The argument
The number the retention industry reports is wrong
Every retention product counts an accepted offer as a save. That number has always included the people who were never leaving, and the better your product is, the more of them there are.
1. The industry counts acceptances and calls them saves
Open any retention product and find the headline number. It will be some version of revenue saved, offers accepted, or customers retained. In every case it is built the same way: a subscriber reached the cancel screen, was offered something, took it, and did not cancel. That subscriber is counted as a save, and the value of their subscription is added to a total the tool reports back to you.
The arithmetic is simple and it is wrong in one specific way. It assumes that every subscriber who accepted an offer would otherwise have left. Some of them would have. Some of them were going to stay regardless — they clicked into the flow to check something, or to see what was on offer, or because cancelling is how people ask for a discount. The tool cannot tell those two groups apart, so it counts them together and bills the difference to your margin.
2. The error is not small, and it runs one way
If the mistake were random it would not matter much. It is not random: it is systematically inflationary. Every subscriber who was staying anyway and took a discount is counted twice over — once as revenue you kept, and never as margin you spent for nothing.
It also compounds with how well your product works. The stickier your product, the higher the share of people who reach a cancel screen and were never truly leaving, and so the larger the fraction of your reported saves that are not saves. The better your retention genuinely is, the more your retention tool overstates its own contribution.
3. There is a standard answer, and the category has not adopted it
The method for separating the two groups is not novel or contested. You withhold the offer from a random slice of the same population and compare. This is how every serious question about cause and effect has been answered for a century, and it costs almost nothing: a small share of cancelling subscribers see no offer, and the difference between what happened to them and to everyone else is the part you caused.
The reason the category has not adopted it is not technical. It is that the number gets smaller. A vendor whose dashboard says you saved $84,000 has a better renewal conversation than one whose dashboard says $44,000 of that was you and the rest would have happened anyway. The incentive runs against the honest number, and it has run that way for a decade.
A hundred subscribers who reached the cancel screen
Offered an offer · 90
53 stayed · 37 left anyway
Held back · 10
1 stayed with no offer at all
Because 10% of the held-back group stayed without being offered anything, roughly 9 of those 53 were never leaving — and the discount they took is margin you can get back. That is what positions every offer on the chart above.
4. Meanwhile the cancellations moved
The second thing the category has not caught up with is where cancelling now happens. Retention tooling was designed around a button on a billing page, because for a long time that was where subscribers went. It is not any more. They message a support agent, which increasingly means an AI support agent with no offer to make and no authority to make one. Their card fails at three in the morning and nobody clicks anything at all. They cancel inside a native app on a platform you do not control.
A widget on a cancel page cannot see any of that, and the products that guard only that page report only on that page. The revenue leaves through doors nobody is standing at, and it does not show up as a retention problem because it does not show up in the retention tool.
5. What follows
Two things follow, and Tenure is both of them. Cover every surface a subscriber can leave through with one set of rules, so a decision made in a support chat knows what was already offered by email. And measure every offer against a group that was shown nothing, so the number you report is the part you caused.
The second is the uncomfortable one. It means our own dashboard will sometimes tell a customer that an offer they like is not earning its cost, and occasionally that a campaign did nothing at all. We would rather build the product that says so. A retention number you cannot defend to your CFO is not an asset — it is a liability with a nice chart on it.
Where this comes from
- The arithmetic in section 3 is the standard two-group comparison. Our exact method — assignment, exclusions, and the interval we report — is written out on the trust page.
- The claims about what competing products report are taken from their own public marketing, and both comparisons say where each of them is ahead of us: vs Churnkey and vs Sirius.
- No customer results are cited here because we have none to cite. The figures in the illustration are a worked example and are labelled as one.
If this is right, it is worth ten minutes of your own data.
Give Tenure your domain and it writes the flow. Run it against a group shown nothing, and find out which of your saves were yours.