out to be the problem, try posting the menu before the office orders and follow the next two mature groups. Record what else changed, including promotions and closures. With numbers this small, a change in the return rate can suggest where to investigate. Whether the menu caused the change requires a closer comparison. In the meantime the owner has a perfectly good decision to make. Someone needs to publish tomorrow’s lunch while the customers nearby are still choosing where to eat.