Missed first payment

The population, the rate, and the money. The operational number to carry around is ~21 misses a working day — everything else on this tab is context for it.

Figure	Value	Period	Source	Read it this way
Agreements judged	27,737	Opened 2023-11-30 → 2026-09-16	`/opening-payment`, built 16 Sep 2026 09:02 from the Square recurring book.	An opening = a customer’s first charge on the RECURRING book, with no recurring charge from them in the preceding 6 months. 252,296 recurring charges resolve to these.
Miss the opening instalment	11,587 41.8%	Whole cohort, 2023-11 → 2026-09	`/opening-payment`, 16 Sep 2026.	This is the whole-history figure and it is flattered by 2023–early 2024, when misses ran 3–8%. Do not quote it as today’s rate.
The CURRENT miss rate	~47%	Aug 2026: 972 openings, 466 missed	`/opening-payment` month-by-month, 16 Sep 2026.	Aug 2026 = 357 never recovered 36.7% + 109 recovered ≤10d 11.2%. The rate has sat in the 31–42% never-recovered band every month since May 2024. `/first-payments` measures the same thing on a Jun–Aug window and gets 46.9% failed.
Misses per month	~455	Jun–Aug 2026 run rate	`/first-payments` (2,912 first instalments in three months) × the 46.9% failure rate.	≈ 21 per working day. This is the single most important operational number on the page: a first-miss call queue is one person’s morning, not a department.
Value of a first instalment	$221 avg / $209 median	Jun–Aug 2026	`/first-payments`, 16 Sep 2026.	Down −9.0% on last year ($243) and −8.3% on two years ago. So the miss is getting slightly cheaper per event while the count rises.
Banked on first instalments	$445,269	Jun–Aug 2026, all six accounts	`/first-payments`, 16 Sep 2026.	+9.0% on last year on +12.8% volume. Like-for-like it is −2.4% — Chicago and Nashville opened studios after Aug 2024, so part of the headline growth is estate, not demand.
Never recovered, whole cohort	9,252 33.4%	2023-11 → 2026-09	`/opening-payment`, 16 Sep 2026.	“Never recovered” here means nothing banked inside 10 days — it does not mean the agreement never paid again. 48.2% of missed openings eventually see money, however long it takes.
What a never-recovered opening banks	$243 by day 365	n = 5,232 with a full year behind them	`/opening-payment`, 16 Sep 2026.	Median $0. Against $1,327 for one that paid first time. The mean is carried by a minority who come good much later.
The 2×2 that predicts the book	Failed → failed = 26.2%	365 days from the opening charge, n = 16,743	`/instalment-recovery`, 16 Sep 2026.	Paid→paid $1,558 (44.1% of the book) · Paid→failed $807 (13.1%) · Failed→paid $994 (11.4%) · Failed→failed $184 (26.2%, median $0, 0.94 instalments ever banked). A quarter of the book is decided in the first two charges.
The SECOND instalment is the sharper signal	$1,050 gap	Same 15,885 agreements, same window	`/instalment-recovery`, 16 Sep 2026.	Splitting on the first instalment opens a $956 gap; splitting on the second opens $1,050. Only the split variable changes, so they are directly comparable. A call process should not stop at miss 1.

By Square account

Agreements opened 2023-12-01 onwards. A studio that opened recently has almost no 365-day runway, so read Chicago on the miss and 10-day columns only — those need no runway.

Account	Agreements opened	Opening charge missed	Recovered within 10 days	Read it this way
Chicago	1,561	734 47.0%	166 22.8%	Newest studio (first recurring charge 2025-08-14), so almost no 365-day runway — read it on the miss and 10-day columns only.
Nashville	2,639	1,167 44.2%	273 23.5%	Worst miss rate of the mature accounts, but the BEST 10-day recovery. Those two together say the misses are timing, not affordability.
Texas	8,753	3,827 43.7%	683 18.0%	Biggest account and the worst recovery. 3,133 of its agreements are never-recovered openings — the largest single pile in the business.
New York	7,039	2,948 41.9%	530 18.1%	Second-worst recovery. Its recurring location is Recurring NYC — the prefix-match trap that has dropped it from ten reports.
Florida	4,342	1,756 40.4%	400 23.1%	Highest banked-by-365 of the mature accounts at $1,064 per agreement.
Boston	3,362	1,153 34.3%	258 22.6%	Best in the business by a clear 6 points. Nobody has explained why — worth asking Boston what they do differently at the point of sale.

The staffing number. Measured 16 Sep 2026 straight from the Square recurring book over the last four complete months, 1 May – 31 Aug 2026 (123 days = 17.57 weeks): 3,735 agreements opened, 1,784 opening charges failed (47.8%), $398,634 of missed first instalments. Same definitions as /instalment-recovery — recurring only, an agreement is a run of charges broken by 12 months of silence, the first instalment is that run’s opening charge. September is excluded because it is not a complete month.

Measure	Per week	Read it this way
New missed first payments	101.5 / week	1,784 over 17.57 weeks. On the 17 complete Mon–Sun weeks: mean 97.3, median 98, and it swings from 43 to 144 — roster for the median, not the mean.
Still unrecovered at day 1	91.6 / week	10 a week come back on their own within 24 hours. They cost nothing and need no call.
Still unrecovered at day 3	86.3 / week	The realistic call queue. Let the free recovery have days 0–2 and you drop 15 people a week off the list for no effort at all.
Still unrecovered at day 7	79.8 / week	Waiting to day 7 only removes another 6.5 a week — and by then 92% of the ten-day return is already spent. Not worth the delay.
Never back within 10 days	78.0 / week	The people a process exists for. Everything after this is won or lost by a person.
Value arriving each week	~$22,684	$398,634 of missed first instalments over the four months. Average miss $223.

Make sure you are sizing the right queue — these three are all called “misses”

Three different populations, three different staffing answers, and they are routinely mixed up. A missed first instalment is the smallest of them by a factor of twelve.

Population	Per week	What it is
Every failed recurring charge	~1,274 / week	46,985 declines Jan 1 – Sep 16 2026 over 36.9 weeks. Includes the same person failing month after month, so it is a workload ceiling and nothing else. `/decline-trends`.
Accounts newly entering arrears (any plan position)	~359 / week	About 72 a working day. This is what `/debt-plan` means by “new misses” and it is the number behind its 2.1 FTE figure. Measured 20 Aug 2026.
Missed FIRST instalments	~102 / week	This page’s number, and the smallest of the three. ~21 a working day. Measured 16 Sep 2026 over the four complete months May–Aug.

Month by month, so you can see it is not one odd month

Month	Agreements opened	Opening charge failed	Misses per week	Still out at day 3
May 2026	822	417 50.7%	94.2	366
Jun 2026	993	486 48.9%	113.4	400
Jul 2026	947	414 43.7%	93.5	351
Aug 2026	973	467 48.0%	105.5	399

Across the 17 complete Mon–Sun weeks inside the window: mean 97.3 misses, median 98, quietest week 43, busiest week 144. That is a 3.3× spread, so a rota built on the average will be underwater one week in four. The quiet weeks are the ones containing a public holiday (6 Jul, 3 Aug); the busy ones are month-end.

The day the miss arrives — this is the rostering fact

Bucketed by the day the opening charge actually failed. The queue is not flat and staffing it flat wastes a third of it.

Day it lands	Misses in the window	Per week	Read it this way
Friday	533 29.9%	30.3	Nearly a third of the whole week lands on Friday. This is the single biggest rostering fact on the page — and Friday is also the best-converting day of the week on the book, so the queue arrives exactly when the money is most likely to be there.
Monday	328 18.4%	18.7	Monday and Friday together are 48.3% of the week.
Wednesday	201 11.3%	11.4
Sunday	200 11.2%	11.4	Weekend arrivals sit until Monday — they are part of why Monday looks busy.
Saturday	191 10.7%	10.9
Tuesday	181 10.1%	10.3
Thursday	150 8.4%	8.5	The quietest day by a distance — a third of Friday. Do not staff the week flat.

By city, and by what the bank said

City	Agreements opened	Opening charge failed	Misses per week
Texas	1,022	515 50.4%	29.3
Florida	879	397 45.2%	22.6
Chicago	536	270 50.4%	15.4
New York	527	263 49.9%	15.0
Nashville	413	196 47.5%	11.2
Boston	358	143 39.9%	8.1

Route	Misses in the window	Per week	What that route is
Soft — a card the issuer might still say yes to	1,555 87.2%	88.5	TRANSACTION_LIMIT 54.2/wk · GENERIC_DECLINE 21.6/wk · INSUFFICIENT_FUNDS 12.6/wk. The main queue.
Issuer has blocked us — they must ring their own bank	203 11.4%	11.6	A short, specific call with one instruction. Worth its own script because no re-charge can ever work.
Dead card — needs a NEW card on file	26 1.5%	1.5	Under two people a week. Any process built around expired and dead cards is built for almost nobody.

The man-hours, with every assumption stated

Two columns: working the queue from day 3 (86.3 people a week, after the free recovery has had its go) and from day 1 (101.5 a week). Assumptions are /debt-plan’s own so the two pages can be compared. Argue with the assumptions, not the arithmetic — and note the contact rate is the one input we do not measure.

Activity	Assumption	Start at day 3 86.3/wk	Start at day 1 101.5/wk	Note
Dial the queue	3 attempts × 2 min	8.6 hrs	10.2 hrs	Three attempts at different times of day per person, ~2 minutes each including the ones that ring out.
Talk to the ones you reach	~9 min each	2.6 hrs	3.1 hrs	At a 20% right-party contact rate — the midpoint of the 15–25% consumer-lending norm. We do not measure our own, so this is the softest input here. At 15% it is 1.9/2.3 hrs; at 25% it is 3.2/3.8 hrs.
Notes and admin	~3 min per account	4.3 hrs	5.1 hrs	Per account touched, not per conversation.
TOTAL		~15.5 hrs/wk	~18.4 hrs/wk	0.41 FTE if you start at day 3, 0.48 FTE if you start at day 1 (38-hour week). Starting on day 1 costs ~3 extra hours a week and buys you the 10 people a week who would have come back on their own — so day 3 is the better buy, but only just, and day 1 has the advantage of catching people while the failure is still news to them.

