Proposal: rebuilding how we collect a missed payment
For the board · prepared 17 September 2026 · every figure is measured from our own
Square and CRM records, with the date and source stated. Nothing in this proposal has been started.
What is being asked for
Approve
Eight decisions — listed in full below
Engineering
3–4 weeks of one developer (estimate — needs the team’s own sizing)
People
No new headcount — the existing five-person collections team re-pointed as 3 + 1 + 1
Timeline
12 weeks to fully live, in four phases
At stake
$1,136,232 a year is missed at the first payment alone — 46.9% of every new agreement
Expected return
$57,000–$118,000 a year from the first-payment work, after halving the observed value for customer selection. Two further items are genuinely unquantifiable today and are marked so.
The one-sentence version: nearly half of all new agreements fail their very first
payment; the customer then hears nothing for a day, receives two text messages a day apart, and is left alone
for a month — while the team who should be calling them averages 34.6 minutes on the phone a day.
The evidence says the first 24 hours are worth more than everything that follows put together, and we are not
in them.
1. The problem, in numbers
New agreements a year
10,943
on the recurring payment book
Fail their first payment
46.9%
5,127 agreements · $1,136,232
Currently owing
14,109
accounts whose last payment attempt failed
First contact after a failure
23.2h
the fastest one all week; median 24.4h
Never reach collections
20%
12 of 60 failures in the week measured
Answer a message in 30 days
8.3%
across the whole collections book
The three findings that justify this proposal.
1. The first day is worth more than the next nine put together. Of everything a missed payment ever
returns within ten days, 48% comes back on day 1, 71% by day 3, 92% by day 7. Today the fastest we
contact anybody is 23.2 hours, because the collections system is fed by a once-a-day spreadsheet. We are
arriving at the end of the most valuable day we have.
2. Catching it early is worth much more than the payment itself — but not all of the difference is ours
to claim. An agreement whose first payment is recovered within ten days goes on to pay $1,165 by day
365; one where it is not pays $243. We are deliberately NOT claiming that $922 gap as the value of
rescuing somebody. Much of it is simply who the customer already was: a person whose payment we can recover
in a few days is, by nature, a better payer than one we cannot, and they would have paid more anyway.
This proposal counts HALF the difference — $461 an agreement — as something the rescue creates,
and treats the rest as the customer, not us. That halving is a judgement, not a measurement, and it is the
single most important assumption in the financial case. It is still a large number.
3. We are not failing to collect. We are failing to ask. In the week measured, of the people who still
had not paid: 22 were sent one message and never followed up, 16 replied to us and nothing happened at all,
and 7 were never contacted by anyone. Two of the sixteen were explicitly waiting on us — one asking for a
payment link, one telling us they had already paid.
2. What happens today
Every step below is measured against last week’s 60 failed first payments
(7–13 September), traced through our own collections system. This is not a characterisation. It is the
whole process, end to end.
Hour 0The payment fails, and that is the last attempt of the month
- Our payment provider attempts the card once, on the due date, and never again — measured at 0.90 failed attempts per missed cycle across the book.
- Nothing further is attempted on that account until next month’s invoice falls due. There is no second attempt, no retry schedule, nothing automatic. If money arrives, a person made it arrive.
Hours 0–23Nothing happens at all
- The collections system is fed by a once-a-day spreadsheet drop, so the customer cannot be contacted until the next run.
- Across all 60 failures, the fastest first contact was 23.2 hours. The median was 24.4 hours. Nothing went out inside twelve hours, all week.
- Set against the recovery curve, this is the expensive part: 48% of everything a missed payment ever returns comes back on day 1. We arrive at the end of it.
Day 1A text — for most, but not all
- 12 of the 60 never reached the collections system at all (20%). No workflow can see them, so no text was ever possible.
- 17 of the 60 were never texted — 71.7% got a message, 28.3% got nothing.
- The message is identical whatever the bank said. The decline reason never reaches the system that sends it, so a customer whose bank has simply blocked recurring payments gets the same words as one with no money.
