DEV Community

Swipe Again for Swipe Again

Posted on

A Calculator Is Only as Honest as Its Worst Input

Every failed-payment calculator promises a number, and the number is only ever as honest as the inputs the merchant feeds it. This is not a criticism of calculators — a rough, honest estimate beats the precise ignorance most subscription businesses operate under — but the inputs deserve scrutiny, because two of them are systematically wrong in the optimistic direction, and a leak underestimated is a leak deprioritized for another year.

The first suspect input is the failure rate itself. Merchants typically quote the figure their dashboard shows, which is usually failed transactions divided by transactions — a number that counts every retry as a fresh trial and dilutes the true rate. The figure a calculator actually needs is episodes: subscribers whose first renewal attempt failed, divided by subscribers who came up for renewal. Counted properly, the rate is commonly half again the dashboard figure, sometimes double. Garbage in is too kind a phrase; this is the right formula fed the wrong population.

The second suspect is the recovery assumption, and here honesty cuts the other way as often as not. Merchants who have never run structured recovery assume near-zero is recoverable and dismiss the estimate; merchants quoting a vendor's brochure assume the headline rate applies to their whole failed pool, including the hard declines no honest process prices as recoverable. The usable middle is a rate applied to the workable pool — soft and fixable failures, with dead accounts priced separately through a working card-update path. That is the estimation discipline behind a failed payment leak estimated from episodes, not transactions: the model's value is in its structure, not its arithmetic.

The third input is the one nobody puts in: time. A subscriber recovered in month one pays again in months two through twelve, and a leak calculation that counts only the first recovered payment understates the stakes by most of an order of magnitude. Conversely, a subscriber recovered sloppily — chased, annoyed, rescued at the third attempt — may pay once and cancel voluntarily, and counting a full year for that save overstates it. Honest models carry a tenure assumption and say so on the page.

Run the calculator twice, then: once with the flattering inputs, once with the audited ones. The gap between the two answers is itself a finding — it measures how much of the business's picture of its own billing health is decoration. The audited number is the one to take into the planning meeting, even when, especially when, it is the larger leak and the more uncomfortable conversation.

A final honesty test costs nothing and catches the most: write the inputs down next to the answer, every time the estimate is quoted. A leak figure travelling through a business sheds its assumptions at each retelling, arriving in the board deck as a bare number with the authority of an audited fact. Keep the episode definition, the workable-pool boundary, and the tenure assumption attached to it, and the number stays what it was — a model, clearly built, open to challenge on its stated terms. That transparency is not a weakness in the estimate. It is what makes the estimate usable by people who were not in the room when it was made, which is to say, by everyone who will ever act on it.

Models that travel with their assumptions earn challenges that improve them; bare numbers earn only belief or dismissal, neither of which makes the next estimate better.

Challenged on its stated terms, an estimate gets sharper every quarter it survives — which is the most any model in a planning meeting can honestly promise anyone.


Published by Swipe Again — failed payment recovery at a flat $179/mo.

Top comments (0)