So: roughly half a person. 15–19 hours a week, 0.4–0.5 FTE on a 38-hour week, to call every new missed first payment three times. For comparison the whole debt team currently produces 20.2 conversations of two minutes or more per day, so this queue is about 17% of one person’s week — affordable, but it is not free capacity, it has to come off something. The number this does NOT give you is what those calls will collect; nothing like this has been run as a measured process, so hold a control arm back or you will never know.

The single most actionable thing on this page. Population: 11,475 agreements whose opening charge failed and which have at least ten days of runway behind them. “Recovered” means a recurring charge for that customer was banked after the failed attempt — a same-day second attempt counts as day 1.

Day after the miss	Recovered that day	Recovered by this day	Money banked by this day	Share of the 10-day return	Note
Day 1	1,099	1,099 9.6%	$338,463	48%	Nearly half of everything the ten-day window ever returns lands on day one. Any process that starts on day 3 has already lost most of it.
Day 2	315	1,414 12.3%	$424,971	61%
Day 3	217	1,631 14.2%	$473,933	71%
Day 4	175	1,806 15.7%	$518,154	78%
Day 5	127	1,933 16.8%	$547,017	84%
Day 6	98	2,031 17.7%	$568,089	88%
Day 7	99	2,130 18.6%	$589,263	92%	A 7-day window captures 92% of what a 10-day one would.
Day 8	71	2,201 19.2%	$608,399	95%
Day 9	53	2,254 19.6%	$620,481	97%
Day 10	56	2,310 20.1%	$638,000	100%	Days 8–10 together add only 180 recoveries — 1.6% of the population.

And after day 10

Each horizon has its own population, because a longer horizon needs more runway. The n column is that population, not a subset of the ten-day one.

Horizon	Misses with this much runway	Recovered by then
Within 15 days	11,435	2,512 22.0%
Within 20 days	11,288	2,587 22.9%
Within 30 days	11,142	2,937 26.4%
Within 60 days	10,739	3,897 36.3%
Within 90 days	10,316	4,337 42.0%
Within 180 days	8,973	4,432 49.4%
Ever, however long it took	11,585	5,579 48.2%

What catching it is worth — average banked per agreement

This is the question the whole area exists for. By day 365 a recovered miss banks $1,165 against $1,327 for an agreement that paid first time and $243 for one we never got back. Catching a miss inside ten days is worth $922 per agreement, and recovers 88% of a clean payer. A recovered miss even runs ahead of a clean payer at 30 and 60 days — because a clean payer’s second instalment has not fallen due yet.

Measured at	Missed, never recovered	Recovered ≤10 days	Paid first time	Recovered as % of a clean payer
Day 30	$18	$303	$282	107%
Day 60	$52	$431	$417	103%
Day 90	$82	$538	$554	97%
Day 120	$108	$633	$676	94%
Day 150	$131	$719	$789	91%
Day 180	$153	$802	$893	90%
Day 270	$200	$1,025	$1,169	88%
Day 365	$243	$1,165	$1,327	88%

The control: strip the opening charge out of both groups and ask only what came in afterwards. The gap survives it — at 365 days the missed group still banks 0.39× what the paid group does. That is what makes this behavioural rather than arithmetic: a missed opening charge is not one lost payment, it is a different agreement.

Why the opening charge was declined, across all 11,475 missed openings with ten days of runway. The shape is not what anyone assumes — this is where a routed script earns its money. Decline codes live on the payment, not the invoice: /v2/payments with an explicit location_id → card_details.errors[].

Reason	Missed openings	What it means	Recovered by day 10	Ever banked anything	Read it this way
TRANSACTION_LIMIT	6,810 59.3%	Card is at its limit	20.3%	55.0%	The biggest single reason by a mile, and it behaves exactly like a no-money decline — measured 16 Sep 2026 across 18,467 of them in 2026: money lands after 20.9% within 30 days and 38.8% ever, against 21.3% and 39.5% for INSUFFICIENT_FUNDS. Two codes, one behaviour. It is not a per-charge cap (the low-value ones recover best, not worst) but treat it as the card having no room, not as a bank quirk on a funded account.
GENERIC_DECLINE	3,200 27.9%	Bank declined and gave no reason	19.6%	41.4%	No information at all. Square’s `detail` field is worthless here — it just echoes the code back (`"Authorization error: 'GENERIC_DECLINE'"`), so there is no human-readable sentence to read to a customer; any plain-English mapping has to be ours. Treat as “could be money”, not “not money”.
INSUFFICIENT_FUNDS	742 6.5%	Not enough money in the account	23.3%	62.0%	The smallest of the money reasons and the best recovering. Only 6.5% of first misses are an explicit no-funds — so a call that opens “your payment bounced” is talking to the wrong 93%.
CARDHOLDER_INSUFFICIENT_PERMISSIONS	558 4.9%	Issuer has blocked this kind of charge	17.4%	37.3%	The blocklist case: the card is fenced off from card-not-present recurring charges and only the cardholder can lift it. A re-charge cannot pass. This one genuinely needs the customer to ring their own bank — it is the only reason where that instruction is correct.
INVALID_ACCOUNT	75 0.7%	Issuer cannot find the account — card is dead or closed	14.7%	23.1%	Needs a NEW card, not an expiry edit. Distinct from expiry and much rarer than people assume.
PAN_FAILURE	53 0.5%	Card number is wrong or dead	15.1%	48.8%	Structural. Meaning inferred, not documented by Square.
CARD_EXPIRED	30 0.3%	Card has expired	33.3%	45.5%	Thirty people out of 11,475. An “update your expiry date” campaign is aimed at almost nobody — and separately, 91.2% of declined cards have 12+ months of life left. Card decay is not this business’s problem, which is why a card-refresh campaign was ruled out on `/debt-plan`.

The no-card list — 100 names, built 2026-09-16

The bucket below is the only one on this page with a live worked list behind it. Swept 30,000 invoices across the six recurring locations: 10,973 SCHEDULED, of which 714 6.5% hold no card on file — an independent rebuild of the 6.5% figure. The CSV carries the 100 soonest to fall due.

Measure	Value	Read it this way
Due window	2026-09-17 → 2026-09-21	The 100 nearest. Sorted soonest-first because these are the ones about to fail next; `Invoice raised` is in the CSV to re-sort by when it was created.
About to not be taken	$16,614	Across the 100. Nothing will be attempted — there is no card to attempt it with.
Already in arrears	61 61%	$17,768 already past due behind these 100 — more than the amount about to be missed. The card came off and the arrears kept building with nothing to stop them.
Have paid us before	86 of 100	This is the finding. Only 14 never paid a penny. The rest were paying normally and their card came off — expired, replaced, removed — and nobody noticed, because a missing card generates no decline code and no alert.
No card ever seen	14	Never had one on file at any point. These are the genuinely new-and-never-set-up.
On an early-access bank	10	Small here — this bucket is not a neobank problem.
By city	Texas 149 · Boston 116 · New York 169 · Florida 126 · Chicago 70 · Nashville 84	Whole population of 714, not just the 100.

Download: https://reports.iconicbyai.com/first-miss-nocard.csv?key=studio1-sales-2026
32 columns: name, phone, email, CRM ref, amount, due date, invoice, when it was raised, past-due count and value, whether they have ever paid, charges / paid / failed / total banked, first and last attempt, days since, last status and decline reason, last known issuer and card, whether Square has any reminder set on it, and a ready-made /pay link per person. It is key-gated because it carries customer contact details, and it is never committed — the file is gitignored and parked in Supabase so the route can serve it. Rebuild with node scripts/reports/report_nocard_list.js.

Four things the reason table does not show

The fourth bucket nobody has: no charge was ever attempted

6.5% of SCHEDULED invoices hold no card on file — 658 of 10,133 sampled across all six recurring locations (Texas 4.2%, Boston 9.2%, New York 7.9%, Florida 6.6%, Chicago 5.4%, Nashville 7.0%). They cannot auto-charge, so they produce no decline code, land in no bucket, and appear on no list. On 2026 first misses that group recovers 6.3% next cycle against 28.2% for a plain no-money decline — the worst on the book. In the Boston sample, 7 of 41 misses had no payment attempt at all. These people must never be told “your card was declined” — nothing was declined. This is the single cleanest call list in the business and it currently does not exist.

The damage is done between miss 1 and miss 2

Given a series has just missed N in a row, does the next cycle get paid? 1 miss → 20.8% (n = 51,548, $12.39M owed at that step) · 2 → 12.9% · 3 → 7.9% · 4 → 5.0%. The collapse is between the first and second miss. Arguing about a cut-off at four misses is arguing about what is left after the loss has already happened. 2,751 series are sitting at 1–3 misses right now, $932,466 owed on those runs.

Never-paid-a-cent climbs with the run, and it beats the decline reason as a predictor

28.5% of payers, 41.5% at one miss, 43.3% at two, 46.5% at three, 53.4% among the 4+ zombies. That is the strongest day-one split available — stronger than the decline reason, where every family clusters between 10% and 31% next-cycle recovery. Sort the call list on “have they ever paid us anything” first, and on the reason second.

Recovery on the ORIGINAL invoice barely moves by reason — and that is the point

TRANSACTION_LIMIT 6.6% · INSUFFICIENT_FUNDS 6.2% · CARD_EXPIRED 6.0% · INVALID_ACCOUNT 5.5% · GENERIC_DECLINE 4.7% (July 2026, 10,903 due invoices). An expired card — trivially fixable — recovers no better than genuine no-funds, because nothing currently treats them differently. That flat line is the case for routing, not evidence that routing would not work.

