“You spent 12% less this month.” Every budgeting app says a version of this sentence. Your spending already swings by 15–25% from one month to the next, for reasons that have nothing to do with anything you decided. So the sentence is not an achievement. It is noise, presented as one.
Kesse reports a typical period with the band that number actually has, built from two separate sources and kept separate because they behave differently: the uncertainty of the estimate, which shrinks as you record more periods, and how much you genuinely vary from period to period, which does not shrink and should not. That variation is not measurement noise — it is the thing being measured.
The typical value is a median, not an average. One move, one dental bill, one flight, and an average describes a month nobody had.
The recent half of your history is drawn against the earlier half as two ranges on one axis. Where the ranges overlap, the two halves cannot be told apart, and no percentage is written — writing one would sell a certainty nobody measured. When they come apart, and only then, you get a number.
Small samples lie about their own spread. Two halves were generated with the underlying level held exactly constant, six thousand times over, and a naive comparison was asked how often it saw a change:
| Periods each side | Naive comparison | Kesse |
|---|---|---|
| 4 | 15.4% | 4.9% |
| 5 | 12.3% | 4.5% |
| 6 | 8.0% | 5.1% |
| 8 | 4.2% | 4.2% |
| 12 | 1.3% | 1.3% |
Kesse widens its band at small sample sizes until that false-alarm rate stays under one in twenty. The widening is a measured table, not a formula somebody invented, and it is re-measured on every build.
Kesse asks what you spent this month, or this week — optionally split into a few categories you name. It is deliberately not a transaction log. A ledger you stop keeping is worse than none at all, because you forget it is incomplete and read its total anyway. And this measurement does not need line items.
A range, not a figure. A single number would assume every period costs the same, which is exactly what this app measures as false. If the spending band reaches zero, there is no upper end, and none is written.
Each category’s share carries its own band, and the bands are deliberately wide: a category is part of the total, so the two move together, and narrowing would mean trusting a relationship nobody measured. Rent barely moves; transport moves a lot; the bands say so.
One line: the verdict, the typical period with its band, and how long your money lasts. The watch performs no measurement of its own — two devices measuring the same periods could show two different bands.
Every other app in this category connects to your accounts, which makes your transaction history readable by whoever runs the service. Kesse cannot connect and will not. No account, no sign-in, no sync, no server, no permissions, no analytics, no ads. Each of those lines is a build gate that fails the build if it stops being true — see Privacy.
Kesse reports what your own entries show and never recommends an action, a product, a provider or an amount. It is not affiliated with any bank, broker or lender.
Free. No in-app purchases, no subscription. Everything runs on your device, so there is nothing to bill for.
iPhone, iPad and Apple Watch. English, German, Spanish, French, Japanese and Turkish.
A kese was a purse you could feel the weight of. This one tells you how much that weight varies.