The customers you are losing are still ordering.

SECTION 01The question

Every owner wants the same list: the customers who have quietly stopped coming, while there is still time to do something about it.

Almost everyone builds it the same way. Anyone with no order in ninety days goes on the list. Sometimes sixty, sometimes a hundred and twenty, but always one number of days.

We built that list on a real business, then built two better ones, and graded all three against what the customers actually did next. The answer changed three times, and the last change was the one that mattered.

SECTION 02One real business, one morning in June

The University of California, Irvine publishes the complete transaction file of a real UK online retailer that mainly sells gift-ware, many of its customers wholesalers: every line of every invoice from December 2009 to December 2011, 1,067,371 lines from 5,853 named customers.[1] We used the file before, to measure how far its sales move on their own.

This time we pretended it was the morning of 1 June 2011. We let ourselves see only the orders placed before that day and built each list from them. Then we opened the next 180 days and checked every name.

Its active customers that morning, anyone who had ordered in the year before, came to 4,324. Of those, 1,880, or 43.48%, placed no order at all in the next 180 days. That is the bar every list has to beat. A list that finds gone customers at that rate is no better than picking names out of a hat.

SECTION 03The usual list

No order in ninety days. On 1 June 2011 that rule put 2,367 customers on the list, 54.74% of everyone active. More than half the customer base, flagged as leaving.

Over the next 180 days, 59.02% of them placed no order. Better than the hat, but not by as much as the word "lost" suggests. The other 970, 40.98%, ordered again, with nobody chasing them.

Look at who is on it. 1,414 of the 2,367 names, 59.74%, had only ever placed one or two orders. Their typical spend in the year before was £303. Most of the list is not customers drifting away. It is people who bought once or twice and never had a habit to break.

The fixed cutoff also hides how smooth the real picture is. The chance a customer is gone climbs steadily with the days since their last order. There is no cliff at ninety.

Even customers who had ordered in the last thirty days placed no order in the next 180 days 16.59% of the time. And at the far end, a whole six months to a year of silence, about a third still came back.

SECTION 04Late, by their own clock

Ninety days means different things to different customers. A wholesaler who orders every week and has gone quiet for a month is very late. One who orders twice a year is right on time.

So the second list measures each regular against their own rhythm. For the 2,375 customers with three or more orders, take the typical gap between their orders, the middle one, and count how many of those gaps have passed since the last order.

The pattern is clean. On 1 June 2011, regulars still within one usual gap were gone 17.87% of the time. Past five of their own gaps, 54.79%. The same climb shows up from the two earlier mornings we tried, 1 December 2010 (24.26% to 61.36%) and 1 March 2011 (22.93% to 57.56%).

That makes a better ranking: the customers at the top of it are the ones furthest off their own pattern. But it is still a list of people who have gone silent. And it turns out silence is not where most of the money goes.

SECTION 05Then count the money

Both lists answer "who stopped?" The question that matters is "where did the money go?", and it has a different answer.

Take the twelve months from December 2009 to November 2010 and the same twelve months a year later. Sales to named customers went from £8,125,231 to £8,250,425, up 1.54%. A quiet, steady year. Then split every customer by what they did.

Under that steady year, £2,965,202 left customers the business already had. 1,531 customers stopped ordering altogether and took £1,028,872 with them. That is the money every lost-customer list is built to find.

The bigger flow is the next bar. 1,524 customers kept ordering and spent £1,936,330 less. They never went quiet. They would never appear on a ninety-day list, or a list of regulars running late, because they were still placing orders.

65.30% of the money lost from existing customers left with customers who never stopped ordering.

New customers and growing ones covered all of it, which is why the year looked calm. That is also why nobody would go looking. The same split holds across each half of the year, compared a year apart.

SECTION 06Is a shrink real?

A fair objection: customers have good years and bad years. Some of that shrinking is just a customer who had a big order last year and a normal one this year. It would come back on its own.

So we tested it the way we test anything, against the ordinary case. On 1 June 2011, take every customer who had ordered in the last ninety days and had spent at least £500 in the same six months a year earlier. Call them shrunk if their last six months came to under half of that, and steady if not. Then watch the next six months, each against the same months a year earlier.

There were 179 shrunk customers. A year earlier they had spent £637,026 in those six months; this time, £203,814. The 573 steady customers were the comparison.

44.69% of the shrunk customers were still under half over the next six months, against 23.91% of the steady ones. And 21.79% of them stopped ordering altogether, against 8.90%. A shrink is the step before a stop, and it shows up while the customer is still buying.

Then we moved the morning. Five more start dates, a month apart, with four-month windows. Across all six, the shrunk group stayed down more often (34.83% to 44.69%, against 19.70% to 29.22%) and stopped more often (19.35% to 27.08%, against 8.90% to 16.63%).

The objection is half right, and the half it gets right is useful. The shrunk group's spending was down 68.01% on the year before; over the next six months it was down 22.82%. A lot of the drop did come back. What did not come back is the part worth a phone call, and it is concentrated in a list of 179 names rather than 2,367.

SECTION 07Three lists, and what each costs

Each of these is a real door, and each is built from the same three columns of a sales file: the date, the customer and the amount.

The silence list. The easiest to build, and the biggest: 2,367 names here. It finds the customers who are fully gone, and it costs attention spread thin, because four in ten of its names come back on their own and an offer sent to all of them pays some people who were already coming.

The late-regulars list. A ranking instead of a cutoff, so the work starts at the top. It costs a little more arithmetic, and it still only sees customers who have gone quiet.

The shrinking list. 179 names on 1 June 2011, every one of them still a customer, and it watches the kind of customer most of the lost money left with. It needs a year of history for the same-season comparison, and it needs a conversation rather than a coupon, because these customers did not forget you. They moved some of their buying somewhere else, for a reason only they can tell you.

Which one fits depends on what you know about your customers that no file can show. Your call.

SECTION 08The point

Nothing in this piece is exotic. Every list came from a date, a customer and an amount, which every sales file in the world already carries. What it took was building each list as of a real morning, grading it against what actually happened, and then counting the money instead of the names.

A list of who stopped buying can only ever see the smaller part of what is walking out.

A data team builds these lists every month and grades last month's against what the customers did, which is how you learn which list your own business can trust. The same file, read for a different question, showed how long it takes to tell whether a change worked.