Who misses, and what that does and does not tell you. The card and issuer figures come from a BIN feed that covers 99.7–99.9% of charges through August 2026 and 0% of September — a month with low coverage is a month the feed has not reached, not a month without prepaid cards.

Figure	Value	Period	Source	Read it this way
Early-access neobank cards	83.3% decline	Jan–Aug 2026, 12,644 declines	`/decline-trends`, built 16 Sep 2026 07:36.	Against 50.3% for traditional banks and 73.9% for Direct Express benefits cards. The sponsor-bank cluster behind Cash App, Chime, Current, Varo, Dave, SoFi and MoneyLion. The 8 worst banks on the whole book are all early-access and the break is clean — best early-access at volume is SoFi at 23.9% success, worst bank of any other kind is M&T at 24.8%, with nothing in between. Two populations that barely overlap. A BIN identifies the ISSUER, not the person.
Share of the first-miss list	30.6% early-access	Current re-charge list, 16 Sep 2026	`/retry-list`, built 16 Sep 2026 08:02.	205 of 674 in the missed-first-payment cohorts. On the forward book, 499 of 2,555 identifiable cards falling due this week (19.5%) are early-access.
It is NOT a “paid two days early” effect	Both peak Friday	Jan–Aug 2026, 78,016 BIN-matched charges	`/decline-trends`, 16 Sep 2026.	If the early-access story were really about wages landing early, those charges would do best mid-week. They peak on Friday and trough on Sunday — exactly like traditional cards. What differs is the size of the swing: early-access 1.8× best day to worst, traditional only 1.1×. These accounts run empty faster, they are not paid on a different clock. Do not build a call or charge schedule on the early-wages theory.
We charge hardest on the worst days	41.2% vs 43.9%	82,673 recurring charges, Jan–Sep 2026	`/decline-trends` calendar tab, 16 Sep 2026.	Best day of the month is the 1st at 49.5%, worst is the 14th at 38.8%. Our four busiest days — the 15th, 31st, 28th and 20th — carry 26.7% of every charge we make and three of them sit in the worse half of the month. Separately, the 31st is simultaneously the most-chosen charge day (9.1% of agreements) and the second worst.
Moving the charge date was RULED OUT as a programme	marginal	451,285 CRM charges	`/debt-plan`, 20 Aug 2026.	Day-of-month success runs 40–49% across the whole month and day-of-week 42–48% — a spread of five to eight points. Real, but marginal against the 46-point swing available from when you chase. And charging ahead of a stated payday did not hold: within 2 days of payday collected 34.5% against 37.5% further away — heavily confounded, so read it as not demonstrated, not disproven.
Card brand on the missed opening	VISA 8,171	Whole cohort	`/instalment-recovery`, 16 Sep 2026.	VISA 8,171 (20.1% recovered ≤10d, $399 banked by 365) · MASTERCARD 2,914 (19.9%, $454) · DISCOVER 278 (24.8%, $895) · AMEX 112 (16.1%, $451). Discover is a small, distinctly better-paying population.
Where they sit in the plan	1st charge 46.4%	Jan–Sep 2026	`/decline-trends`, 16 Sep 2026.	Decline rate by plan position: 1st 46.4% · 2nd 44.6% · 3rd–6th 55.6% · 7th–12th 59.3% · 13th+ 66.7%. That rise is survivorship, not decay — clean payers finish and leave the book, so the deeper you look the more the remainder is people who were already struggling. The honest reading is “a charge deep in a plan is far more likely to fail”, useful for where to put effort, not “plans decay”.
Prepaid cards	81–85% decline	Jan–Aug 2026	`/decline-trends`, 16 Sep 2026.	Against debit 54–60% and credit 38–43%, and it has worsened every month. Prepaid wins over debit/credit where a card is both — a prepaid debit card behaves like prepaid.

The neobanks — Cash App, Chime, Current, Varo, Dave, SoFi, MoneyLion — and the sponsor banks that actually issue their cards. A BIN identifies the ISSUER, not the person. “Early access” means the population that can be paid wages early, not a proven list of people who are. Measured 16 Sep 2026 on first instalments only, 1 May – 31 Aug 2026: 3,733 of 3,735 carry an identified card (99.9%). The window ends 31 Aug because that is where the BIN feed ends.

Measure	Value	Read it this way
Share of first instalments	19.0%	708 of 3,733 with an identified card.
Share of the MISSES	32.3%	576 of 1,784. One miss in three is a neobank card.
First instalment failure rate	81.4%	Against 39.8% for a traditional bank — more than double.
Misses a week	32.8	Of the ~102 total. Traditional banks contribute 68.0.
Come back within 10 days	18.8%	Against 25.2% traditional. They fail more AND recover less.
Ever bank anything after the miss	28.3%	Against 43.3% traditional. The gap widens with time, it does not close.

Is it the neobank, or is it just a prepaid card?

It is the bank. Prepaid makes everything worse, but it does not explain the gap.

Class	Card	First instalments	Failed	Read it this way
Early-access	Prepaid	316	275 87.0%	The worst combination on the book.
Early-access	Normal card	392	301 76.8%	This is the row that settles it. A perfectly ordinary non-prepaid card from a neobank still fails 76.8% of the time — nearly double a normal high-street card. So it is not the prepaid product, it is the account behind it.
Traditional	Prepaid	47	28 59.6%	Prepaid is worse everywhere, but it adds ~20 points, not 37.
Traditional	Normal card	2,956	1,167 39.5%	The baseline.

Why they fail — and the one difference worth a script

576 early-access misses against 1,195 traditional ones. Four of the five reasons are the same shape in both. The fifth is not, and it is the most fixable thing on this whole page.

Reason	Early-access	Traditional	Read it this way
TRANSACTION_LIMIT	307 53.3%	640 53.6%	Identical. The biggest reason is the same for everyone.
CARDHOLDER_INSUFFICIENT_PERMISSIONS	138 24.0%	64 5.4%	The one real difference, and it is 4.4×. The issuer has fenced the card off from card-not-present recurring charges. Only the cardholder can lift it — usually a toggle in the neobank's own app. This is the single most fixable thing in the whole first-miss population and it has its own script: 138 people in four months, ~8 a week.
INSUFFICIENT_FUNDS	72 12.5%	145 12.1%	Identical. Neobank customers are not more likely to be explicitly out of money.
GENERIC_DECLINE	52 9.0%	325 27.2%	Three times LOWER. Neobanks tell us why; high-street banks often do not. That is a reporting difference, not a behavioural one.
INVALID_ACCOUNT	7 1.2%	8 0.7%	Dead cards are rare in both.

Named banks, worst first — 15 or more first instalments to qualify

The class was never really the finding; the individual bank is. Sutton Bank alone — the sponsor behind Cash App and Chime — is 309 first instalments and 269 of them failed. At the other end, a Chase card opens a plan successfully 93.5% of the time.

Issuing bank	Class	First instalments	Failed
GREEN DOT BANK DBA BONNEVILLE BANK	Early-access	23	22 95.7%
SUTTON BANK	Early-access	309	269 87.1%
SOFI BANK	Early-access	23	19 82.6%
THE BANCORP BANK NATIONAL ASSOCIATION	Early-access	50	39 78.0%
THE BANCORP BANK	Early-access	112	86 76.8%
STRIDE BANK	Early-access	99	74 74.7%
COMERICA BANK (Direct Express)	Benefits	19	13 68.4%
NAVY FEDERAL CREDIT UNION	Traditional	140	78 55.7%
AMERICAN EXPRESS US CONSUMER	Traditional	47	10 21.3%
TRUIST BANK	Traditional	39	7 17.9%
BANK OF AMERICA - CONSUMER CREDIT	Traditional	56	10 17.9%
JPMORGAN CHASE BANK N.A.	Traditional	124	8 6.5%

Does the money really land early? Not in a way we can charge against

If the early-wages story were true, neobank cards would do best mid-week — after the early money lands and before the Friday payroll. Share of first instalments that went through, by the weekday the charge was actually made.

Day charged	Early-access charges	EA went through	Traditional charges	Traditional went through	Note
Monday	116	12.9%	542	58.7%	The worst day for neobank cards by a distance. Weekend spending has landed and Friday's wages are gone.
Tuesday	69	20.3%	303	59.1%
Wednesday	85	16.5%	328	60.7%
Thursday	74	23.0%	234	60.3%	Nominally the best day — but on a third of Friday's volume, and the difference against Friday is well inside the noise.
Friday	212	21.2%	972	62.7%	The best-powered cell in the table (n = 212) and it is statistically indistinguishable from Thursday. Both classes peak here.
Saturday	73	15.1%	328	60.7%
Sunday	79	20.3%	296	55.1%

The verdict: do not move charge dates on the early-wages theory. Thursday reads highest at 23.0% but on n = 74; Friday is 21.2% on n = 212 and the difference is roughly 1.7 standard errors — noise-compatible. And the far better-powered whole-book measurement (/decline-trends, 78,016 BIN-matched charges) has both classes peaking on Friday and troughing on Sunday, the same shape. What genuinely differs is the size of the swing — early-access moves 1.78× from its best day to its worst against 1.14× for traditional banks. These accounts run empty faster; they are not on a different clock. The real signal here is Monday at 12.9%, the worst cell in the table: the weekend has already taken the money.

What this means for the call process

A third of the queue is one kind of customer, and they are the hardest third

~33 of the ~102 weekly misses. They fail at 81.4% and only 28.3% ever bank anything afterwards, against 43.3% for everyone else. Failing more and recovering less compounds — this cohort is where a disproportionate share of the never-paid-a-cent pile comes from. Worth its own queue, its own script, and its own measurement.

“Go into your banking app and allow recurring payments” — ~8 people a week

24.0% of early-access first misses are the issuer blocking card-not-present recurring charges, against 5.4% on traditional banks. No re-charge at any amount can pass one of these, and only the cardholder can lift it. It is a sixty-second instruction with a specific destination, and it is the only place in this whole report where “contact your bank” is the correct thing to say. Today nobody says it, because the decline reason never reaches GHL.

