No — they have gone up. 5,140 failed charges in Jan 2026 to 9,343 in Aug 2026 (+81.8%). Part of that is a bigger book — active accounts grew from 11,691 to 13,670 (+16.9%) — but not all of it: the share of charge attempts that fail rose from 37.5% to 42.2%. More accounts, and a higher proportion of them missing.
Yes — but not for the reason you would expect. The 30-day recovery rate fell from 26.4% in Jan 2026 to 17.7% in Aug 2026 — -32.7% in six months. The obvious explanation would be that the pool has aged — that we have worked through the easy ones and what is left is chronic. That is not what happened. First-time missers held at 16.5%–11.4% of all misses across the whole window. What changed is that the same kind of debtor pays back less often: repeat missers went from 18.9% to 13.9%, first-timers from 43.7% to 45.7%. The fruit is not higher up the tree — there is less of it on the same branches.
The chase itself still works. Whoever does come back comes back just as fast as ever — median 0.8 days, 52.3% of recoveries land same-day or next-day. Nothing has slowed down. There are simply fewer people responding at all.
Yes — but only in the pile we keep re-texting. The fresh work has not changed at all. Split every month by whether it was the person's first month in the account or one we were re-chasing, and the two lines go opposite ways. A conversation in its first month has answered at 51.4%–57.1% every month since Feb 2026 — 54.6% in Aug 2026, against 52.6% back in Oct 2025. It has not decayed. The carried-over pile went the other way, 37.1% → 12.8%, and it grew from 280 people in Oct 2025 to 3,308. So the blended rate falling 48.4% → 21.1% is a mix effect, not a decline. Nothing about either group changed; we simply put a far bigger dead weight into the denominator. Intake is flat too — 819 new debtors reached the account in Aug 2026, inside the 712–842 band it has held all year. The real answer to "are the staff quieter": 447 people replied out of a first month in Aug 2026 against 398 in Oct 2025 — that is the answerable work, and it is the same size it has always been. What fell is the re-chase: it peaked at 834 replies in Jan 2026 and is 425 now, out of a pile that is no smaller (3,489 people then, 3,308 now). Delivery is not the problem: 97.4% of outbound is still delivered by the carrier, so these are landing and being ignored, not blocked. Full table on the Texts tab.
| Measure | Jan 2026 | Aug 2026 | Change | What it means |
|---|---|---|---|---|
| Failed charges | 5,140 | 9,343 | +81.8% | More misses landing on the team every month |
| Active accounts charged | 11,691 | 13,670 | +16.9% | The book grew — some of the rise is just scale |
| Miss rate (per charge attempt) | 37.5% | 42.2% | +12.6% | Scale-adjusted — this is the real deterioration |
| Recovered within 30 days | 26.4% | 17.7% | -32.7% | The headline: the chase converts far less often |
| Of the money missed | 38.3% | 32.9% | -14.0% | Dollars hold up better than people — bigger misses still get cured |
| Median days to recover | 1.2 | 0.8 | flat | Speed is unchanged — it is conversion that fell, not pace |
A raw miss count goes up when the book grows. The miss rate is the one to read. The newest month is part-built and is marked as such.
| Month | Charge attempts | Active accounts | Missed | Miss rate | $ missed | $ taken |
|---|---|---|---|---|---|---|
| Jan 2026 | 13,704 | 11,691 | 5,140 | 37.5% | $1,468,217 | $2,018,138 |
| Feb 2026 | 13,881 | 11,842 | 5,270 | 38.0% | $1,449,444 | $2,094,442 |
| Mar 2026 | 14,480 | 12,324 | 5,646 | 39.0% | $1,608,102 | $2,089,411 |
| Apr 2026 | 14,286 | 12,376 | 5,844 | 40.9% | $1,543,854 | $1,958,986 |
| May 2026 | 14,791 | 12,874 | 6,124 | 41.4% | $1,671,095 | $2,275,126 |
| Jun 2026 | 15,553 | 13,583 | 6,421 | 41.3% | $1,716,584 | $2,232,290 |
| Jul 2026 | 15,567 | 13,675 | 6,503 | 41.8% | $1,722,619 | $2,274,370 |
| Aug 2026 | 22,117 | 13,670 | 9,343 | 42.2% | $2,513,692 | $3,059,730 |
| Sep 2026 part month | 5,415 | 3,631 | 2,087 | 38.5% | $521,963 | $716,059 |
A miss counts as recovered when that same customer's next completed charge lands inside the window. A month only appears once its window has closed — a miss from last Tuesday cannot have had 30 days, and counting it would invent a decline.