Day 2A reminder — and then it stops
- The median customer receives exactly two texts. 25 of the 43 who were contacted got two and no more.
- The median span from first message to last is 1.0 days. 39 of the 43 had every contact they will ever get inside two days.
- After that the account goes quiet until the next invoice fails a month later, by which point the chance of the next payment succeeding has almost halved.
If they replyRoughly a third do — and we do not handle it properly
- 19 people replied — 44.2% of those texted, 31.7% of all 60 failures. That is a good response rate to a debt message and it is being wasted.
- 6 of those 19 got no answer back from us at all. They wrote to us about a missed payment and the conversation stopped there.
- Among those still unpaid, the pattern is worse: 16 had replied and no money followed. Two were plainly waiting on us — one asking for a payment link, one saying they had already paid.
PhoneWe cannot tell — and that is itself the finding
- We cannot say how many times any of these 60 customers was rung, because nothing records it. Our phone system reports call volumes per agent per day and nothing else — no customer, no number. It does not write back into the collections record either, so the account shows nothing.
- Do not read that as “nobody rang them”. The team made 5,709 calls in 30 days and those calls reached somebody. Some may well have been these people. We simply have no way to know.
- What we can say is about the team as a whole: over the same period the six collections lines averaged 34.6 minutes of talk time per person per day, the busiest made 58 outbound calls a day, one line recorded no activity at all, and between all of them they produced 20.3 conversations of two minutes or more per day.
- Being unable to answer “has this customer been called?” is a serious gap in its own right. It means no manager can check whether a queue was worked, and no measurement of any of this can separate a call that happened from one that did not. Closing it is a Phase 0 item.
Put plainly: a customer misses their first payment, hears nothing for a day, gets two
text messages a day apart, and is then left alone for a month. A third of them write back, and one in three
of those gets no reply. Meanwhile the collections team averages 34.6 minutes of talk time a day and says
it is not busy — and we cannot even establish whether any of these particular customers was ever called.
This is not a collections process that is underperforming. It is the absence of one.
| | Today | Proposed |
|---|
| First contact | 23.2 hours at best | Within 2 hours, worded by what the bank actually said |
| Who gets contacted | 71.7% — a fifth never even reach the system | 100%, reconciled daily |
| Messages sent | 2, one day apart | A planned sequence to day 30, and a text after every call attempt |
| How long contact lasts | 1 day | 30 days, to the next invoice |
| Phone calls | Unknown — nothing records who was called | 2 a day on days 1–3, one on day 5, every attempt logged against the account |
| If they reply | 1 in 3 gets no answer | Named owner, answered the same working day |
| Staff | 34.6 minutes of talk time a day | Same five people, re-pointed |
3. What we are proposing
One contact process with three entry points, replacing what is currently a single
once-a-day list.
The evidence for treating it as one process is strong and it was a surprise: the reason the bank gives for
refusing the payment is almost identical across all three groups — roughly 85% is “not enough
money / over a limit” whether the customer is brand new, three weeks in, or a year lapsed. The groups
do not differ by why the card failed. They differ by how long ago anybody asked, and whether anybody
asked at all.
| Group | Size | Money | What we would do differently |
|---|
New agreements
first payment fails |
5,127
a year | $1,136,232 |
Prevent what we can at the till, message before the payment is taken, then contact within hours rather than a day. This is where the money is. |
First two months
missing payments early on |
849
of 1,599 live | $254,182 |
One message, placed in front of the second missed payment. After one miss the next payment succeeds 44% of the time; after two, 23.5%. Nothing sits in that gap today. |
Long-lapsed
four or more failures running |
9,801 | $14,601,622 |
Stop charging the dead ones, then a priced settlement offer to the 1,469 who have paid us before. Not a calling list — it is the largest number here and the worst use of an hour. |
The biggest number is the worst place to spend the effort, and the board should expect
that to feel counter-intuitive. The long-lapsed group is $14,601,622 on paper. But $6,691,378
of it belongs to 4,687 accounts that never paid us a single cent, and 3,936 of the group last paid
something over 240 days ago. The new-agreement group is a fraction of the size on paper and worth far more per
hour of effort, because of the $922 difference above.