The lever is at the point of sale, not on the phone

A Chase card opens a plan successfully 93.5% of the time; a Sutton Bank card 12.9%. That is decided when the card is taken, months before anyone rings. Asking for a second card at signing — or steering to bank debit, which is /debt-plan action 2 — is worth more than any call script, and it is the only intervention that acts before the failure instead of after it. Not modelled, and not costed — but it is where this evidence points.

What actually happens to a missed first payment right now, end to end. Read this before designing anything — three of the pieces a new process would need already exist, and one of them (the GHL feed) will double-text people if a second entry point is added beside it.

Stage	What it does	Detail
Square	Charges once and stops	The scheduled invoice carries `automatic_payment_source: CARD_ON_FILE`; every past-due one reads `NONE`. 0.90 failed charges per missed cycle. The series keeps minting a new invoice every cycle forever — series 00440822 minted invoice 17 on the same day invoice 16 fell due unpaid, fifteen misses in. Failures never feed back into the schedule. Square’s Invoices API cannot pause, end or even read a recurring series — ending one is Dashboard-only, a human job.
GHL — the automated side	`Debt not debt` / `2nd Missed payment`	Missed payments enter the US Debt Collection sub-account (`pcTyOadbzBqZO7v3cbwn`) via the Debt AI G-Sheet API — a Google Sheet feed, not our code. New lead with no owner → Assign states for debt lead → Debt not debt (first-time missers: offers three options to find out why they missed, handles it with AI, assigns a team member). Everyone else → 2nd Missed payment Automation. Workflows cannot be built or edited via the API — the canvas, branching and message bodies are GHL UI only. Code can upsert contacts, set fields and tags, and enrol a contact into a workflow.
GHL — what it does NOT know	The decline reason	No decline-reason field exists on that sub-account. 23 contact custom fields, none reason-related; of 881 tags the only card-ish ones are `card_issue` and `card handling - <person>`. So the AI conversation that asks “why did you miss” is asking a question Square already answered. Getting the reason into GHL is the actual build behind any routed process — and the preference is to decorate the existing contact so Debt not debt can branch on it, rather than start a second feed that double-texts the same person on the same day.
The daily list	`/chase-daily` → one CSV a day	Built on Render straight after `square-declines`, one file per day for Hamad to push to GHL. The cut rule was removed on 3 Sep 2026 — Neil: “i didnt say to hold anyone back. in the report i said if they miss a payment you add them in.” A day’s file is now everyone who missed that day. Before that fix the hand-over ledger never released anyone, so on 31 Aug 683 of 897 people were cut as duplicates, 677 of them last handed over on 6 Aug. The recovery list that created is `/missed-chase` — 1,020 people, $1,254,900 past due.
The phone	~20 real conversations a day	Last 28 calendar days (18 active), all six debt lines together — Nick, Emily, Maria, Charles, Collections 1 NY, Collections 2 FL: 3,395 calls, 3,169 outbound, 50.0 hours of talk time, and only 364 calls lasted 2 minutes or more. That is 20.2 real conversations a day for the entire team. Weekends are zero, correctly — they do not work them.
Who they reach	968 live of 11,605	Across the whole US Debt Collection sub-account: 10,038 accounts owe $16.79M, and 968 have answered a text in the last 30 days — 8.3%. 6,706 have replied at some point, which reads as 58% engaged; quote 968, never 6,706. Voice calls are not in the GHL thread, so a phone-worked card reads as unchased.
Who does the collecting	The robot, increasingly	Jun→Aug 2026 total collections were flat (~$846k → $808k) while the automated run went +19.7% ($385,656 → $461,743) and every human desk fell −24.8% ($461,219 → $346,876). Both the “we are quiet” and “the totals look the same” reports were right. Only two accounts clear the habitual-batch bar: Natalia Accounting and Studio1 Collections.
The escalation ladder Neil already approved	S0 → S5, Caet LAST	Agreed 31 Jul 2026: S0 Square miss + decline reason (day 0) → S1 automated self-serve recovery (d0–7) → S2 automated persistent chase (d7–21) → S3 collector calls (d21+) → S4 Caet’s reduction offer (d45+) → S5 Monterey. Payment at any stage drops the account out. Caet is the last human stage, not the front door — “one cog in the wheel and her cog is almost at the end”. Timings (7/21/45) and whether S3 is the wider team were left open.

We do not send one today. I checked the whole corpus rather than assuming. 439,694 note lines, 43,274 distinct bookings, 2.05 billion characters of CRM notes — which carry the pasted email correspondence as well as staff notes — searched for anything resembling a message about an upcoming payment. There is nothing. Every hit is reactive: after a decline, a complaint, a cancellation request or a reduction offer.

What was searched for	Bookings carrying it	What those hits actually are
“your/the first payment”	971 2.2%	Every one read is after the fact — a reduction offer, a missed payment, a cancellation request. Typical: “please find the invoice link provided below for your first installment toward your reduced balance” — that is a rescue, not a heads-up.
“payment plan / recurring starts”	33 0.08%	Thirty-three bookings in six years, and they are internal notes a collector typed after setting a plan up on the phone (“Set up on recurring starting 2/26 of $50.00”), not messages sent to anyone.
“upcoming / scheduled payment”	361 0.8%	All reactive. The only automatic notification that appears anywhere in the corpus is quoted in our own words: “This is an automatic notification concerning multiple missed payments* on your account.”* We have an automated message for failure and none for anything else.
“card on file / confirm your card”	2,106 4.9%	The closest thing that exists, and it still fires only once a problem is already visible: “I was trying to give you a call to set your account up for recurring payments. Unfortunately, I am showing an error with the card on file.” Triggered by a detected error, not by a due date.
“welcome / thank you for your purchase”	1,630 3.8%	No onboarding message. The word appears inside collections correspondence — “you’re welcome to pay the balance in full”.

The clearest evidence is a customer telling us so, unprompted, in 2022: “This is the first email or contact I have received in regards to the contract. I was under the assumption that the bill was being automatically debited monthly. Please email me back in regards to getting a payment plan started, as it must’ve been going to another email.” Someone who wanted to pay, could not tell whether they were paying, and had to chase us.

Does Square send anything itself? Read straight off the Invoices API

Checked 16 Sep 2026 because it was worth ruling out before writing anything: the 100 most recent recurring invoices in each of the six accounts (600 in total) plus the 2,671 scheduled invoices on /due-next-7. Square sends by email, and on an auto-charge invoice it sends no reminder at all.

What was checked	What Square actually does	Detail
Delivery method	EMAIL, not SMS	600 of 600 recurring invoices sampled are `delivery_method: EMAIL`. On the forward book it is 2,667 of 2,671 — just 2 invoices in the entire business are set to SMS (both Texas) and 2 are share-manually. Square is not texting anybody about an upcoming payment.
Invoices with any reminder configured	23 of 600 3.8%	96.2% of recurring invoices have `reminders: []` — an empty array. Nothing is scheduled to go out at all.
On AUTO-CHARGE invoices (card on file)	0 of 557	This is the finding. Not one single invoice that will be automatically charged has a reminder on it. The entire population this page is about gets nothing from Square before the charge — ever.
On NO-CARD invoices (emailed to pay)	23 of 43 53%	Reminders exist only where there is no card to charge. Exactly backwards: the people who need no reminder because they must act anyway are the only ones who get one.
The schedule where it does exist	−7d, 0d, +1d, +3d	And the −7 and the 0 both read `status: NOT_APPLICABLE`, so they never fire. Only +1 and +3 are `PENDING` — both AFTER the due date. Even on the invoices that have reminders, every reminder that actually sends is a chaser, not a heads-up.
The message text	(none)	Every reminder has `message` unset, so Square sends its own default invoice template. Nobody has ever written a word of this.

If somebody has seen a text from Square, it is almost certainly a payment RECEIPT — Square texts a receipt to customers who have opted into SMS receipts with Square itself, across every merchant they buy from. That fires after a successful charge, we do not write it, we do not control it, and it never reaches anyone whose payment failed. It is not a reminder and it cannot do this job.

And do not fix this by switching Square's own reminders on for auto-charge invoices. Square's default template is an “you have an invoice to pay” email. Send that to someone whose card is about to be charged automatically and a proportion will pay it manually — then the stored card fires too. A duplicate charge on a first payment is the fastest route to a chargeback there is. The message below is worded to avoid exactly that.

The message — SMS, first payment due tomorrow

Roughly 290 characters, two segments. Merge fields in braces.

Hi {FirstName} — Studio 1 here. Your first payment of ${amount} is scheduled for tomorrow, {date}, on the card ending {last4}.

Please make sure that card's good to go.

If it's changed, or tomorrow isn't a good day, just reply to this message or call {phone} and we'll sort it out with you.

Reply STOP to opt out.

The email version

Same message with room for the bank-app line, which will not fit in an SMS.

Subject: Your first payment is scheduled for tomorrow

Hi {FirstName},

Just a quick heads-up so nothing catches you out: the first payment on your {studio} portfolio, ${amount}, is scheduled for tomorrow, {date}. It'll come off the card ending {last4}.

Please make sure that card is ready to go.

If anything has changed — a new card, a different bank, or tomorrow just isn't a good day — reply to this email or call us on {phone} and we'll sort it out. It's much easier to move it now than to fix it afterwards.

One thing worth checking if you bank with Chime, Cash App, Current, Varo, Dave, SoFi or MoneyLion: some of these apps block recurring payments by default. If yours does, switching that on now will save the payment being declined.

Thanks,
{SenderName}
Studio 1 — {phone}

The variant that is not optional — no card on file

6.5% of scheduled invoices hold no card at all (658 of 10,133 sampled across the six recurring locations). Nothing will be charged, so the main message would be a lie to them — and this is the worst-recovering group on the whole book, 6.3% next cycle against 28.2% for an ordinary money decline. They also generate no decline code, so no existing list has ever seen them. This is the cheapest win on this page.