| Month | Missed | Window closed | Paid ≤7d | Paid ≤30d | $ recovered | % of $ | Median days | Same/next day |
|---|---|---|---|---|---|---|---|---|
| Jan 2026 | 5,140 | 5,140 | 19.7% | 26.4% | $562,390 | 38.3% | 1.2 | 45.4% |
| Feb 2026 | 5,270 | 5,270 | 19.7% | 30.2% | $519,396 | 35.8% | 2.2 | 37.6% |
| Mar 2026 | 5,646 | 5,646 | 18.6% | 23.7% | $557,124 | 34.6% | 1.0 | 51.7% |
| Apr 2026 | 5,844 | 5,844 | 16.4% | 23.0% | $495,707 | 32.1% | 1.1 | 46.1% |
| May 2026 | 6,124 | 6,124 | 15.8% | 19.6% | $582,781 | 34.9% | 0.7 | 52.4% |
| Jun 2026 | 6,421 | 6,421 | 15.9% | 21.8% | $571,291 | 33.3% | 1.1 | 47.9% |
| Jul 2026 | 6,503 | 6,503 | 13.9% | 17.2% | $479,913 | 27.9% | 0.9 | 50.8% |
| Aug 2026 | 9,343 | 4,193 | 12.8% | 17.7% | $409,751 | 32.9% | 0.8 | 52.3% |
| Sep 2026 | 2,087 | 0 | 10.5% | — | $0 | — | — | — |
"Repeat" means this customer already had a failed charge in the 90 days before this one. A fixed 90-day window, not "have they ever missed" — an all-history lookback would make January look full of first-timers and July full of chronics purely because January had no history behind it. That fixed window is why this table starts at Apr 2026.
| Month | First-time missers | They pay ≤30d | Repeat missers | They pay ≤30d | Blended |
|---|---|---|---|---|---|
| Apr 2026 | 966 16.5% | 43.7% | 4,878 83.5% | 18.9% | 23.0% |
| May 2026 | 983 16.1% | 43.4% | 5,141 83.9% | 15.0% | 19.6% |
| Jun 2026 | 1,081 16.8% | 47.1% | 5,340 83.2% | 16.7% | 21.8% |
| Jul 2026 | 1,008 15.5% | 40.4% | 5,495 84.5% | 12.9% | 17.2% |
| Aug 2026 | 1,064 11.4% | 45.7% | 8,279 88.6% | 13.9% | 17.7% |
| Sep 2026 | 222 10.6% | — | 1,865 89.4% | — | — |
Read the two share columns first. If the first-time share were collapsing, the pool would be ageing and the fall in recovery would be a mix effect — we would have worked through the easy ones. It is not collapsing. Both segments' own recovery rates are falling, which is a harder problem than an ageing pool: it is not who is missing that changed, it is what they do when we chase them.
Two different rates, because they answer different questions. Per message is chase efficiency — of the texts we sent, how many got a text back. Per person is whether anyone is home — of the people we texted that week, how many answered at all. The second is always higher and is the one worth quoting. Weeks with fewer than 25 outbound texts are dropped as noise, and the current part-week is excluded.
| Week beginning | Texts sent | Replies | Reply rate (per message) | People texted | People who replied | Reply rate (per person) | Delivered |
|---|---|---|---|---|---|---|---|
| Jun 15 | 2,375 | 580 | 24.4% | 1,516 | 275 | 18.1% | 98.3% |
| Jun 22 | 1,997 | 492 | 24.6% | 1,312 | 223 | 17.0% | 98.0% |
| Jun 29 | 2,465 | 628 | 25.5% | 1,679 | 301 | 17.9% | 98.3% |
| Jul 6 | 1,770 | 547 | 30.9% | 1,110 | 197 | 17.7% | 98.6% |
| Jul 13 | 1,695 | 491 | 29.0% | 1,065 | 226 | 21.2% | 98.1% |
| Jul 20 | 2,216 | 562 | 25.4% | 1,505 | 261 | 17.3% | 98.4% |
| Jul 27 | 1,875 | 446 | 23.8% | 1,152 | 226 | 19.6% | 98.1% |
| Aug 3 | 2,384 | 558 | 23.4% | 1,496 | 266 | 17.8% | 98.2% |
| Aug 10 | 1,600 | 412 | 25.8% | 965 | 187 | 19.4% | 97.5% |
| Aug 17 | 2,678 | 679 | 25.4% | 1,454 | 301 | 20.7% | 95.7% |
| Aug 24 | 1,808 | 384 | 21.2% | 1,126 | 199 | 17.7% | 97.0% |
| Aug 31 | 2,409 | 525 | 21.8% | 1,500 | 247 | 16.5% | 96.4% |
| Sep 7 | 1,126 | 253 | 22.5% | 743 | 120 | 16.2% | 97.4% |
Watch the Delivered column alongside the reply rate. If delivery is holding near 100% and replies are falling, the texts are arriving and being ignored — a message and cadence problem. If delivery drops, it is carrier filtering or dead numbers, which is a completely different fix.
