The retention discount line

1. 018%Derived in this reportCopy quote

   At a 20% discount held for three months on a 70% margin subscription with a nine-month remaining lifetime, 8% of accepters must be genuine saves before the offer breaks even.

   The margin given away is small and immediate; the margin recovered is larger but only exists for the fraction of accepters who were actually leaving. The break-even share is the ratio between the two, and it is far lower than most teams assume, which is why shallow, short offers survive scrutiny more often than deep ones.

   Do thisCompute your own break-even share from price, discount depth, discounted months and expected remaining lifetime before the next save campaign ships.

2. 028% to 33%Derived in this reportCopy quote

   Depth punishes the offer faster than duration does: moving from 20% to 50% off, holding everything else, raises the break-even incremental share from 8% to 33%, a four-fold increase in how right the targeting has to be.

   A 50% offer has to be right about one accepter in three. Almost no untargeted save flow is. This is the single most useful line in the model: if a deep discount is on the table, the targeting has to be far better than the discount is deep.

   Do thisIf a deep discount is on the table, improve the targeting first. Going from 20% to 50% asks the model to be four times more right.

3. 03unchangedDerived in this reportCopy quote

   Margin, not price, decides how forgiving a discount is. At 85% gross margin the break-even share for a 20% offer is 8%; at 50% margin it is unchanged.

   Margin cancels out of the ratio when the discount is a percentage of price, because it scales both the giveaway and the recovery. What margin does change is the absolute size of the prize, and therefore how much analytical effort the decision deserves. Low-margin businesses should discount less because they have less to recover, not because the break-even moves.

   Do thisUse margin to decide how much analysis the decision deserves, not to set a different threshold.

4. 04Independent researchCopy quote

   The failure mode is structural, not analytical: cancellation-flow offers are shown to people who reached the cancel button, a population enriched in customers who are leaving and in customers testing whether a discount appears.

   Selection into the offer is not random, so the acceptance rate carries no information about incrementality. Independent research on incrementality testing finds a large share of active next-best-action programmes deliver marginal or negative lift once measured against a randomised control, and save offers are the archetype.

   Do thisStop reporting acceptance rate as a save rate. Move the offer out of the cancellation flow and onto a risk score.

5. 05Anonymised Markin deploymentsCopy quote

   Across Markin deployments where a save offer was moved from an untargeted cancellation flow to a risk-scored decision with a randomised holdout, the incremental share of accepters rose enough to clear break-even in most, but not all, cases.

   We are deliberately not putting a single number on this. The sample is small, the bases are very different, and a mean across them would be a rounding of something rather than a finding. What travels is the mechanism: the decision moves from 'who clicked cancel' to 'who is likely to leave and likely to respond', and the second population is smaller and cheaper to serve.

   Do thisHold out 10% of eligible customers for the full window, and compare the measured incremental share against your break-even line.

• Head of Retention

  Set the break-even share before the next save campaign ships, and treat it as the acceptance threshold. If the measured incremental share sits below it, cut depth before cutting volume: a 20% offer needs the targeting to be right one time in twelve, a 50% offer one time in three.

• Finance

  Ask for the break-even share alongside every save-offer business case. It is four inputs and one division, so its absence is a choice. Retained revenue reported without it is a gross number that includes the customers who were never leaving.

• Data and analytics

  Hold out 10% of eligible customers for the full discounted window plus the expected remaining lifetime. A holdout closed at the end of the discount measures acceptance, not retention, and will overstate the save rate.

Incremental save

A customer who accepted the offer and would have cancelled without it. Not a customer who accepted and would have stayed regardless. The two are indistinguishable without a randomised holdout, which is the whole difficulty.

Break-even incremental share

The minimum proportion of accepters that must be incremental saves for the discount to return more margin than it gives away, over the remaining lifetime.

Remaining lifetime

Expected months the saved customer keeps paying after the save, at the discounted price. Deliberately conservative here: a save that lasts one billing cycle is not a save.

Gross margin

Contribution margin on the subscription revenue, after variable delivery cost. Content, licensing and network costs belong here. Sales and marketing do not.

There is no sampling error here, because there is no sample: this is arithmetic. The uncertainty is entirely in the inputs, so every figure is published with the range it takes across the plausible input grid rather than as a point.

• 8% break-even at a 20% discount

  Holding d = 20% and k = 3 months, across remaining lifetimes L of 6 to 15 months. The point estimate uses L = 9.

  5.0% to 12.5%

• 33% break-even at a 50% discount

  Same grid. At L = 6 the offer needs half of all accepters to be genuine saves, which no untargeted cancellation flow achieves.

  20.0% to 50.0%

• Effect of discounted months

  At d = 20%, L = 9. Duration moves the threshold linearly; depth moves it faster than linearly.

  2.8% at k = 1 to 16.7% at k = 6

• Effect of gross margin

  Exactly zero, not approximately. Margin cancels in s* = G / R. It scales the size of the prize, never the threshold.

  zero

• This is arithmetic, not an empirical finding. It tells you where the line is, not where your programme sits relative to it. Only a randomised holdout does that.
• It ignores second-order effects in both directions: price-anchoring damage from repeated discounting, and referral or usage value from a saved customer.
• It assumes the discount is the only thing that changed. Where a save offer is bundled with a service fix or a plan change, the attribution is no longer clean.
• The plausible input ranges come from consumer subscription businesses. Contract telecom, insurance and utilities have longer lifetimes and different answers.

• When is a retention discount worth sending?
• How do I know if my save offer is actually working?
• What percentage of save offers go to customers who would have stayed?
• How deep should a retention discount be?
• How do I measure the incrementality of a cancellation offer?

• PlaybookRun a retention offer before renewal without paying customers who were staying anywayPre-renewal save offers on Stripe, sized by risk instead of by list membership.
• ResearchARPU and churn, measured consistentlyMost cross-sector ARPU comparisons are comparing denominators. Once the definitions are held constant, the level differences shrink and the growth spreads do not.
• ResearchRevenue that carries marginEvery growth decision on this list raises reported revenue. Four of them raise contribution margin. They are not the same four most teams prioritise.
• GuideHow Markin measures incrementalityThe holdout design, the measurement window and the stopping rules Markin uses before any result is allowed to scale. Written to be checked.
• GuideControlled growth experiments: proving an action caused the resultEligible population, randomised control, primary metric, observation window and guardrails, and the scale, refine or stop decision that follows the readout.