Hi {FirstName} — Studio 1 here. Your first payment of ${amount} is due tomorrow, {date}, but we don't have a card on file to take it.

You can add one here: {payLink}

Or call {phone} and we'll do it with you — takes a minute.

Reply STOP to opt out.

Eight rules the copy is built on

This is a service message, not a collections message — and the difference is the whole point

They have not missed anything. Nothing is overdue. Nothing in this message may read like a debt letter: no “declined”, no “overdue”, no “failure to pay”, no consequences. The current single first-miss message treats everyone as a debtor and that is the likeliest reason the reply rate halved between January and April 2026. If this one inherits that tone it will do more harm than sending nothing.

Name the amount, the date, the last four digits and the studio

Specificity is what separates a real message from spam, and it is what lets someone spot a problem — “that's not my card any more” is only possible if we tell them which card. It also makes the message impossible to mistake for a phishing text, which a bare “you owe us money, click here” very much is.

Offer a way out that is not a phone call

Right-party contact on a dial runs 15–25% in this industry and we do not measure our own at all. A reply to a text costs the customer nothing and reaches us either way. Reply-to-text must route to a human the same working day — if it does not, this message is worse than no message. There is precedent: a customer wrote in agreeing to pay, the email sat unread, and Caet opened her reply with “Apologies as your communication was missed and just discovered this morning”. $885 was nearly lost to an unopened inbox, not to the customer.

Do NOT put a pay-now link on the main message

A stored card is going to be charged tomorrow. Give someone a pay link today and you risk taking the money twice — and a duplicate charge on a first payment is the fastest possible route to a chargeback. The link on the main message is for updating a card. The pay link belongs only on the no-card-on-file variant, where nothing is scheduled to fire.

No discount, no offer, not ever on this message

A reliable pay-now discount on a recurring base trains people to miss — a one-off cost becomes a permanent price cut. And the evidence is already in: reducing an agreement returns $0.12 per $1 given up, and 600 of 945 accounts we reduced had no failed charge at all. This message exists to prevent a miss, not to buy one back.

The early-access line earns its place — it is the one instruction that is actually correct

24.0% of early-access first misses are the issuer blocking card-not-present recurring charges, against 5.4% on traditional banks, and only the cardholder can lift it — usually a toggle in the neobank's own app. Carrying one line about it in the email reaches roughly 8 people a week who could otherwise not be saved by anything we do. Send that line to everyone rather than gating it on the BIN: it is harmless to a Chase customer, and the BIN feed runs a fortnight behind.

Hold 10% back and send them nothing

This has never been done, so its effect is UNKNOWN — and there is a real chance it is negative. Telling someone a charge is coming also gives them the chance to move the money out or cancel. Pre-dunning is standard practice in card-billing businesses, but standard is not the same as proven here. A fixed 10% hold-back, assigned per person so a rebuild never moves anyone between arms, is the only thing that will ever tell us the net effect. /retry-list already implements exactly this pattern.

Measure payments, not replies

The metric is the share that pays on the due date, in each arm. Reply rate is the wrong measure — a link needs a tap, not an answer, and an actively-entered payment succeeds at 99.4% at any lag. A message that produces no replies and a higher pay-through rate has worked perfectly.

How it runs

	Setting	Volume	Why
Who gets it	First instalments due tomorrow	~30 a day	~971 first instalments open a month. Start here rather than the whole book: it is the highest-value moment (a caught first miss is worth $922 per agreement), it is small enough to run a clean control arm on, and it is the population this entire page is about. The same message scales to every instalment afterwards — that is ~380 a day (2,671 due in the next 7 days).
When	24 hours before the due date	9am–6pm local	We bill across Central and Eastern, so send in the customer's timezone, not the studio's. Note 29.9% of first instalments fall on a Friday, so most of these go out on a Thursday, and Monday is the worst-performing day for neobank cards (12.9%) — a Sunday send is going to the hardest cohort on its worst day.
Channel	SMS first, email alongside	both	SMS is what GHL already does and what the collections team already uses (“sent txt/em” is all over the notes). The email carries the bank-app line, which will not fit in two SMS segments.
Where the list comes from	`/due-next-7`	already built	The forward schedule already exists — every instalment Square will attempt in the next seven days, with the card, the issuing bank, the arrears and the full payment history. Filter it to first instalments falling due tomorrow. No new data pull is needed; this is a send, not a build.
What must exist before it goes out	An inbound route	blocking	Replies have to reach a person the same working day, and STOP has to be honoured. Without both, do not send it.

Nothing has been sent and nothing is scheduled. This is copy and a plan; putting it in front of a customer is an outward-facing act and needs Neil’s explicit go-ahead.

Day −5 to collections, built on the recovery curve rather than on a calendar someone liked. The stated aim is opening the lines of communication, so every step has a reply path that is not a payment and the ladder is judged on contact rate, not dollars. A step that collects nothing and gets someone talking has worked.

Why the cycle is 30 days: Square charges the card once, on the due date, and never again — 0.90 failed charges per missed cycle. The series then mints a fresh invoice for the next cycle. So day 30 is not something we trigger and it is not a re-charge; it is the next invoice falling due. The 30 days are the billing cycle, not a policy choice.

Day	What fires	Channel	Who	The ask	Why that day
−5	Pre-notice — the relationship message	SMS	Automated	Nothing. “Anything we should know?”	This one exists to open the line before there is a problem. No ask, no link, no amount to find. Neil’s main aim in one message — the only step in the whole ladder that reaches someone while nothing has gone wrong, which is the only time a reply costs them nothing.
−1	Card-ready message	SMS + email	Automated	Check the card is good	Names the amount, date and last four. Carries the neobank line — 24.0% of early-access first misses are the issuer blocking recurring charges and only the cardholder can lift it. No pay link: the stored card fires tomorrow and a manual payment on top is a chargeback.
0	The charge fires and fails	—	Square	—	Square attempts the card once. It will never attempt it again — measured at 0.90 failed charges per missed cycle. Everything after this point is us or nobody.
0 +2h	Reason-routed text + the pay link	SMS	Automated	One tap — `/pay?c={id}`	Day 1 is 48% of everything the ten-day window ever returns. Opens with what the bank actually said, not with “you failed to pay”. Four routes: no-money/limit (87% of the queue), issuer-blocked, dead card, no-card-on-file. An actively-entered payment succeeds at 99.4% at any lag.
1	Second touch, different channel	Email	Automated	Same link, more room to explain	Different channel because a dead phone is not a dead customer. Suppressed instantly if the money has landed.
2	First call	Phone	Save desk	A conversation, not a demand	The single highest-value hour in the ladder. By day 3 the ten-day window has already given up 71% of everything it will ever return. Call before the trail goes cold, not after.
3	Second call, different time of day	Phone	Save desk	A conversation	Right-party contact runs 15–25% in this industry, so one attempt reaches roughly one person in five. Three attempts at different hours is the standard answer and it is why the ladder budgets three.
5	Third call + text	Phone + SMS	Save desk	Last conversational attempt of the hot phase	Day 5 is 84% of the ten-day return. After this the curve flattens hard — days 8–10 together add only 180 recoveries out of 11,475.
7	The communication ask — three buttons	SMS + email	Automated	Pay · Arrange · Can’t this month	The most important message in the ladder for Neil’s stated aim. No payment demand at all. “Can’t this month” is a successful outcome — it tells us who to stop chasing and who to restructure, and it converts a collections touch into a retention one. 92% of the ten-day return is already in by now, so nothing is lost by asking.
10	Arrangement offer	SMS	Automated	Move the date, or split it	Not dead, just slow: recovery runs 20.1% at day 10 → 26.4% at day 30. Moving the charge date is the cheapest fix we own — the 31st is our most-chosen charge day and the second worst-performing (38.9% vs 49.5% on the 1st).
14	Check-in	SMS	Automated	Still nothing? One line.	Low cost, keeps the thread alive. No new ask.
21	Next-charge warning — the cliff message	SMS + email	Automated	Avoid a second miss	The highest-value automated message in the whole ladder. Given one miss, the next cycle is paid 20.8% of the time; given two, 12.9%. The collapse is between miss 1 and miss 2 and this is the only message that sits in front of it. Names both the arrears and the amount about to be taken.
25	Pre-notice for cycle 2	SMS	Automated	Anything we should know?	The day−5 message again, now with a live arrears balance behind it.
29	Card-ready for cycle 2	SMS + email	Automated	Check the card	Same as day −1. This is the last free intervention before the cliff.
30	Next cycle’s invoice falls due	—	Square	—	Not a re-charge — a new invoice. PAID → the cycle resets to day −5 and the arrears become a separate arrangement. FAILED → miss 2, and the tone changes.
31	Miss-2 message	SMS + email	Automated	Both amounts, named	Firmer, still not a threat. The customer now owes two instalments and most do not know it — 44% of the debt book never paid a single instalment after the deposit and they got there one silent cycle at a time.
33	Call — named collector	Phone	Recovery desk	A plan, in writing	Hand-off from the save desk. A retention conversation and a debt conversation are different jobs, different tone, different comp.
35	Settlement conversation opens	Phone	Recovery desk	Shallow catch-up offer	Only inside a conversation. Of the accounts that ever paid after a reduction, 200 of 381 paid the SAME DAY — the money was on the table in that call, the discount did not unlock it later. See the discount ladder below for the caps.
38	Final notice before escalation	SMS + email	Automated	Tells them exactly what day 40 is	No surprises. A person who knows what happens next has a reason to reply now.
40	ESCALATION — assigned, worked daily	All	Recovery desk	Settle or arrange	Neil’s threshold. By now the next-cycle hazard is 7.9% at three misses. Do not cancel the Square series here — 68.7% of money landing after a deep run is the schedule firing on a later cycle, so cancelling removes the mechanism, not just the paperwork.
45	Deeper settlement authorised	Phone	Recovery desk	Capped — see the ladder	Still capped. Cut more than half the balance and recovery falls to $0.14 per $1 given up.
60	Pre-collections notice	Email + post	Recovery desk	Final internal offer	The last point at which a settlement is worth more than the handover.
75–90	External collections (Monterey)	—	External	—	Historically accounts went to Monterey at ~120 days / 5+ misses. By then the account is worth about 15 cents on the dollar. Moving the handover earlier is the argument this ladder makes.

