What Moves a Recovery Rate

This page is about dunning recovery. Cancel flow save rates have their own drivers, offer design and flow UX among them, and are out of scope here; see Two Measurement Frames.

Most movement in a recovery rate has nothing to do with the recovery process. Before working on the process, rule out the things that move the number on their own.

Part 1: What moves the number without any process change

Sample size

Fewer failed payments means a wider swing from one window to the next. The population that matters is failing subscribers, not subscribers, so a small one swings even at a large company.

Seasonality

Holidays, promotions, and surges in new subscribers show up in recovery months later. From late November through December, more cards fail and subscribers are harder to reach. In January, subscribers acquired over the holidays hit their first renewal and decide whether to stay.

Price points

Customers on lower prices can be more likely to churn. In a dollar-weighted rate, a few high value failures move the number on their own, whichever way they resolve.

Billing cycles

Annual and quarterly renewals fail and churn differently from monthly ones. They renew less often and decline more often when they do, so a period with more of them in it looks different with no change to the process.

Customer acquisition

Deep discount and unintentional signups churn more. They never intended to stay; the failed payment is where it shows up.

Service quality

Product, support, and onboarding set how much customers want to stay. Improve them and churn falls and recovery rises, with no change to a single email or retry.

Customer mix, and why recovery rate is partly a loyalty measurement

Two outcomes depend on a customer choosing to act. A Card Update is a customer deciding to stay. A Cancellation is a customer deciding to leave. Passive Churn is a customer who never acted on the request, whether they ignored it or never saw it. A Successful Retry can also resolve without the customer doing anything. So the forces that drive voluntary churn drive recovery rate too, and each period's rate carries the mix of customers who failed in it.

Two similar coffee subscription companies run an identical recovery process. One recovers 80%, the other 20%. The 80% company's customers opted in knowingly and like the product. The 20% company's customers didn't realize they'd subscribed, or took a steep discount planning to cancel before renewal. The lower one thinks it has a payments problem. It has a loyalty problem showing up as a payments number.

Segmentation: telling a process change from a mix change

Segmentation is how you narrow that variance into something actionable. Same data, sharper question.

Separate customers on their first renewal from everyone else. Separate a recent signup surge from the base. Look at long tenured customers alone. Much of the noise leaves. If the blended rate moved and the long tenured segment didn't, the mix moved. If every segment moved together, something changed in the process or in the business. The method is on Natural Variance.

Processing anomalies

A spike in one decline code, or a batch of old charges reprocessed weeks after they first ran, can move the number with no customer behavior behind it. A processor returning a wave of vague declines from a temporary issue looks like a failure spike. Reprocessed old charges flood the funnel with failures that were never likely to recover. Check the decline distribution for the period before changing strategy.

Part 2: The levers that do move the number

The levers that move a recovery rate: deliverability, decline-code handling, campaign length, and the card-update experience.

Deliverability

If the email doesn't reach the inbox, nothing downstream matters. The card update email is the one that brings a failing customer back, so its inbox placement caps how much is recoverable. Sender authentication, a sending reputation kept separate from marketing email, and not emailing dead cards indefinitely all protect it.

Decline-code handling

Soft declines get retries before anyone contacts the customer. Hard declines get few retries or none, and the customer hears from you sooner. Routing by code sends effort where it can work and keeps futile retries from hurting your standing with the card networks. Definitions are on Failure Reasons.

Campaign length

How long a campaign runs before it ends in Passive Churn sets how much of the long tail you recover and when your cohorts complete. Too short misses recoveries that happen late in the window. Too long keeps emailing and charging cards that will never recover, which hurts your reputation with the card issuing banks, raises the risk of chargebacks, and lowers engagement on your emails, dragging down deliverability for the ones that would convert. Returns diminish the longer the window runs. The full treatment is on Campaign Length and Churn Recognition.

The card-update experience

Friction here turns recoverable failures into cancellations. The update should save the new card to the subscription, so the same customer doesn't fail again next cycle on the old one. Digital products such as SaaS and online memberships usually capture the failed payment right away. Subscriptions for physical goods may delay the charge until the customer adjusts the order, the fulfillment window opens, or the next scheduled charge attempt runs.

Part 3: What can't be measured

Retry timing

There is no way to know whether a retry at a specific minute performed better than the same retry an hour later. The alternate version of that charge never ran, so there is no counterfactual to measure it against. Every other number in this methodology traces to a row in a table; this one doesn't.

Retry timing has diminishing returns as an optimization target. Attention spent there is attention not spent on the four levers above. No system, adaptive or otherwise, and no volume of data changes this, because more data about the retries that ran says nothing about the retries that didn't.

The size of the prize is small even in principle. Start from all failed payments. Narrow to insufficient funds declines, the usual target for a specific retry time. Then look for a marginal gain beyond what a sound retry strategy already recovers. You're looking for a percent of a percent that can't be validated, and it's unknown whether the timed retries are harmful or helpful as a strategy.

Copy, once you're inside the band

Transactional messaging has a narrow band of what good looks like. There are many ways to write it badly and few meaningfully different ways to write it well.

When a messaging A/B test runs long enough on enough volume and shows no separation, the account is already optimized there. Some tests show a lift, and others stay even. An even result on enough sample means the messaging is at its ceiling.

A flat A/B result on transactional messaging is a finding, not a failure. The messaging isn't hurting you, and it isn't the lever either. The move is to keep the messaging clear, on brand, and recognizably from you, so customers trust it enough to act, and to spend the attention on the four levers.

The general principle

Churn prevention has diminishing measurable returns. Once the four levers are handled and a copy test comes back flat, the account is close to its ceiling. Say so, rather than implying there's always more to tune.

What remains is holding the ceiling. That means avoiding the same mistakes on repeat, protecting the downside, and adapting as the business changes, with new plans, new pricing, and new billing models. That work is not about a measurable lift, and yet its value grows as the business scales.

Don't measure it this way

• Attributing a swing to a process change before ruling out the six drivers. Check sample size, season, price mix, billing cycle mix, acquisition source, and service changes first. Then segment.
• Claiming a specific retry time outperformed another. The counterfactual doesn't exist. Report the retries that ran; don't rank the timings.
• Rerunning copy tests after a flat result on sufficient volume. The flat result is the finding. Move to the four levers.
• Mistaking a mix shift for a performance change. A signup surge or a wave of first renewals moves the blended rate on its own. Segment before you conclude.
• Reporting a retry-to-card-update split as a fixed ratio. The split varies by account and by period. Report the four outcomes you counted for the period, not a ratio.

The full list of counter-rules is on Practices This Methodology Rejects.

---

Prerequisites: Natural Variance, which measures the swing this page explains; Recovery Campaigns, which defines the structure the four levers act on.

Next: Two Measurement Frames, the first page in Cancel Flows. This is the last page in How Dunning Works.