"Contacted within 24 hours" is not a usable metric here. The outbound is
automated, so reach is close to 100% by construction and moves for nobody. What varies is how much
human effort goes in, and whether anyone engages at all. GHL marks each outbound message
workflow (the automation fired it) or app (a person typed it) — so the two
can be told apart. Do not use userId for this: workflow messages carry the assigned
user's id too, which reads as 74% human when the true figure is nearer 28%.
| Week beginning | Automated | Sent by a person | Human share | People given a human touch | Calls logged | Call answered |
|---|---|---|---|---|---|---|
| Jun 15 | 1,596 | 778 | 32.8% | 502 33.1% | 42 | 7.1% |
| Jun 22 | 1,347 | 650 | 32.5% | 441 33.6% | 41 | 9.8% |
| Jun 29 | 1,853 | 612 | 24.8% | 408 24.3% | 46 | 4.3% |
| Jul 6 | 1,084 | 685 | 38.7% | 471 42.4% | 24 | 8.3% |
| Jul 13 | 980 | 715 | 42.2% | 477 44.8% | 29 | 3.4% |
| Jul 20 | 1,305 | 911 | 41.1% | 651 43.3% | 29 | 13.8% |
| Jul 27 | 1,078 | 797 | 42.5% | 575 49.9% | 34 | 5.9% |
| Aug 3 | 1,537 | 847 | 35.5% | 636 42.5% | 43 | 4.7% |
| Aug 10 | 682 | 918 | 57.4% | 641 66.4% | 15 | 13.3% |
| Aug 17 | 1,681 | 997 | 37.2% | 677 46.6% | 32 | 6.3% |
| Aug 24 | 995 | 813 | 45.0% | 596 52.9% | 33 | 9.1% |
| Aug 31 | 1,354 | 1,055 | 43.8% | 726 48.4% | 35 | 2.9% |
| Sep 7 | 525 | 601 | 53.4% | 428 57.6% | 19 | 0.0% |
"People given a human touch" is the number to manage — the share of everyone texted that week who got a message an actual person wrote. That is the effort metric the automated reach figure was pretending to be. Calls logged in GHL are only the ones dialled from inside it; the bulk run through GoTo, whose puller currently discards the phone number and keeps per-agent totals only.
Every person we texted in a month is in exactly one of two groups: this was their first month in the account (a new missed payment arriving), or we were re-chasing someone who first appeared earlier. Read the two rate columns before the blended one. The blended rate is the only number that fell, and it fell because the second group grew — not because anybody got quieter.
| Month | New in the account | First month: texted | …replied | …rate | Carried over: texted | …replied | …rate | Blended |
|---|---|---|---|---|---|---|---|---|
| Sep 2025 | 1,118 | 1,118 | 456 | 40.8% | 0 | 0 | — | 40.8% |
| Oct 2025 | 757 | 757 | 398 | 52.6% | 280 | 104 | 37.1% | 48.4% |
| Nov 2025 | 1,910 | 1,909 | 877 | 45.9% | 288 | 82 | 28.5% | 43.7% |
| Dec 2025 | 1,514 | 1,514 | 865 | 57.1% | 1,816 | 324 | 17.8% | 35.7% |
| Jan 2026 | 1,158 | 1,158 | 518 | 44.7% | 3,489 | 834 | 23.9% | 29.1% |
| Feb 2026 | 712 | 712 | 405 | 56.9% | 3,462 | 689 | 19.9% | 26.2% |
| Mar 2026 | 720 | 716 | 368 | 51.4% | 3,843 | 662 | 17.2% | 22.6% |
| Apr 2026 | 817 | 815 | 463 | 56.8% | 3,877 | 534 | 13.8% | 21.2% |
| May 2026 | 775 | 770 | 440 | 57.1% | 4,168 | 564 | 13.5% | 20.3% |
| Jun 2026 | 842 | 841 | 462 | 54.9% | 4,448 | 529 | 11.9% | 18.7% |
| Jul 2026 | 718 | 715 | 403 | 56.4% | 4,221 | 547 | 13.0% | 19.2% |
| Aug 2026 | 819 | 818 | 447 | 54.6% | 3,308 | 425 | 12.8% | 21.1% |
| Sep 2026 | 331 | 330 | 177 | 53.6% | 2,198 | 246 | 11.2% | 16.7% |
A conversation is dated by its first text, so the group a person sits in never changes retrospectively. The 67 conversations that hit GHL's 100-message API cap are dated by the conversation's own created-at instead, because their message tail is truncated and their cached "first" text would read too recent. The newest month is part-built and its first-month rate reads low until it closes — a debtor who arrived last week has had days, not a month, to answer.
