Share:

Re-charge list — from 2026-09-23

5,026 people whose latest recurring payment failed and has not been collected since. Same amount, no reduction. Built 2026-09-19 08:00 UTC.

We have never done this before. Nothing in this business re-charges a failed payment automatically — every second attempt on record is a debt collector on the phone, or a customer paying a reminder link we sent. So this run is a new capability with no baseline, and no number from our history predicts what it will collect. In particular: a same-amount second attempt converts at 92.5% in our data, and that figure is a collector ringing someone who then moved money across. It is not what a silent re-charge will do. Expected recovery here is UNKNOWN, and the point of the control arm below is to find out.
People
5,026
latest charge failed, not collected since
Value of those charges
$1,014,092
$202 average
To charge
4,491
$907,135
Held back as control
535
10% — charge nothing, measure them anyway
Charles's batch — 120 people, $25,402

20 from each of the six cohorts, so one person can work the whole thing in a sitting. Picked by a stable hash, not by biggest debt — taking the largest balances first would make the batch unrepresentative of the list it came from, and the point of a first run is to find out what a re-charge does to a normal person on the list. The same 20 come out on every rebuild, so nobody can be charged twice by a re-run. Control rows are excluded, so everything in this file is meant to be charged.

At 120 people this is an operational pilot, not a measurement. It will tell you whether the process works and roughly what happens; it is far too small to compare cohorts against each other. Judge it on "did it run cleanly and did money arrive", then scale.

Download the batch — 120 people

The six lists

The full size of the opportunity. Charles works the batch above; these are what is behind it. Everyone here failed on a soft decline — a card that is dead or blocked (expired, closed, wrong number, issuer-blocked) is excluded outright, because no re-charge at any amount can work on one. Benefits cards are excluded too: Direct Express pays on the 1st/3rd and on a Wednesday set by birth date, so a Wednesday run means something different for that population.

Card classGroupPeopleValueTo chargeControlAvg chargeFile
Early-access missed first payment 201 4.0% $38,744 182 19 $193 CSV
Early-access never paid anything 543 10.8% $104,156 475 68 $192 CSV
Early-access on/off payer 518 10.3% $89,607 466 52 $173 CSV
Normal bank missed first payment 474 9.4% $104,764 428 46 $221 CSV
Normal bank never paid anything 823 16.4% $188,048 742 81 $228 CSV
Normal bank on/off payer 2,467 49.1% $488,771 2,198 269 $198 CSV

Download all 5,026 as one CSV — the Arm column marks the control rows, and every grouping on this page is also a column.

Why they failed

Soft reasons only, so every one of these is a card the issuer might still say yes to.

CodeWhat it meansPeople
TRANSACTION_LIMIT Issuer says the amount does not fit — in practice the card is at its limit 2,311 46.0%
GENERIC_DECLINE Bank declined and gave no reason 2,202 43.8%
INSUFFICIENT_FUNDS Not enough money in the account 513 10.2%

Read the two halves separately

2041 of these people have never paid a penny on their plan, and 2985 have paid before and stopped. Those two will not behave the same, and a single blended recovery rate would hide it. Judge them apart.

The control arm is what makes this worth doing twice. 10% of every group is held back and charged nothing. Without it, whatever the run collects cannot be separated from what those people would have paid anyway — and we would end up quoting a number as a lever that is nothing of the sort, which is the mistake this whole area keeps producing. The hold-back is fixed per person, so rebuilding this page never moves anyone between arms.