One genuinely encouraging finding. Of the 11,622 accounts that have ever gone four
or more payments down, 2,286 — 19.7% — later paid us again, contributing $1,024,771
after they had lapsed, an average of $448 each. A lapsed account is not a dead one. Every one of those
recoveries happened because a person made contact: nothing in our systems ever attempts a card a second time
by itself.
4. What has to change, in order
Phase 0Weeks 1–4Plumbing. Nothing customer-facing changes.
- Event-driven feed into the collections system, replacing the once-a-day spreadsheet drop
- Carry the bank’s decline reason through, so a message can say something true about why the payment failed
- Daily reconciliation so nobody falls out of the process unnoticed
- A named owner and a same-working-day answer for anyone who replies
- Fix the reply-rate measurement (phone reactions are currently counted as replies)
- Log calls against the customer record — today the phone system reports totals per agent per day and nothing else, so nobody can tell whether a given account has ever been rung
Who: Engineering · Effect: No revenue effect on its own. Everything after this depends on it.
Phase 1Weeks 3–6Stop the bleed, and build the tools the calls need.
- Nightly sweep to stop charging the 2,867 dead agreements
- Apply the payment-portal credit that is promised on screen today and never applied — blocking for anything that sends the link
- The one-button follow-up text, wired to the call record
- Point the existing notification code at the collections account it was always meant to use
Who: Engineering + Finance · Effect: Cost and risk reduction, not new revenue. Should pay for itself in avoided chargebacks.
Phase 2Weeks 5–10The first ten days after a missed payment.
- Same-day message, routed by what the bank actually said, carrying the payment link and its credit
- Two calls a day on days 1, 2 and 3, one on day 5 — the highest-value hours on the book
- A one-button “we tried to reach you” text after every single call attempt
- A day-7 message that offers three replies, one of which is “not this month”
- A day-21 warning in front of the second missed payment
- Re-point the team 3 + 1 + 1 and hand each role a ready-made list each morning
Who: Collections + Engineering · Effect: This is where the money is. See the financial case.
Phase 3Weeks 8–12Prevention, and the back book.
- Pre-payment messages at day −5 and day −1, with the 10% control group
- Card-quality step at the point of sale
- Priced settlement campaign on the 1,469 lapsed accounts that have paid us before
Who: Sales + Collections · Effect: Prevention compounds; the settlement campaign is one-off.
Phase 0 is not optional and cannot be skipped to get to the revenue faster. If we
begin sending more messages before the reply handling and the reconciliation are fixed, we will increase the
number of customers who contact us and get no answer. That is measurably worse than the current position,
because it spends goodwill we cannot convert. In the week measured, sixteen people replied to a message
about a missed payment and nothing came back.
5. Why the payments actually fail — and the one lever that costs nothing
We used to think most first-payment failures were a technical problem. They are not.
Roughly 85% of them are the bank saying there is not enough money, and that share is almost identical
whether the customer is brand new, three weeks in, or a year lapsed. But who the customer banks with
predicts failure better than almost anything else we hold — and unlike affordability, that is something we
can influence, at the till, for nothing.
