The headline
The median person in this sample pays $69 a month in small recurring charges. Small means each individual charge is under $100 on a roughly monthly cadence, with rent, utilities, loan payments, credit card payments, and transfers between a person's own accounts all stripped out. The mean is $104, because the top of the distribution is ugly.
The median person has four of them.
Three of those four go to a merchant that person had zero other transactions with across the entire five months. No purchase, no refund, nothing. The only relationship left is the charge. We call those autopilot charges, and the median person is running $61 a month through them, which is $728 a year.
The distribution is the actual story
Nobody notices a $16 charge. That is the point.
- 61.3% of people have more than $50 a month in these small charges.
- 38.7% have more than $100 a month.
- 14.5% have more than $200 a month in charges where no single charge is over $100.
One in seven people is spending over $2,400 a year in increments too small to trigger a second look.
For scale, if you put rent, loans, and insurance back in, the median total recurring load is $609 a month and the mean is $1,444.
What we did not find
Duplicate streaming is real, and it is smaller than the internet says. 41.8% of this sample pays for at least one video streaming service. 13.8% pay for two or more. 7.1% pay for three or more. The "everyone has five streaming services" story is not in our data.
The money is in the boring tail instead. The single most common small recurring charge is not Netflix. It is whatever Plaid files under general services: 47.4% of people have one, and the median is $16.
Category table
Prevalence is the share of the 196-person cohort with at least one small monthly charge in the category. Categories are Plaid's, unmodified.
| Category | Prevalence | Median charge |
|---|---|---|
| General services (misc) | 47.4% | $16 |
| TV and movies | 41.8% | $18 |
| Electronics and software | 29.1% | $13 |
| Music and audio | 26.0% | $15 |
| Insurance | 20.4% | $30 |
| Internet and cable | 15.8% | $60 |
| Phone | 14.8% | $57 |
| Video games | 14.3% | $11 |
| Other entertainment | 14.3% | $13 |
| Gyms and fitness | 13.8% | $28 |
| Bank fees | 12.8% | $8 |
| Financial services | 10.7% | $16 |
Four more categories (online marketplaces, donations, home security, medical) cleared our detector but not our n>=20 publishing floor, so they are suppressed.
Methodology, including the parts that hurt
The sample. 196 people, all Cashews users with a live bank connection, 54,826 expense transactions. Selection bias runs one direction and it is worth saying out loud: these are people who chose to install a personal finance app. If anything they are more on top of their money than average, not less.
"Recurring" is not our opinion. It is transactions.recurring, set by the detector Cashews already ships: name similarity at or above 0.5, amount within 5%, at least 28 days apart, at least $4.00. We did not build a classifier for this post. Because the detector needs a repeat to fire, subscriptions that started recently are missing. Every number here is a floor.
"Forgotten" is a proxy, not a fact. We cannot see inside anyone's head. What we can see is that a merchant appears in the feed only ever as an automatic charge, with no other transaction in five months. That is evidence of no active relationship, not evidence of forgetting.
The averaging choice, since somebody will ask. A person's monthly figure is their total across the months their feed actually covered, divided by the number of those months. Collapsing each person by their median month instead reads higher ($81 rather than $69 for the headline) because it rounds a skipped billing date up to a full month. Both are in the published JSON. We led with the lower one.
The window is lumpy. 173 to 186 people have covered data in March, April, and May. Only 68 and 55 do in June and July, after roughly 342 bank connections dropped into a re-authorization state in May 2026. The sample leans spring.
Suppression. Any statistic resting on fewer than 20 people is suppressed rather than published. That rule is unit tested, because it is the one thing that must never silently break.
Every number above is reproducible from index_queries.rb, which is read-only and prints aggregates only. No user-level row leaves production.
Pull quotes
- The median person pays $69 a month in recurring charges that are individually under $100.
- Three of the median person's four small recurring charges go to merchants they never otherwise interact with.
- $728 a year, per person, to companies that leave no other trace in the feed.
- 14.5% of people spend over $200 a month in charges no single one of which is big enough to notice.
- Only 13.8% pay for two or more video streaming services. The duplicate-Netflix story is mostly a myth.
One line about us
Cashews is a $3.99-a-month app that shows you one number a day and no homework. The median person in this index is running about 17 times that through charges they never look at. If you want to see your own version of this table, the Leak Finder is at cashews.finance/leaks.
Data: 196 Cashews accounts, 54,826 expense transactions, March 1 to July 31, 2026. Aggregate only, n>=20 floor on every published cell. Questions: [email protected].