Skip to content
allanninal.dev/build
Collaborate

Part 5 of 7 · Win-back campaigner series ~5 min read

How a win-back is honestly counted

A win-back campaign will always report a recovery rate that looks fine, because some share of lapsed customers return regardless and every one of them lands in the numerator if nobody arranges otherwise.

Key takeaways

  • Some lapsed customers return on their own. Hold back a share and find out how many.
  • Report margin recovered, not revenue recovered. They tell different stories.
  • The attribution window should be one buying interval, not a fixed thirty days.
  • Count the replies as an output in their own right.
  • A campaign whose result is ’seventy open complaints’ has still been worth running.

Revenue is the wrong number

How win-back results are computed from raw orders to net marginA vertical chain of five steps entered by a box labelled A campaign ends, four hundred and fourteen messaged. Step one counts orders in the window, say forty-seven. Step two subtracts the holdout rate, where seven of forty-one returned anyway, drawing on a side box labelled Holdout, ten per cent, never messaged. Step three computes incremental orders as forty-seven minus seventeen per cent of four hundred and fourteen, giving about twenty-four. Step four multiplies by margin rather than revenue. Step five subtracts what was given away in discounts and delivery. A note says half the headline orders were people who were coming back regardless.AWS ACCOUNTA campaign ends414 messagedOrders in the windowsay 47Minus the holdout rate7 of 41 returned anywayHoldout10%, never messagedIncremental orders47 - 17% of 414 = ~24Times marginnot revenueMinus what was givendiscounts, deliveryHalf the headline orders were people who were coming back regardless.
Fig 1. How a recovery number becomes a real one. Each step removes something the raw count was quietly claiming credit for.
  • App integration
  • Security & identity
  • Analytics

Margin, not revenue

A win-back that recovers eleven thousand pounds of revenue at fifteen per cent off has recovered considerably less margin than the headline suggests, and on a low-margin product line it can recover almost none. Reporting revenue makes discount-led campaigns look best; reporting margin makes them look like what they are.

The change is trivial — one column — and it reliably changes how the next campaign is designed, which is a good return on a column.

The window is theirs too

A thirty-day attribution window is the default and it has the same flaw as a ninety-day lapse rule: it fits one buying frequency. The window should be one of that customer’s own intervals, so a monthly buyer is counted over a month and an annual buyer over a year.

That does mean the annual segment’s results are not known for a year, which is genuinely inconvenient and is the truth. Reporting them at thirty days would produce a number that means nothing, faster.

What a quarter looks like

One quarter of win-back activity reported honestlyA horizontal row of five boxes. Lapsed: six hundred and twelve. Suppressed: one hundred and ninety-eight, including seventy open complaints. Messaged: three hundred and seventy-three, with a holdout of forty-one. Incremental: about twenty-four orders. Net margin: about one thousand nine hundred pounds. A note says seventy open complaints were found, which is arguably the more valuable output.ONE QUARTER, HONESTLYLapsed612Suppressed198, incl. 70 complaintsMessaged373, holdout 41Incremental~24 ordersNet marginabout £1,900And 70 open complaints found, which is arguably the more valuable output.
Fig 2. A quarter reported end to end. The suppression count carries a finding that the campaign was not looking for and that outranks the revenue.
  • App integration
  • Security & identity
  • Analytics

Nineteen hundred pounds of net margin in a quarter is a modest, real result from a system that costs two dollars a month to run, and it is worth reporting as modest rather than inflating it to the eleven thousand of gross recovered revenue that the same quarter could also be described as.

The finding that was not the goal

Seventy unresolved complaints surfacing from a suppression check is the kind of result that justifies the whole exercise on its own. Those customers were lost for a reason somebody wrote down and nobody acted on, and the campaign is the first process that looked.

It is worth putting that in the report explicitly rather than leaving it as a suppression count nobody reads. “Seventy customers stopped buying after a complaint we have not resolved” is a sentence that produces action; “suppressed: 198” is not.

The pressure that arrives at month four

Three requests to expect, and the answers

  • “Drop the holdout, we know it works.” Knowing it still works next year requires it to keep running. Ten per cent is a cheap insurance premium.
  • “Send a second message to non-openers.” Converts slightly, complains considerably, and it is the same argument as the second abandonment nudge.
  • “Lower the lapse multiplier to get more volume.” That contacts people who are merely a bit late, and the response rate falls faster than the volume rises.
  • “Just discount everyone, it is simpler.” It is, and the margin column shows what it costs.
  • All four are reasonable-sounding and all four are answered by numbers the report already carries, which is the point of carrying them.

Next: what all of this costs to run.

All posts