What the bank tells us when it refuses
| What the bank said | Share | What it actually means |
|---|
| Over a limit on the account | 55% | A money code, not a technical one. It recovers at the same rate as “insufficient funds”, and when we ask customers to pick a reason themselves, almost all choose “money is a bit tight”. |
| Refused, no reason given | 23% | The issuer told us nothing. A conversation is the only way to find out. |
| Not enough money | 10% | Exactly what it says. |
| Bank has blocked recurring payments on the card | 7% | The only code with a thirty-second customer-side fix — a toggle in their banking app. Nobody is ever told this, because the reason never reaches the people making contact. |
| Card expired, wrong details, or the bank refused the instalment | 5% | Genuinely technical. Needs new card details. |
A claim we previously made and have withdrawn — the board should hear it from us
rather than from somewhere else. We used to say “87% of customers who miss their first payment are not
broke”, because the biggest code is a limit rather than insufficient funds. That was wrong. Checked
properly, a limit decline recovers at 20.9% within 30 days and an insufficient-funds decline at 21.3% —
indistinguishable. The big code is a money code. This proposal is built on affordability and timing, not
on telling customers their bank made a mistake.
The finding that matters most: which bank the card comes from
Measured across 3,733 first instalments between 1 May and 31 August 2026, matched to the
issuing bank. “App bank” means the sponsor banks behind the app-based accounts that release wages up
to two days early — Cash App, Chime, SoFi, Varo, Dave and similar.
| Card is from | First instalments | Failed | Share of ALL failures |
|---|
| An early-access app bank | 694 (18.6%) | 81.1% | 31.6% |
| Any other bank | 3,039 (81.4%) | 40.1% | 68.4% |
Under a fifth of our new customers, and nearly a third of everything that fails. They
also recover worse afterwards — 18.8% are back within ten days against 25.2%, and 28.3% ever pay against
43.3%. Failing more often and recovering less compounds.
It is the bank, not the prepaid card — we checked
| Card | First instalments | Failed |
|---|
| App bank, ordinary card | 380 | 76.1% |
| App bank, prepaid card | 314 | 87.3% |
| High-street bank, ordinary card | 2,984 | 39.7% |
| High-street bank, prepaid card | 55 | 61.8% |
An ordinary, non-prepaid card from an app bank still fails 76.1% against
39.7% for the same kind of card from a high-street bank. Prepaid adds roughly ten points on top of that,
and about twenty points at a traditional bank — a second, smaller effect. So this is not a
“prepaid cards are bad” finding. It is about the account behind the card.
A second, separate signal: credit beats debit
| Among high-street banks only | First instalments | Failed |
|---|
| Credit card | 786 | 27.2% |
| Debit card | 2,251 | 44.7% |
Compared like with like, with the app banks stripped out, so this is not the same finding
twice. The best card we can be handed — a high-street credit card — fails 27.2%. The worst
— an app-bank debit card — fails 82.5%. A threefold difference, settled in the ninety seconds when
the card is taken.
The honest caveat, and it is the same one that halved the financial case. Somebody
who offers a high-street credit card is, on average, in a better financial position than somebody offering an
app-bank debit card. Asking for a different card does not turn one customer into the other. The real and
smaller opportunity is the customers who hold both and hand over whichever is nearest — for them,
taking the better card is free. This is why decision 7 asks for a trial and a measurement, not a
rollout.
Named banks, so the sales floor knows what it is looking at
| Issuing bank | First instalments | Failed |
|---|
| Sutton Bank — Cash App, Chime | 308 | 87.3% |
| The Bancorp Bank — various app accounts | 112 | 76.8% |
| Stride Bank — Chime | 99 | 74.7% |
| Navy Federal Credit Union | 140 | 55.7% |
| Wells Fargo | 220 | 44.5% |
| Bank of America | 377 | 40.1% |
| JPMorgan Chase — debit | 382 | 36.6% |
| JPMorgan Chase — credit | 124 | 6.5% |
One theory we tested and had to drop — do not move charge dates on it. The obvious
explanation for the app-bank failures is that those customers are paid early, run out early, and should
therefore be charged on a different day. The data does not support it. Across 78,016 charges both groups
peak on the same day of the week and trough on the same day. What differs is the size of the swing
— 1.78 times best-to-worst against 1.14 — so these accounts run empty faster, they are not on a
different clock. Moving their charge date would cost effort and change nothing.
