dfmchn_

stress test · coffee · subscriptions

Where coffee-subscription churn actually goes: the defection map

We read 8.3 million public posts and comments about coffee, kept the 121,112 that talk about buying and subscribing, and extracted every sentence where someone quits, switches, or threatens to. This is where coffee-subscription customers actually go, and which exits a retention budget can still reach.

2013 to jun 2026 · 8.3M public posts · 1,599 typed exits · no internal data

read this first

  • Public data only. Built from public posts and comments in twelve coffee communities plus global brand sweeps. No brand internal data was used or available.
  • Observed language, not a survey. Every number counts real exit sentences, typed into source → destination → trigger edges.
  • Density, not presence. Brand shares are frequencies within each brand's own corpus, so bigger brands don't look worse by volume.
  • Decay-weighted. Complaints age by trigger type. Price friction fades with the market; mechanics and trust friction doesn't age out.

The headline: 1,599 exits, typed and mapped

dfmchn's defection map, built from 1,599 typed defection edges across 8.3 million public posts, finds 13% of coffee-subscription exits landing where no retention spend can follow. The rest split between recoverable rotations and open doors.

defection edges · typed

1,599

Who left, why, and where they went.

already gone

80%

Of defection posts describe an exit that has already happened.

recoverable cohort

16%

Leave the door open (“might come back if…”).

gone for good

13%

Of exits land somewhere no win-back campaign can follow.

Price complaints fired for eleven months, then cooled

For a decade, price grumbling was a steady background hum: roughly one in ten subscription conversations. When the coffee commodity price broke its all-time record in February 2025 ($4.41/lb after the Brazil drought), the hum became a siren: price-complaint density jumped to 1.7× its 2021 to 2024 average and stayed elevated for about eleven months. Then, as futures eased through late 2025 and 2026, it fell most of the way back. dfmchn measured price-complaint density at 24.6% of all coffee-subscription conversation in Q3 2025, against a 2013 to 2018 baseline near 7%.

That decay is the finding. Price friction is perishable: it spikes with the market and cools with it. The mechanics, trust, and freshness complaints in the sections below don't do this; they accumulate. A brand that panicked about price in 2025 and rebuilt its pricing today is solving last year's complaint; a brand with cancellation friction has the same complaint at every price.

Price-complaint density by quarter, 2013 to mid-2026. Roughly 5 to 10 percent through 2018, stepping up from 2019, spiking above 20 percent after the February 2025 C-price record, then easing back through 2026. 10%20% Feb 2025: C-price record $4.41/lb 2013201520172019202120232025
share of all subscription conversation that is price complaint, by quarter. elevated for ~11 months after the feb 2025 C-price record, then most of the way back.
data table · price-complaint density by quarter
price-complaint density by quarter
quarterprice-complaint density
2013-Q15.9%
2013-Q27.3%
2013-Q34.6%
2013-Q48.1%
2014-Q18.0%
2014-Q29.6%
2014-Q35.1%
2014-Q48.6%
2015-Q112.6%
2015-Q26.8%
2015-Q37.4%
2015-Q46.6%
2016-Q16.2%
2016-Q28.2%
2016-Q36.5%
2016-Q44.4%
2017-Q19.5%
2017-Q28.1%
2017-Q310.1%
2017-Q47.9%
2018-Q16.7%
2018-Q25.6%
2018-Q37.0%
2018-Q46.7%
2019-Q114.3%
2019-Q214.7%
2019-Q312.6%
2019-Q412.4%
2020-Q111.9%
2020-Q212.4%
2020-Q314.6%
2020-Q412.7%
2021-Q110.1%
2021-Q26.6%
2021-Q38.6%
2021-Q414.5%
2022-Q114.2%
2022-Q29.4%
2022-Q314.8%
2022-Q411.1%
2023-Q115.8%
2023-Q212.4%
2023-Q314.3%
2023-Q421.0%
2024-Q120.4%
2024-Q214.6%
2024-Q314.3%
2024-Q413.0%
2025-Q122.3%
2025-Q220.6%
2025-Q324.6%
2025-Q423.3%
2026-Q116.4%
2026-Q214.5%

“$21 per shipment and they sent me a $13 bag of beans. I asked why they did it and they said, "Well some you will lose money on, but some you will gain money on." I found it hard to believe they would make this happen. So I canceled.”

a price / value exit · destination: quit / drink less

The defection map: where they actually go

Every exit sentence names (or implies) a destination. Three destinations dominate: quitting good coffee altogether, rotating to another subscription, and going direct to a roaster's own website, each roughly a quarter of all defection. The map splits sharply by brand:

Defection destinations by brand. Trade Coffee exits concentrate on roasters' own sites. Mistobox, Blue Bottle, and Counter Culture leak to other subscriptions. Unbranded category-level exits mostly quit altogether. Roaster's own siteAnother subscriptionLocal roasterQuit / drink lessBought equipmentHome roastingGrocery beansOtherTrade Coffee3012161715810Unbranded / category-level121064972411Other named subscriptions16186432312Onyx31263355Mistobox1729439316Blue Bottle17321820552Atlas Coffee Club7191142938Counter Culture1734251735Angels' Cup42810367214Bean Box2425152718
cell = percent of that brand's decay-weighted defection edges landing on each destination. rows are brands (defection volume ≥12 weighted edges), columns are destinations.
data table · defection destinations by brand
defection destinations by brand
Roaster's own siteAnother subscriptionLocal roasterQuit / drink lessBought equipmentHome roastingGrocery beansOther
Trade Coffee30%12%16%17%1%5%8%10%
Unbranded / category-level12%10%6%49%7%2%4%11%
Other named subscriptions16%18%6%43%0%2%3%12%
Onyx31%26%3%35%0%0%0%5%
Mistobox17%29%4%39%0%3%1%6%
Blue Bottle17%32%18%20%0%5%5%2%
Atlas Coffee Club7%19%11%42%0%9%3%8%
Counter Culture17%34%25%17%0%0%3%5%
Angels' Cup4%28%10%36%0%7%2%14%
Bean Box24%25%15%27%0%0%1%8%

Trade Coffee churners don't rotate. They cut out the middleman. Only ~13% of Trade exits go to a competitor subscription (half the category norm), while the roaster's-own-site column lights up: the marketplace introduces you to a roaster, then loses you to that roaster. Mistobox, Blue Bottle, and Counter Culture leak sideways instead: a third of their exits rotate to another subscription, which also means they're winnable in the other direction.