The forks — where somebody leaves the ladder

More important than the steps. A ladder that cannot let people off it is a ladder that turns customers into debtors.

If this happens	Then	Why
Money lands, at any step	STOP EVERYTHING	Payment is detected from Square, not from what they reply. Every queued message and call is cancelled the moment a payment posts. A chase text arriving after someone has paid is the single fastest way to lose the thread you just opened.
They reply “can’t this month”	OUT of the ladder, INTO an arrangement	This is a success, not a failure. It routes to a human who resets the date or the amount. Keeping them on a collections ladder after they have told us the truth is how a retention conversation gets turned into a debt one.
Cycle-2 charge is PAID (day 30)	RESET to day −5	The arrears do not vanish — they become a separate, gentler arrangement thread. Do not run two ladders at the same person.
No card on file	DIFFERENT LADDER FROM DAY −5	6.5% of scheduled invoices have no card to charge. They generate no decline code, appear on no list, and recover 6.3% next cycle against 28.2% — worst on the book. They must never be told a card was declined. Their day 0 is “we have nothing to take this from” plus a link.
Dead or blocked card	SKIP the payment asks	Under two people a week are a dead card, but ~8 a week are issuer-blocked and no amount can ever pass. Route them straight to the one instruction that works, and never send them a “try again” message.
Cancellation, complaint or dispute	HARD STOP to a human	Out of the ladder immediately. The AI classifies and picks approved copy; it never composes free text about someone’s debt.

The discount ladder — shallow, late, and only in a conversation

Every number here is measured, and they all point the same way: depth of the cut is the strongest signal in the data. Under 25% of the balance written off returns $0.74 per $1; 50–75% returns $0.14; clearing a balance outright returns −$0.03. And timing matters as much — of the accounts that ever paid after a reduction, 200 of 381 paid the same day. The discount does not unlock money later; it closes a conversation that was already going well.

When	What is on offer	Detail
Day 0 – 30	NO discount	The portal’s bonus credit only: +15% paying the full monthly, +10% at 75%, +5% at 50%. That is a reward for paying, not a discount for missing — which is the entire difference. Nothing here reduces what they owe.
Day 31 – 40	Catch-up bonus	Clear both instalments and the bonus credit applies to the whole amount. Still no reduction of the balance.
Day 40 – 60	Settlement — capped at 25% of the balance	This is where the evidence says the money is. Writing off under 25% of a balance returns $0.74 per $1 given up; a cut under $500 returns $0.99. Shallow, inside a conversation, once.
Day 60+	Deeper settlement — needs an approver	25–50% returns $0.41 per $1. 50–75% returns $0.14. Clearing the balance outright returns −$0.03. Beyond 50% you are not collecting, you are closing an account — which is sometimes right, but call it that and book it that way.

The bonus credit is a better instrument than a reduction and it is already built. A reduction rewards failing; the portal’s bonus rewards paying, costs the same headline percentage, and shortens the plan instead of writing the tail off. It is live at /pay today at +15% / +10% / +5% — but nothing currently applies the credit, which is the first item on the build list below.

Six guard-rails, and none of them are optional

Never discount someone who has not failed a charge

600 of the 945 accounts we reduced had no failed charge at all — 72% of the money given away, returning $0.08 per $1. We discounted people who were never fighting a card. The ladder structurally prevents this: nothing below day 31 can offer a reduction.

Once per customer per rolling 12 months, enforced by the system

Not by a collector’s memory. A reliable pay-now discount on a recurring base trains people to miss — a one-off cost becomes a permanent price cut. The word “one-off” goes in the copy and the limit goes in the code.

Track repeat-miss rate among discount takers against everyone else

This is the tell, and it is the only one. If takers miss again at a higher rate than non-takers, the discount is buying this month from next month. Agreed as a standing check on 31 Jul 2026 and never built.

Never on the no-card or admin path

They were going to pay. A discount there is pure margin loss.

Do not cancel the Square series to “stop the noise”

68.7% of money landing after a deep run is the schedule firing on a later cycle — the invoice IS the collection mechanism. And Square’s API cannot end a series at all; it is a Dashboard job, one account at a time, and it emails the customer when you do it.

Every voice call must be logged to GHL

Voice calls do not appear in the GHL thread today, so a card worked hard by phone reads as never contacted. Run this ladder without fixing that and the reporting will tell you the opposite of the truth.

The copy, step by step

Merge fields in braces. Nothing has been sent and nothing is scheduled.

Day −5 — the pre-notice

This is the relationship message. No amount to find, no link, no ask. Its only job is to make a reply cheap before anything has gone wrong.

Hi {FirstName} — Studio 1 here.

Quick heads-up: your {studio} payment of ${amount} is set to come out on {date}.

Nothing to do if that's all fine. If anything's changed your end, just reply here and we'll sort it.

Reply STOP to opt out.

Day −1 — card ready

Names the amount, date and last four. No pay link — the stored card fires tomorrow.

Hi {FirstName} — Studio 1 here. Your first payment of ${amount} is scheduled for tomorrow, {date}, on the card ending {last4}.

Please make sure that card's good to go.

If it's changed, or tomorrow isn't a good day, just reply to this message or call {phone} and we'll sort it out with you.

Reply STOP to opt out.

Day 0 — no money / limit 87% of the queue

TRANSACTION_LIMIT, GENERIC_DECLINE and INSUFFICIENT_FUNDS together. Opens with what the bank did, not with what they failed to do.

Hi {FirstName} — Studio 1. Your payment of ${amount} was declined by your bank today, so nothing has come out.

No problem — you can sort it whenever suits: {payLink}

Pay the full amount and we'll add {bonusPct}% of it back as credit on your balance.

If money's tight this month, reply and tell us — we can move the date or split it. We'd rather know.

Reply STOP to opt out.

Day 0 — issuer blocked ~8 a week

CARDHOLDER_INSUFFICIENT_PERMISSIONS. No re-charge at any amount can ever pass this. The only step in the ladder where "contact your bank" is the right instruction.

Hi {FirstName} — Studio 1. Your bank blocked your payment of ${amount} today. It isn't about funds — they've stopped this card being used for recurring payments.

Only you can switch that back on. It's usually a toggle in your banking app (look for recurring, subscriptions, or card-not-present), or one quick call to your bank.

Once it's on, you can pay here: {payLink}

Stuck? Reply and we'll talk you through it.

Day 0 — no card on file 6.5% of invoices

Nothing was declined because nothing was attempted. They must never be told a card failed.

Hi {FirstName} — Studio 1. Your payment of ${amount} was due today, but we don't have a card on file to take it from — so nothing was attempted.

You can add one and pay here: {payLink}

Or call {phone} and we'll do it with you. Takes a minute.

Reply STOP to opt out.

Day 7 — the communication ask

The most important message in the ladder for opening the lines. No payment demand. "Can't this month" is a SUCCESS.

Hi {FirstName} — Studio 1 again. We still haven't been able to take your ${amount}, and we'd rather talk than keep texting.

Just reply with a number:

1 — I'll pay now ({payLink})
2 — I want to change the date or the amount
3 — I can't manage it this month

Whichever you pick, we'll work with it. Option 3 is a real answer and we won't chase you for saying it.

Day 21 — the cliff message

The highest-value automated message in the ladder. Given one miss, the next cycle is paid 20.8% of the time; given two, 12.9%.

Hi {FirstName} — Studio 1. Your next payment of ${amount} is due on {nextDate}, and ${arrears} from last month is still outstanding.

We'd rather not have two go wrong. If the next one is going to be a problem, tell us now and we'll move it — much easier than fixing it afterwards.

Sort last month here if you can: {payLink}

Or reply and we'll put a plan together.

Day 31 — miss 2

Firmer, still not a threat. Names both amounts, because most people do not know they are two behind.

{FirstName}, your payment of ${amount} was declined again yesterday. That's two missed now and ${arrears} outstanding on your {studio} portfolio.

We need to hear from you. Reply to this message, call {phone}, or pay here: {payLink}

If you can't clear it all, say so — there are options, but they need a conversation.

Day 38 — final notice before escalation

Tells them exactly what day 40 is. A person who knows what happens next has a reason to reply now.

{FirstName} — this is the last message before your account moves to our recovery team on {escalationDate}.

${arrears} is outstanding and we've had no reply.

You can stop that happening today: pay at {payLink}, reply to this message, or call {phone}.

We would much rather sort this with you directly.

What it is judged on

Metric	Role	Detail
Contact rate	THE headline	Any reply on any channel inside the cycle. Neil’s stated aim is opening the lines, so this is the number the ladder is judged on — not dollars. Today the whole debt book runs at 8.3% answering a text in 30 days, so there is an enormous amount of room.
Right-party contact rate	NOT MEASURED YET	Of dials made, how many reached the actual person. Industry runs 15–25%. `goto_call_pull.js` discards the phone number and keeps per-agent counters — one script change, not a new integration. Without it the call steps cannot be judged at all.
Cure rate at miss 1	The money metric	Share of day-0 misses that bank something inside 30 days. Baseline today: 26.4%. This is the number the whole ladder is trying to move.
Miss-2 rate	The cliff metric	Share reaching day 30 who also fail cycle 2. Baseline: 79.2% of one-miss accounts do not pay the next cycle. The day-21 message exists purely for this.
Repeat-miss rate among discount takers	The safety metric	Against non-takers. If it is higher, stop the discount.
Payments, never replies, on the link steps	Do not confuse them	A link needs a tap, not an answer. A step that produces no replies and a higher pay-through has worked perfectly. (The exception is the day-7 message, which is measured on replies by design.)