6. The operating model — who does what, and does it fit
No new headcount. The five collections people are re-pointed, three ways. The evidence
for this is the team’s own call records: over 30 working days to 16 September they averaged
34.6 minutes of talk time per person per day, the busiest made 58 outbound calls a day, and one
line recorded no activity at all. Between all of them the team produced 20.3 conversations of two
minutes or more per day. The hours exist. They are pointed at the oldest, coldest accounts on the book.
| People | On what | Volume | Why they sit there |
|---|
| 3 | The back book
everything four or more payments down |
9,801 accounts | The existing job, plus the settlement campaign. Largest pile, lowest return per hour — so it gets the people, not the priority. |
| 1 | Missed first payments
the first ten days |
~100 a week | The highest-value hour on the book. Needs one person whose whole job is speed, because the value halves within days. |
| 1 | Irregular missers
first two months, missing here and there |
~119 a week | Catching the second missed payment before it happens. After one miss the next payment succeeds 44% of the time; after two, 23.5%. |
The contact rhythm — two calls a day, and nobody is left wondering
Every call attempt is followed by a text. Not typed by hand — one button in the
system that sends “we tried to reach you” with the customer’s own payment link on it. A
missed call on an unknown number is nothing; a missed call plus a text is a message.
| When | What happens | By whom |
|---|
| Day 0, within 2 hours | Text, worded by what the bank actually said, carrying the payment link and the credit | Automated |
| Day 1 | 2 calls, each followed by a one-button text | Person |
| Day 2 | 2 calls + texts | Person |
| Day 3 | 2 calls + texts | Person |
| Day 5 | 1 call + text | Person |
| Day 7 | Three-button message — pay / arrange / not this month | Automated |
| Days 10, 14, 21, 25, 30 | Arrangement offer, check-in, the warning before the next payment, pre-notice, next payment falls due | Automated |
Does one person actually fit 100 a week at that rhythm? Yes — and the arithmetic
is shown rather than asserted.
20 arrive per working day. Running two calls on days 1–3 and one on day 5 means about 80 accounts live
at any moment and 140 dial attempts a day. At the team’s own measured outbound contact rate of
9.1% — not the 20% a textbook would assume — that is 12.7 real conversations a day.
Costed at 1.5 minutes a dial, 9 minutes a conversation, 3 minutes of notes and 15 seconds for a one-button text,
the day comes to 6.6 hours. It fits.
What the board should know: 140 dials a day is roughly two and a half times what the busiest
collector does today. The hours are there, but this is a real change in how the day is worked, and it only
holds if the list is handed to them ready — which is what Phase 0 buys.
The irregular-misser role is lighter and that is deliberate. About 71 dials a day,
6.5 conversations, 3.4 hours. The remaining time is the buffer for the first-payment role on a heavy week
— arrivals swing from 43 to 144 a week, so a rota built on the average is underwater one week in four.
7. What the board is being asked to decide
1. Engineering time for five plumbing fixes
Approve
Three of them are blocking: nothing else in this proposal reaches its stated value without them. Estimated 3–4 weeks of one developer, and that estimate needs the team’s own sizing before it is relied on.
Why: Without this the first contact after a failed payment cannot leave inside 23 hours, and day 1 is 48% of everything a missed payment ever returns.
2. Re-point the collections team — 3 + 1 + 1, no new headcount
Approve
Three people stay on the existing back book. One takes new missed first payments (about 100 a week). One takes irregular missers — customers in their first two months who miss here and there. The capacity is already in the building; it is pointed at the wrong things.
Why: Measured over 30 working days to 16 September, the six collections lines averaged 34.6 minutes of talk time per person per day, and the busiest made 58 outbound calls a day. One line recorded no activity at all across the whole period. The team produces 20.3 conversations of two minutes or more per day between all of them. This is not a team without hours.
3. The payment link, the credit on it, and a follow-up text after every call
Approve and fund the build
Every call attempt is followed by a text saying we tried to reach them, sent with one button from the system rather than typed. The text carries the customer’s own payment link with the bonus credit already applied — pay in full and 15% is credited to the balance, 75% earns 10%, 50% earns 5%.
