Recurring charges
Subscriptions, bills, and habits — how streams are detected and what the numbers on them mean.
A nine-dollar charge you have been paying since 2021 is nearly invisible in a statement and hard to miss on a timeline. The Recurring screen finds the charges that keep coming back, groups them into streams, and sorts each stream into one of three behaviours.
How streams are found
Transactions are clustered by how similar their descriptions are — nothing else. No date rule, no amount rule. Charges whose descriptions share enough words become one stream, and the most common description in the group becomes its name.
That's deliberately loose. A subscription that renames itself between NETFLIX.COM and NETFLIX 866-579-7172 is one thing, and a stream that quietly split in two would hide exactly the charge you're hunting.
You can tighten or loosen it with the Matching control. Higher means descriptions must be more alike to merge; lower merges more aggressively. Each stream shows its own Match J figure so you can see how tight a particular grouping actually is.
Subscriptions, bills, and habits
Three behaviours wearing similar clothes, told apart by timing and amount stability:
- Subscription — arrives roughly monthly or slower, at a consistent amount. Netflix. The gym. A domain renewal.
- Bill — arrives on the same cadence but the amount moves. The power bill in August is not the power bill in April.
- Habit — everything faster or looser. Coffee two or three times a week, the grocery run, the takeaway. Not a subscription in any useful sense, but a real and often surprising line in your month.
The strip across the top gives you a monthly figure for each of the three, plus a combined total — "this is what recurs, before you have decided anything."
Reading a stream
Each stream carries a few numbers, and they measure different things. Click a stream to see them all — on a narrow window the table keeps only the merchant and its totals, and the rest live in that expanded view:
- Predictability — a 0–100 score, weighted toward timing regularity, with cadence coverage and amount stability behind it. High means you could set a calendar reminder for it.
- On-plan — the share of charges landing within the band of the stream's average. The Band control sets how wide that is; the default is ±12%.
- Spread — how much the amount varies. Near zero is a fixed subscription; a wide spread is a bill or a habit.
- Match J — how alike the descriptions merged into this stream are.
- Next charge — a predicted date, shown only when the cadence is confident enough to be worth predicting: a recognizable period, at least three occurrences, and reasonable regularity. When the timing is too irregular to call, no date is shown rather than a guessed one.
A price increase flag appears when a stream that had been stable stepped up to a new stable amount. That's the subscription that quietly went from $9.99 to $15.99 — the single most common thing people find here.
Controls
- Horizon — how many months back to compute over, up to a chosen end date. A shorter horizon reflects current behaviour; a longer one is more confident about cadence.
- Matching — the description-similarity threshold described above.
- Band — the ± tolerance that defines "on-plan".
- Categories — the same rail as the Transactions screen, sharing your selection with it, so narrowing to Subscriptions in one place narrows it in the other. On a narrower window it sits behind a Categories button beside the page title.
Getting more out of it
Recurring detection reads descriptions, not categories, so it works even on transactions nothing has categorized. But categorizing helps: with the category rail you can ask "what recurs inside Subscriptions" separately from "what recurs inside Groceries", and the two questions have very different answers.
If a stream looks split — the same merchant appearing twice under slightly different names — loosen Matching slightly. If two unrelated merchants have been merged, tighten it.