And hold 10% back at day 0 and run them on today’s process. None of this has ever been done, so its effect is UNKNOWN and could be negative — a pre-notice also gives someone the chance to move money out. A fixed hold-back, assigned per person so a rebuild never moves anyone between arms, is the only thing that will ever tell you the net. /retry-list already implements exactly that pattern.

What has to be built first

Item	Status	Detail
Apply the bonus credit	BLOCKING	The `/pay` portal already promises “$X credited to your balance” and nothing applies it — it is written into the Square payment note and a GHL field, and no CRM or ledger write exists. The ladder cannot use the portal until this is real.
Repoint the portal’s GHL calls	BLOCKING	`notifyGHL` reads `ghl_config.json` = `XcjF95QScPf2rxRSgp2e`, the near-empty “AI Debt Collection” account. The live one is `pcTyOadbzBqZO7v3cbwn`.
Get the decline reason into GHL	BLOCKING	No decline-reason field exists on that sub-account. Without it every message on this ladder is generic, and the four-way routing that makes day 0 work cannot happen. This is the single biggest build item.
An inbound route with a same-day answer	BLOCKING	Replies must reach a person the same working day and STOP must be honoured. Without both, do not send step one — a ladder whose replies go unread is worse than silence. A customer’s agreement to pay $885 once sat unread until Caet found it by accident.
Do not build a second feed beside the G-Sheet	CAREFUL	Missed payments already enter the debt GHL via the Debt AI G-Sheet API and the `Debt not debt` workflow. A parallel entry point means the same person gets texted twice on the same day. Decorate the existing contact so that workflow can branch on the reason rather than competing with it.
The forward list already exists	NO BUILD	`/due-next-7` is the day−5 and day−1 source; `/chase-daily` is the day-0 source and already emits a per-customer `/pay?c=` link. This ladder is a send, not a data build.
Workflows are GHL UI only	CONSTRAINT	`GET /workflows/` returns metadata only; every detail path 404s and there is no create/update/delete. Code supplies the data and the enrolment; a human builds the branching.

Volume, so the desk can be sized: ~102 first-payment misses a week enter at day 0, 86 still unrecovered at day 3 — that is the call queue, and at three attempts plus notes it is ~15.5 hours a week, about 0.4 FTE. The automated steps cost nothing per person. Extended to every instalment rather than just first payments, day 0 intake is ~380 a day instead of ~30. Start on first payments: it is the highest-value moment ($922 an agreement), the smallest clean population, and the one this evidence actually covers.

Not a process — the constraints the measured data puts on one. Each of these is a thing the evidence already settles, so a process that contradicts it is starting from a known-wrong premise. The costed plan that sits on top of these is https://reports.iconicbyai.com/debt-plan.

The queue is ~21 people a working day — that is the whole design envelope

455 first misses a month against a debt team currently producing 20 real conversations a day in total. So a dedicated first-miss desk is affordable, but it cannot be bolted on top of the existing load without something coming off. /debt-plan sized the alternative: calling all first misses with no automation in front of it is three dial attempts at ~2 minutes, ~390 conversations at ~9 minutes, plus admin — 268 hours a month, about 2.1 FTE. Fix the automation first and the identical job is 0.75 FTE, because two thirds of the accounts cure before anyone picks up a phone.

Day 1 or it barely matters — but do NOT spend the call on the 20% who self-cure

48% of everything ten days ever returns lands on day one, and 20.1% come back inside ten days with nobody doing anything at all. A human call on day 1 is therefore competing with a free recovery. The evidence-shaped split: let the cheap mechanisms have days 0–2, and put the phone on the people who have not come back by day 3 — that is 85.8% of the misses and still inside the window where 29% of the ten-day return is left. A process starting at day 5 has forfeited 84% of it.

Route on the reason — but the useful split is soft vs dead, NOT broke vs not broke

59.3% TRANSACTION_LIMIT · 27.9% GENERIC_DECLINE · 6.5% INSUFFICIENT_FUNDS · 4.9% issuer-blocked · 1.5% dead card. Two routes genuinely earn their own script: “ring your bank and lift the block” reaches 558 people and is the only case where that instruction is correct, and “we need a new card” reaches under two people a week. Everything else — 87.2% of the queue — is one soft pile, and the honest framing for it is affordability and timing, not “your bank made a mistake”. TRANSACTION_LIMIT and INSUFFICIENT_FUNDS recover identically (20.9% vs 21.3% at 30 days), so the big code is a money code. The reason is not in GHL at all yet — that routing is the actual build.

The people with NO card on file need their own script, and they are invisible today

6.5% of scheduled invoices have no card to charge. They generate no decline code, so they are on no list — and they recover 6.3% next cycle against 28.2% for an ordinary money decline, the worst outcome on the book. They must never be told their card was declined. Building this list is a Square invoice read, not a new integration.

Sort on “have they ever paid us anything” before anything else

Never-paid-a-cent is 41.5% at one miss and rises with every miss, and it out-predicts the decline reason (all reason families cluster 10–31%). On the current re-charge list 2,027 have never paid a penny and 2,984 paid before and stopped — a retention call and a first-ever-payment call are different jobs with different scripts. /debt-plan action 8 says the same thing: a save desk on misses 1–2 measured on cure rate, a recovery desk on miss 3+ measured on dollars.

Do not lead with a discount — and never on a first miss

Reducing an agreement recovers $0.12 back per $1 given up; 60.4% pay nothing at all afterwards; $2,000+ reductions recover negative money. Of 945 accounts reduced, 600 had no failed charge at all — we discounted people who were never fighting a card. And a reliable pay-now discount on a recurring base trains people to miss: a one-off cost becomes a permanent price cut. The five agreed guard-rails: never on the no-card / admin path, once per customer per rolling 12 months enforced by the system, “one-off” stated in the copy, first two misses only, and track repeat-miss rate among discount takers against everyone else as the tell.

Offer a choice rather than quietly halving the charge

The next attempt at the same amount succeeds 28.8–45.7% by reason; at about half it succeeds 70.3–74.5%. But charging MORE succeeds at 71.5% — which gives the game away. What works is that a human arranged it and the customer agreed, not the size. So put three buttons on the page (pay it all / pay half now / set a new lower monthly) and capture whatever they chose as an agreement, with the schedule updated to match — otherwise it fails again next month at the old figure. Cap how far the monthly can drop or a 6-month plan silently becomes a 3-year one.

Measure effort and engagement, not dollars — and hold a control arm back

Dollars collected tells you about last quarter. The three that move first: % of first-miss accounts touched within 24 hours, contact-to-conversation, conversation-to-payment — per person, weekly, ranked, plus a 5pm exception report of what did not happen. And because there is no baseline, hold back 10% of every group and work them not at all. The current /retry-list already does this: 533 of 5,011 held back, fixed per person so a rebuild never moves anyone between arms. Without it, whatever the process collects cannot be separated from what those people would have paid anyway.

The phone is the last layer, not the first

Consumer lending measures right-party contact rate — of dials made, how many reach the actual person. It runs 15–25%, which is exactly why no serious operation builds its strategy on the phone. We do not measure it at all: goto_call_pull.js keeps per-agent counters and discards the number. Changing that one script is the whole integration. Self-service is the layer that works and we already have it — any actively-entered payment succeeds at 99.4%, at any lag — the gap is that the /pay link is not pushed at the moment of failure.

Claims about missed payments that have been made, were wrong, and were corrected. Several of these reached Neil before they were caught. If you are about to say one of these, stop and read the correction.