| Month | Texts sent | Replies | Per message | People texted | Per person | Delivered |
|---|---|---|---|---|---|---|
| Sep 2025 | 2,622 | 1,023 | 39.0% | 1,118 | 40.8% | 95.3% |
| Oct 2025 | 2,619 | 1,207 | 46.1% | 1,037 | 48.4% | 97.4% |
| Nov 2025 | 5,076 | 2,468 | 48.6% | 2,197 | 43.7% | 96.2% |
| Dec 2025 | 6,359 | 2,780 | 43.7% | 3,330 | 35.7% | 97.2% |
| Jan 2026 | 10,110 | 3,136 | 31.0% | 4,647 | 29.1% | 97.6% |
| Feb 2026 | 9,195 | 2,988 | 32.5% | 4,174 | 26.2% | 98.6% |
| Mar 2026 | 9,848 | 2,768 | 28.1% | 4,559 | 22.6% | 98.6% |
| Apr 2026 | 8,833 | 2,261 | 25.6% | 4,692 | 21.2% | 98.5% |
| May 2026 | 8,577 | 2,305 | 26.9% | 4,938 | 20.3% | 97.9% |
| Jun 2026 | 9,246 | 2,310 | 25.0% | 5,289 | 18.7% | 98.2% |
| Jul 2026 | 8,995 | 2,400 | 26.7% | 4,936 | 19.2% | 98.3% |
| Aug 2026 | 8,891 | 2,143 | 24.1% | 4,126 | 21.1% | 97.0% |
| Sep 2026 | 4,252 | 920 | 21.6% | 2,528 | 16.7% | 96.5% |
Pulled 2026-09-17 11:06 UTC · 12,463 of 12,463 conversations (100.0% coverage) · 69 hit the 100-message API cap and are counted only to their most recent 100.
| Term | Exactly what it counts |
|---|---|
| A miss | One Square card charge with status FAILED. Attempt-level, not account-level — but retries are stable at ~1.1 failed attempts per failing account per month across the whole window, so the count is not being inflated by a change in retry cadence. |
| Miss rate | Failed ÷ (failed + completed) charge attempts in the month. This is the scale-adjusted measure: it does not move just because the book grew. |
| Recovered | That same customer's next completed charge lands within the window. Next, not any — a customer who later pays an unrelated instalment on time has not cured the miss. |
| Window closed | The miss is old enough for the full window to have elapsed. A month is not shown until its window has closed. This is the single most common way a recovery chart invents a decline in the newest month. |
| $ recovered | Capped at the amount missed — a $600 catch-up against a $200 miss is credited $200, not $600. Without the cap, one large catch-up payment makes a bad month look good. |
| Repeat misser | Had a failed charge in the 90 days before this one. Fixed 90-day window, deliberately not "have they ever missed": an all-history lookback grows month by month and manufactures a trend out of nothing but observation time. |
| Text reply rate | GHL TYPE_SMS only, in US Debt Collection (pcTyOadbzBqZO7v3cbwn). Calls, opportunity rows and appointment activity are excluded — they are not texts. Per-message and per-person are both shown because they answer different questions. |
| Limit | Why |
|---|---|
| No 2025 comparison | The Square pull holds 2025, but that side is under-filled — about 1,100 rows a month against roughly 14,000 in 2026. It is a partial backfill, not a quiet year. Any year-on-year read off it would be fiction, so the page starts at Jan 2026. |
| Square misses and GHL texts are not joined | They are two independent reads of the same problem, not a funnel. The debt GHL is fed by the Debt AI G-Sheet, not by this pipeline, so a contact there cannot be assumed to be the same person as a Square decline without a matching key. Reading the two trends side by side is sound; multiplying them is not. |
| No per-person list on this page | Names and amounts belong in the daily chase list, which is built for working, not for reading a trend. → /chase-daily |