A dormant subscriber is a complaint we have not yet received
Digital publishers celebrate low churn without asking how many paying readers have stopped reading. A subscription that survives because cancelling is forgotten or difficult is not loyalty, and boards should stop counting it as such.
Every digital publisher I know reports churn. Most report it as a single figure, the share of paying subscribers who cancelled in a month or a year, and most boards treat a falling figure as good news. I have presented that figure to our own board, and I now think it hides the most important fact about a subscription business: how many of the people paying are still reading.
At Vesterlund Mediegruppe the answer, when we first measured it carefully in the spring of 2023, was uncomfortable. About 12 per cent of our paying digital subscribers had not opened an article, a newsletter or an e-paper edition in ninety days. They were not cancelling. They were paying by card each month, often for a subscription taken out during an election or a local story that had mattered to them, and they had stopped thinking about it. On our dashboards they looked like loyal customers. In practice they were complaints we had not yet received.
What the churn model was really predicting
Our churn model was competent by the standards of the industry. It used reading frequency, payment history, plan type, tenure and a few dozen other signals to estimate the probability that a subscriber would cancel in the next ninety days, and the retention team used its scores to decide whom to contact with offers.
The model had learnt something true and unhelpful. Inactive subscribers rarely cancel within ninety days, because cancelling requires remembering the subscription exists. So the model rated them as low risk, and the retention team left them alone. The people it rated as high risk were mostly active readers unhappy about a price change or a particular story, who were therefore paying attention. The model was predicting who was alert, not who was getting value.
I had seen the same pattern twenty years earlier at a mobile operator, where a churn model sent retention discounts to customers who would have stayed anyway. The remedy there was to measure the effect of acting, not merely the risk of leaving, by holding back a random group from every campaign. We applied that method at Vesterlund. It showed that our offers to active, unhappy readers worked moderately well, and that ignoring silent readers carried a cost the model could not see. Silent subscribers did leave eventually, often after a bank statement reminded them, and when they left they were angrier, more likely to dispute the card payment and more likely to tell friends that our titles were hard to cancel.
There is a wider point about what a model optimises. A churn model trained only to minimise cancellations will, left alone, favour every kind of friction that delays them: longer cancellation journeys, renewal reminders that arrive late, offers designed to confuse. Nobody in a publishing house intends that. It is simply what the measure rewards. European consumer policy is also moving towards making cancellation as easy as signing up, and I would rather a company reached that position because it measured honestly than because it was told to.
Writing to the people who stopped reading
In May 2023 we began writing to every subscriber who had been silent for ninety days. The letter said plainly that they had not used their subscription and offered three choices, each one click from the email: pause for up to three months, move to a cheaper plan with fewer titles, or cancel. We also showed them what they had missed in their own area, chosen by the regional desk rather than by a model, because the best argument for staying is usually a story about one's own town.
The commercial case against this was strong, and the colleagues who made it were not being cynical. Our chief financial officer estimated a loss of up to DKK 30 million a year. Our commercial director pointed out that no competitor was doing anything similar, and that we would report slower growth while others reported faster. Both were right about the short term.
The full outcome is set out among my cases on this site. In brief, of 41,000 silent subscribers about a quarter cancelled, about a seventh paused or downgraded, and about a third began reading again. The net loss was about DKK 23 million of annual revenue. Card chargebacks fell by 60 per cent. Churn among new subscribers who joined after the change also fell, from 31 to 24 per cent in their first year, which surprised the people who had argued against the programme and, honestly, surprised me. Our best explanation is that an easy exit makes people more willing to come in. Readers tell us they subscribe more readily when they are confident they can leave.
What a board should count instead
I do not suggest that boards stop looking at churn. I suggest they never look at it alone. Three further measures have changed how our board discusses subscriptions, and I think they would help any publisher, streaming service or other subscription business.
The first is the silent share: paying subscribers who have not used the product in a defined period, ninety days in our case. It belongs beside the subscriber total every quarter, with its trend. If the total rises while the silent share also rises, the growth is weaker than it looks.
The second is the measured effect of retention activity, not its volume. A board should know how many subscribers stayed because of a retention action, judged against a random group that received none. A retention team that cannot show this is spending money on people who would have stayed anyway, or keeping people in ways it would not want described.
The third is the cancellation experience itself: how many steps it takes, how long it takes, how often customers dispute payments and how many complaints mention it. These are not customer-service details. They are evidence of whether revenue depends on attention or on inertia.
Silent subscribers are now 6 per cent of our base, and dormancy letters go out automatically after ninety days. Our subscriber count grows more slowly than it could. I would rather explain a slower figure to our board than a sudden fall to readers who have discovered, all at once, what they were paying for.