The claim	Verdict	What is actually true
“A second attempt converts at 92.5%, so a call process should get about that”	WRONG	That is a collector ringing someone who then moved money across. It is selected by construction — the card only goes through after the customer agreed. Expected conversion for anything automated or newly staffed is UNKNOWN. Write UNKNOWN.
“Step-down payments convert at 77% against 43% for the same amount”	MISLEADING	The MORE than the failed amount row converts at 73.5% — just as well as the smallest step-down. If shrinking the amount were the mechanism, asking for more could not work equally well. What separates the groups is that a changed amount means a human agreed it. The 33pp gap is engagement, not arithmetic.
“Most first misses are insufficient funds / expired cards”	WRONG	6.5% insufficient funds. 0.3% expired. 59.3% is TRANSACTION_LIMIT — a live card the bank is refusing. The “expired card vs no money” framing has been wrong every time it has been checked, in every account, at every window.
“87% of first missers are not broke”	WITHDRAWN — it was mine	Built on “only 6.5% are INSUFFICIENT_FUNDS”, which treats TRANSACTION_LIMIT as a bank quirk on a funded account. Measured 16 Sep 2026, it is not. Across 18,467 TRANSACTION_LIMIT declines in 2026, money lands after 20.9% within 30 days and 38.8% ever — against 21.3% and 39.5% for INSUFFICIENT_FUNDS. Identical. If the two meant different things, they would not recover the same. What is actually true is the soft/dead split: 87.2% of the queue is a card that might still say yes, and 1.5% is a dead card. That is a routing fact, not a statement about anyone’s bank balance — and it points the script at affordability and timing rather than at “your bank made a mistake”.
“We see TRANSACTION_LIMIT on $25 charges, so it is not about money”	WRONG — and it was mine	Neil asked the right question: why would anyone put a $25 payment through on the phone if they have no money? Nobody did. 248 of the 253 $25-and-under TRANSACTION_LIMIT declines in 2026 are `entry: ON_FILE` — the schedule firing a stored card on the due date with nobody present. Only five were `KEYED`, and across all TRANSACTION_LIMIT declines only 1.6% are KEYED. They are small instalments (median $20, 86 people), not a collector trying a token amount. And they are the best-recovering band, not the worst — 27.7% within 30 days against 18.8% at $200–300. The observation was real; the inference drawn from it was not.
“TRANSACTION_LIMIT means the charge is too big, so charge less”	WRONG	It is not a per-charge cap, and the recovery data says so from the other direction: after a TRANSACTION_LIMIT decline, the SMALL bands recover BEST — $0–25 sees money 27.7% of the time within 30 days against 18.8% at $200–300. If the amount were the obstacle, that would run the other way. And when money did land, 81.8% of the time it was the SAME amount, only 14.7% smaller and 3.5% larger. Stepping the amount down is not the mechanism. (Kiara Mahoney had nine consecutive TRANSACTION_LIMITs on one card: she had emigrated and the US account behind it was no longer funded. A limit code can mean an emigration, not a maxed-out borrower.)
“58% of the debt book has engaged with us”	WRONG	6,706 accounts replied at some point and the flag never expires. 4,551 of them last spoke over 90 days ago. The live figure is 968 — 8.3%.
“Chasing adds close to nothing — only 5.6% of failed charges are ever recovered”	WRONG — and it reached Neil	That is recovery on the original invoice only. Collectors take payments over the phone as separate payments, so they never touch the failed invoice and are invisible to an invoice-status view. Find them on `entry_method: KEYED` + `team_member_id`, across every location: July 2026 was 613 keyed payments into Recurring = $138,105, 16.5% of all recurring money collected, and that is a floor.
“Miss #3 on `/missed-payments` means their third miss”	WRONG	It parses `/R-(\d+)/` out of the invoice number, which is the recurring period sequence — invoice 12 of the plan. It agreed with reality on only 12 of 41 Boston rows and is wrong in both directions. The correct miss number = count that customer’s earlier payment_requests whose status is UNPAID/FAILED/CANCELED/PAYMENT_PENDING, +1. Any escalation logic keyed off that column is keyed off noise.
“Give them a reduction and they will start paying”	WRONG	$0.12 back per $1 given up. 60.4% pay nothing afterwards. Half the money goes on credits that clear the balance to zero and can never recover a cent. Of 381 that ever paid again, 200 paid the same day — the money was on the table in that conversation, not unlocked by the discount.
“Cancel the dead series at 4 misses”	EXPENSIVE	68.4% of all recoveries land at exactly 4–5 misses, while a collector is still working the file. Cutting at 4 refuses $391,740; cutting at 9 refuses $27,306 and still kills $4.67M of dead billing. And 68.7% of post-4-miss money is the schedule firing on a later cycle, so the invoice is the mechanism — cancelling removes it.
Summing unpaid Square invoices as debt	DOUBLE TRAP	CANCELED invoices are not debt — counting them added $1.52M of fake arrears and 834 fake misses in one July pull. And exposure must be per CUSTOMER, counted once: summing arrears per row re-adds the whole balance for anyone who missed twice. Also: a paid-in-full client can still show a $1,724 “balance” across stale invoices nobody cancelled.

Questions a call process will raise that look answerable and are not. Saying UNKNOWN is the correct answer to every one — borrowing a nearby number to fill the gap is how most of the corrections on the previous tab happened.

The question	Status	Detail
What will a first-miss phone call actually convert at?	UNKNOWN	It has never been run as a measured process with a held-back control. Every historical second-attempt rate is selected. The only honest answer is UNKNOWN, and the only way to get a number is the control arm.
What will an automated re-charge collect?	UNKNOWN	Nothing has ever re-charged a failed payment here, so there is no baseline at all. `/retry-list` is the first one and it says so on its own front page. The $581 vs $221 figure behind `/debt-plan` action 1 is observational — accounts that happened to get an early second attempt may differ from those that did not, and history cannot rule that out.
What is our right-party contact rate?	NOT MEASURED	`goto_call_pull.js` pulls per-call records every morning and discards the phone number, keeping only per-agent counters. So we know 20 calls a day run past two minutes; we do not know how many reached the person we were ringing. One script change, not a new integration.
Was the customer ever actually told?	CANNOT TELL	Voice calls do not appear in the GHL conversation thread. A card worked hard by phone reads as unchased. So “1,354 silent debtors have had two texts or fewer” is a count of texts, not a count of contact attempts — do not call the low end “never contacted”.
How many first misses are never chased at all?	CEILING ONLY	23.1% of FAILED payments carry no `customer_id` (74.7% of 2023 failures, 11.3% of 2026), so they cannot be joined to anyone. Any “never attempted” or “never chased” figure is a ceiling, not a count. Completed payments are fine — 44 blank of 285,691 — so collected figures are unaffected.
Does outreach-style working lift first payments?	UNDERPOWERED	Against a tenure-matched control: 55.0% vs 56.6% on the first payment — inside the ±2.6pp standard error, and a dead heat on the second. The one real signal is chargebacks at roughly half the rate (1.01 vs 1.88 per 100), ~1.5 SE, suggestive not proven. And only ~a quarter of who was worked can be matched at all, because half the GHL contacts carry neither phone nor email.
Why is Boston 6 points better than everyone else?	UNEXPLAINED	Boston misses the opening instalment 34.3% of the time against 40–47% everywhere else, and it is not a runway artefact. Nobody has looked at what Boston does differently at the point of sale. Worth an afternoon before anyone builds anything.

Every list and report behind this page. A call process does not need a new data pull — the lists already exist; what is missing is the routing of the decline reason into GHL and a queue that starts on day 1.

Report	URL	What it gives you
`/opening-payment`	https://reports.iconicbyai.com/opening-payment	The three outcomes of the opening instalment and what each is worth, age-matched to 365 days. The page this handoff is mostly built on. Off `square_attempts.jsonl`, no live call.
`/instalment-recovery`	https://reports.iconicbyai.com/instalment-recovery	First vs second instalment, the 10-day curve day by day, the 2×2, and the split by decline reason. Row detail in `agreements.csv`.
`/first-payments`	https://reports.iconicbyai.com/first-payments	Three matched Jun–Aug windows across 2024/25/26 — volume, value and failure rate of the first instalment, with the like-for-like estate adjustment.
`/first-miss-nocard.csv`	https://reports.iconicbyai.com/first-miss-nocard.csv?key=studio1-sales-2026	The no-card call list — 100 people, soonest due first. 32 columns of everything we hold on each: contact details, CRM ref, amount and due date, past-due count and value, full payment history, last known issuer and card, whether Square has a reminder set, and a `/pay` link. SALES_KEY-gated (PII) and never committed — gitignored, parked in Supabase. `scripts/reports/report_nocard_list.js`.
`/retry-list`	https://reports.iconicbyai.com/retry-list	The nearest thing to a call list today. 5,011 people whose latest recurring charge failed and has not been collected since, in six cohorts — early-access / normal bank × missed-first-payment / never-paid / on-off. 674 are missed-first-payment, $145,431. Soft declines only; dead and blocked cards excluded. Carries a 10% control arm and a 120-person pilot batch. Columns: name, phone, email, city, amount, decline reason, plain English, issuer, charges to date, CRM ref, Square id.
`/due-next-7`	https://reports.iconicbyai.com/due-next-7	The forward list — everything else looks backwards. Every instalment Square will attempt in the next seven days with the issuing bank, arrears and full payment history. This week: 2,671 payments, $518,843 scheduled, 1,703 already in arrears ($1.81M past due before the charge). 184 rows where nothing will charge them at all. This is where pre-emptive contact lives.
`/decline-trends`	https://reports.iconicbyai.com/decline-trends	Every recurring charge since Jan 2026 bucketed by action, raw reason, plan position, payment history, card type, issuer class and named bank, month by month. The early-access and calendar evidence.
`/decline-buckets`	https://reports.iconicbyai.com/decline-buckets?key=studio1-sales-2026	Soft 82.6% / hard 10.5% / structural 6.9%, flat across every month and every city. 2,052 follow-up charges failed a second time, re-charging $385,210 to cards the issuer had already declared finished.
`/zombie-reasons`	https://reports.iconicbyai.com/zombie-reasons?key=studio1-sales-2026	The hazard curve (miss 1 → 20.8%, miss 2 → 12.9%), the no-card-on-file finding, and why a shut-off is a nightly sweep rather than a switch. `early.csv` is the 2,751 series at 1–3 misses.
`/chase-daily`	https://reports.iconicbyai.com/chase-daily?key=studio1-sales-2026	One CSV a day of everyone who missed that day, for pushing into GHL. `owner:"render"` — never run it from the laptop, it would write the hand-over ledger twice. The ledger in Supabase is the one file here that cannot be rebuilt.
`/missed-chase`	https://reports.iconicbyai.com/missed-chase?key=studio1-sales-2026	1,020 people, $1,254,900 past due — everyone the old ledger cut out before the 3 Sep fix. One deduped row each, biggest debt first.
`/debt-plan`	https://reports.iconicbyai.com/debt-plan	The nine-action collections plan with the evidence under each line, three scopes costed in hours and FTE, the five-part dunning playbook, and four ideas that did not survive the data.
`/debt-status`	https://reports.iconicbyai.com/debt-status?key=studio1-sales-2026	The whole US Debt Collection sub-account: 11,605 accounts, 10,038 owing $16.79M, and the five-rung contact ladder that stops the lifetime reply flag flattering it.
`/pay`	https://reports.iconicbyai.com/pay	The live self-serve payment page — Square Web Payments with lookup, charge and schedule endpoints. Already built, barely used. An actively-entered payment succeeds at 99.4% at any lag; the gap is that the link is not pushed at the moment of failure.

“First payment” means the first RECURRING instalment — never the studio deposit

NOTHING re-charges a failed payment. Every second attempt is a person

Therefore every conversion rate on a second attempt is SELECTED — do not quote one as an expected rate

Lead with the things that did NOT happen — those are the unselected numbers

Age-match everything, or you are measuring the calendar

Split never-paid from stopped-paying before you design anything