Why: The portal already exists and already shows this offer. It has never once applied it. The credit is written into a note and nothing else — no ledger entry, no credit, no change to what the customer owes. We checked all 83,623 recurring payments taken in 2026: not one carries it. Nobody has been short-changed yet because no traffic has been sent there, but this must be fixed before a single link goes out.
4. A settlement policy with a hard ceiling
Approve the policy
Settlement opens at day 40, capped at 25% of balance. Anything deeper needs a named approver and only after day 60. Nothing before day 31 is discountable at all.
Why: Measured on our own book: a reduction under 25% returned $0.74 per $1 given away; at 50–75% it returned $0.14; clearing a balance outright returned −$0.03. And 600 of the 945 accounts we have reduced had no failed charge at all — 72% of the money given away went to people who were never struggling.
5. Stop charging agreements that cannot pay
Approve
A nightly sweep to stop attempting 2,867 accounts whose cards have failed four or more times running. Square cannot end a recurring series through its API, so this has to be a sweep rather than a switch.
Why: 3,100 failed charges were fired at these accounts in the last 30 days and $611,841 a cycle is being attempted into the void. This is a chargeback and processing exposure, and it is also a revenue-recognition question the board should see rather than an operational one.
6. Contact customers BEFORE the first payment
Approve in principle
A short service message five days before and one day before the first instalment. Nobody sends anything today — not us, and not Square. No pay-now link, because a duplicate payment on top of a stored card is the fastest route to a chargeback there is.
Why: This is the only step in the whole process that reaches a customer while nothing has gone wrong. Its effect is UNKNOWN and could be negative, which is why decision 8 matters.
7. A card-quality step at the point of sale
Approve a TRIAL
Where the card offered is an app-bank card or a debit card, ask whether they have another — a high-street card, or a credit card. One sentence in the sales conversation. Trial it and measure it; do not roll it out on the raw numbers alone.
Why: App-bank cards are 18.6% of first instalments and 31.6% of every failure, failing 81.1% against 40.1%. Separately, among high-street banks alone, credit fails 27.2% against debit 44.7%. Best card to worst is 27.2% against 82.5%. It costs nothing to ask — but it changes the sales conversation, and part of the gap is the customer rather than the card, which is why this asks for a trial.
8. Hold back a control group
Approve — and this is the uncomfortable one
Deliberately send 10% of customers nothing at the new steps, so the effect can be measured against a like-for-like group.
Why: We have never run an automated pre-notice or an automated warning message, so their conversion rates do not exist. Without a control they get measured against a human conversation rate and will look like failures however well they do. The cost is real: a small number of people who might have paid will not be contacted.
8. The financial case
| Line | Per year | The assumption behind it |
|---|
First-payment recovery improves by 5 percentage points
from 26.4% to 31.4% recovered within 30 days |
$57,000–$118,000 |
5,127 failures a year × 5 points = 256 more agreements rescued. The lower figure values each at the missed payment alone (about $222). The upper values each at $461 — half the $922 whole-agreement difference we can observe, because much of that difference is the customer rather than the rescue: somebody we can recover quickly was always the better payer. Halving it is a judgement, and it is the assumption most worth arguing about. Both ends depend entirely on Phase 0. |
| Second-miss warning message | not quantified |
Deliberately left blank. 353 agreements are sitting one payment down right now and the success rate halves at the second. But we have never sent this message, so its effect is unknown, and borrowing a rate from our collectors’ results would measure a human conversation rather than an automated message. Decision 8 is how we find out. |
| Stop charging dead agreements | cost, not revenue |
3,100 failed attempts in 30 days against cards that cannot succeed. Saves processing fees and, more importantly, reduces chargeback exposure — $385,210 was put at already-dead cards in a single quarter. |
| Settlement campaign on lapsed payers | ~$130,000 pool |
1,469 accounts × the measured 19.7% return rate × $448 average. Read this as the size of the pool, not as new money — 19.7% is what the book already achieves with the effort already spent. Whether a priced offer beats that is untested. |
| Cost | No new headcount
+ 3–4 dev weeks |
The people are already employed — this is a re-pointing, not a hire, and the capacity check in section 6 shows the two new roles fit inside a working day at the team’s own measured contact rate. The only new cash cost is the engineering. The board should note the opposite risk: if the back book is left with three people rather than five, the recovery on the oldest accounts will fall, and that reduction is not modelled here. |
9. What this proposal does not claim
It does not claim a return on the two new automated messages. We have never sent a
pre-payment notice or a second-miss warning. Their effect is unknown, and could be negative —
telling somebody a payment is coming also gives them the opportunity to move the money. That is why a control
group is decision 7 rather than an afterthought.
