Two Measurement Frames: Dunning vs Cancel Flows

Are you looking at a dunning trend or a cancel flow trend?

Recovery and churn come from two systems with different measurement properties. Dunning recovery is lagging and cohort-based. Cancel flows are real-time and definitive-session-based.

Dunning recovery

The unit is the campaign: the recovery sequence triggered when a subscription payment fails. It begins with the failure and ends in one of the four outcomes, or exits as a voided campaign when the underlying charge no longer exists.

A campaign runs for days or weeks. At any given moment, a meaningful share of campaigns are still active. Outcomes resolve over time, not when the campaign starts.

• Cohort math. Group campaigns by start date and wait for the cohort to complete before publishing a rate. See Daily Cohorts and Complete Cohorts.
• Rolling windows for performance analysis. Calendar boundaries cherry-pick variance; rolling smooths it. See Rolling Analysis.
• An explicit in-progress state. Active campaigns aren't lost or recovered, and the rate doesn't count them until they resolve. See The Four Outcomes.

The metric is recovery rate. Recovery Rate: the Formula defines it.

Cancel flows

A cancel flow is a real-time, session-based system that intercepts cancellation attempts and presents save offers. We measure it on definitive sessions, which settle within a 24 hour window, with no cohort waiting period.

A session is a single instance of a customer entering the cancel flow. One subscription's sessions within a 24 hour window collapse into one definitive session, which is the unit of measurement for cancel flows, analogous to a campaign in dunning.

A customer clicks cancel, moves through the flow, and either cancels or doesn't. The definitive session resolves within the 24 hour window: cancellation beats an offer, and an offer beats abandonment. There's no cohort lag, no in-progress state, and no complete-cohort waiting period.

• Outcomes settle quickly. Every definitive session resolves within the 24 hour window.
• No cohort waiting period. A definitive session from yesterday is settled today.
• Rolling windows still help. They're useful for smoothing day to day volume noise, not for waiting on campaign cohorts.

The three cancel flow outcomes

Each definitive session reaches one of three primary states.

• Cancelled. A definitive session where the customer completed the cancellation through the flow. Counts as a loss in the save rate denominator.
• Saved by offer. A definitive session where the customer accepted a save offer (pause, swap, discount, content) and didn't cancel through the flow in the 24 hour window. Counts as Saved.
• Abandoned. A definitive session where the customer left without cancelling or accepting a save offer, and didn't cancel through the flow in the 24 hour window. The subscription stays active, so it counts as Saved, in the numerator and the denominator.

Save Rate = Saved / (Saved + Cancelled)

• Saved = definitive sessions that didn't end in cancellation: accepted offers plus abandons
• Cancelled = definitive sessions where the customer cancelled through the flow

Save Rate = Saved / (Saved + Cancelled). We set nothing aside: every definitive session is in the denominator. The full treatment is on Save Rate.

"Cancellation" means two things

Same word, two denominators.

• Dunning. The customer actively cancelled the subscription while a recovery campaign was running. Counts as Lost in the recovery rate denominator.
• Cancel flows. A definitive session where the customer completed the cancellation through the flow. Counts as a loss in the save rate denominator.

Side by side

Why it matters when you evaluate performance

"Is churn going up?" routes to a different metric depending on which system is responsible.

• Dunning. Is the recovery rate trending down? Calculate it on completed cohorts and evaluate it on rolling windows. The number is sensitive to in-progress exclusion.
• Cancel flows. Is the save rate trending down, or is cancel volume up? Calculate both on definitive sessions and evaluate them on rolling counts to smooth daily variance.

A change to one system won't appear in the other system's data. That's why you judge a cancel flow's performance by save rate, not recovery rate, and why the complete-cohort rule doesn't apply to definitive sessions. Dunning vocabulary (in-progress exclusion, cohort completion) doesn't apply to cancel flow numbers, and cancel flow assumptions (24 hour definitive-session settlement) don't apply to dunning.

When both systems change at once

When a new dunning setup and a new cancel flow go live together, separate the two questions before you read anything. A drop in recovery rate is a dunning question. A drop in save rate or a spike in cancel volume is a cancel flow question. Reading both through one "recovery system" lens produces confident wrong conclusions.

Don't measure it this way

• Applying the complete-cohort wait to cancel flow sessions. A definitive session settles within 24 hours.
• Reading a dunning rate as if outcomes were immediate. Recent campaigns may still be active; wait for the cohort to complete.
• Leaving abandoned definitive sessions outside the save rate denominator. They count as Saved because the subscription stays active.
• Comparing a recovery rate to a save rate as if they measured the same thing. Different units, different outcomes, different denominators.
• Reading a cancel flow effect in dunning data, or a dunning effect in cancel flow data.

The dunning-side list is on Practices This Methodology Rejects.

---

Prerequisite: The Four Outcomes, which defines the dunning outcomes this page contrasts with the cancel flow ones. Next: Save Rate.