Angels' Cup shut down in 2026; its defection profile in the years prior is in the map above.

“Just avoid and find yourself a quality roaster than offers a subscription, they will also probably be cheaper and higher quality than the sludge Trade is pushing.”

a variety exit · destination: roaster's own site

What triggers the exit, and which triggers are still winnable

Variety fatigue and price are the two biggest triggers, but they behave differently in time (see the price section above). The ratio that matters for retention teams is threatening vs already gone: a “still threatening” post is a customer who can still be saved.

Exit triggers ranked by volume: variety, other, price and value, mechanics, trust, freshness. Each bar splits into completed exits and still-threatening posts. Variety426 (103 open)Other410 (71 open)Price / value395 (80 open)Mechanics135 (35 open)Trust125 (19 open)Freshness108 (14 open)already gonestill threatening (winnable)
each trigger's defection posts, split into exits already completed and posts still in the threatening stage.
data table · exit triggers
exit triggers: completed vs still threatening
triggercompletedthreatening
Variety323103
Other33971
Price / value31580
Mechanics10035
Trust10619
Freshness9414

“they renewed my annual subscription a day early without sending a renewal reminder ($500+) and then when I reached out to cancel immediately they said they couldn't because the new subscription had already started.”

a trust exit · destination: quit / drink less

Recoverable vs gone-for-good

Not all churn is equal. An exit to another subscription is a rotation, recoverable with a better offer. An exit to an espresso machine, a home roaster, or the grocery aisle is structural: the budget line item is gone, and no retention spend follows it there. Across the category, 13% of defection edges are this second kind. The unbranded, category-level exits (“done with coffee subscriptions”) are unrecoverable at more than twice the branded rate. When a customer churns from a brand, some other brand usually wins them; when they churn from the category, everyone loses them.

Share of each brand's defection edges that are gone for good versus rotated to another subscription. Unbranded category-level exits are unrecoverable at 24 percent, Trade Coffee at 20 percent, Onyx at 3 percent. Unbranded / category-level24%Trade Coffee20%Counter Culture14%Atlas Coffee Club12%Angels' Cup11%Blue Bottle7%Bean Box5%Mistobox5%Other named subscriptions5%Onyx3%Driftaway0%gone for goodrotated within categoryother exits
share of each brand's defection edges whose destination is unreachable by retention spend (equipment, home roasting, grocery) vs rotation to another subscription. number at right = unrecoverable share.
data table · recoverable vs gone-for-good by brand
recoverable vs gone-for-good by brand
scopeedgesunrecoverablerotation
Unbranded / category-level29124%12%
Trade Coffee40920%9%
Counter Culture6414%33%
Atlas Coffee Club7212%26%
Angels' Cup8311%24%
Blue Bottle1347%36%
Bean Box605%18%
Mistobox1415%30%
Other named subscriptions1735%24%
Onyx1223%27%
Driftaway410%27%

“there is no way to cancel a subscription through the website, you have to email them. I thought that was weird but sent them an email asking to cancel my subscription. No reply.”

a mechanics exit · destination: quit / drink less

The recoverable cohort

16% of all defection language carries an open door: conditional phrasing, “paused for now,” “might come back if.” That's the cohort a win-back program is actually addressing, and it clusters on specific triggers: price exits leave the door open far more often than trust exits, which slam it. dfmchn found 29% of Atlas Coffee Club's defection posts still in the threatening stage, the category's highest, which reads as unhappiness that hasn't converted. The retention window is open right now.

“i recieved a first order from a monthly program... and unsubscribed immediately. despite the fact that in private the roaster insisted i would get recent roasts, all i got wasn't weeks old coffee, no, but coffee quite possibly as old as 3-5 months.”

a freshness exit · destination: quit / drink less

Method note

The analysis used 8.3 million cleaned public posts and comments (full history through June 2026) and no brand internal data.

  • Corpus: 8.3M cleaned public posts and comments from twelve coffee-adjacent online communities via full-archive dumps, plus global full-text brand sweeps beyond those communities. 121,112 documents passed the subscription-relevance gate.
  • Schema: a 95-cluster category schema was induced on the whole corpus (never per brand); 38 clusters survived relevance labeling.
  • Edges: 1,599 defection-bearing documents were parsed into 1,599 typed edges (source → destination → trigger, completed vs threatened, reversible vs final) by a constrained-output LLM pass with a calibration prompt.
  • Density, not presence: brand shares are population frequencies of each brand's own corpus, so bigger brands don't look worse by volume.
  • Decay: complaints decay by trigger type (price: ~10.5-month half-life; variety: ~15; mechanics/trust/freshness: none; structural friction doesn't age out).
  • Regimes: boundaries follow the C-price record of Feb 2025 and subsequent easing, verified against market data.
  • Small brands: under ~150 usable mentions (Yes Plz) are scored against the category schema but carry no brand-level claims.

Related dfmchn field notes: density, not presence · the half-life of a complaint · churn has a grammar.

this is one run on one population. the same pipeline reads any corpus of human text. read the other stress tests and field notes.

Bring us a population