It does not claim the whole observed value of an early rescue. Customers whose
first payment we recover quickly go on to pay far more than those we cannot — but they are also, by
nature, better payers, and would have paid more regardless. We have halved the observed gap to keep only
what is plausibly ours. If the board thinks that is still generous, the lower end of the range assumes we
recover nothing but the missed payment itself, and the case is positive there too.
It does not claim that contacting people causes recovery. Customers who pay
immediately never get chased, so any comparison of “contacted” against “not contacted”
is distorted at both ends. We are proposing this because of the timing evidence, which is not
distorted in that way, not because contacted customers pay more often.
It does not assume anything recovers itself. Our payment provider attempts a card
once, on the due date, and never again — measured at 0.90 attempts per missed cycle. There is no
automatic second attempt anywhere in this business. Every pound in this proposal arrives because a person
asked for it, which is why the headcount line is not optional.
It cannot say whether any individual customer has been telephoned. Our phone system
reports call counts per agent per day and nothing else — no customer, no number — and it does
not write back into the collections record. So every per-customer contact figure in this paper covers text
messages only, and nothing here should be read as evidence that a particular person was or was not called.
The team-level call figures are sound; the per-account ones do not exist. Closing that is a Phase 0 item, and
until it is closed none of the measures in section 10 can separate a call that happened from one that did
not.
10. How we will know whether it worked
| Measure | Today | What good looks like |
|---|
| Time from failed payment to first contact | 23.2 hours at best | Under 2 hours for the automated step; a call attempt inside 48 hours |
| Share of failures that enter the process at all | 80% | 100%, reconciled daily with an alarm on the gap |
| Customers who reply and then receive an answer | Not measured, and 16 went unanswered in one week | 100% answered the same working day |
| First payment recovered within 30 days | 26.4% | 31.4% — the basis of the financial case |
| Answering a message within 30 days | 8.3% | Rising — and measured excluding phone reactions, which currently inflate it |
| Second missed payment after a first | 56% go on to miss again | Falling — measured against the held-back control group |
| Call attempts per person per day | 58 at the busiest line; 34.6 minutes of talk time | ~140 on the first-payment role, with the follow-up text sent on every one |
| Calls attributable to a customer | None — the phone system records per agent per day only | Every attempt logged against the account, so a queue can be audited |
| Call attempts that get a follow-up text | None — there is no such step today | 100%, one button, logged against the account |
| Payment-link credit actually applied | 0 of 83,623 payments in 2026 | Every payment made through the link |
A note on how these numbers were produced. Everything here comes from our own payment
records and collections system, measured between 1 and 17 September 2026, and each figure is reproducible from
the detail behind this page. Where two ways of counting the same thing disagree, both are shown in the
underlying analysis rather than averaged. Nothing in this proposal has been implemented, no customer has been
contacted as part of it, and no system has been changed.
Supporting detail is on the other tabs of this page: the week
that was measured, whether anyone chased those customers, the conversations verbatim, the full three-group
analysis, and a worked calling list of 90 people. Prepared 2026-09-17.