Catch the Weird Charge: How WIMM's Anomaly Detection Works

You know the feeling. You glance at your account, something looks off, but you cannot quite say what. A charge a little bigger than usual, a merchant you half-recognize, a category that crept up this month. Most of the time it is nothing. Occasionally it is the thing you wish you had caught a week ago. WIMM's anomaly detection is built for exactly that gap: it watches the patterns in your own spending and surfaces the charges that do not fit, so the weird one does not slip past you.

Photo by Markus Winkler on Unsplash

Stage one: the math does the first pass

The first stage is pure statistics, with no AI cost at all. WIMM compares each new charge against the history it already has for you, and it looks for four specific kinds of strange.

Photo by Artem Podrez on Pexels

An amount spike is a charge more than two standard deviations above that merchant's own history. If your usual grocery run is $90 and one suddenly clears at $400, that stands out against your pattern, not some generic threshold.

A likely duplicate is the same merchant and the same amount within about 48 hours of each other, the classic accidental double charge.

A new large merchant is a first-time merchant whose charge is well above your normal spending. A brand new payee for a small amount is unremarkable. A brand new payee for a big amount is worth a look.

A category spike is a category whose monthly spending is running about double its historical average. It catches the slow creep that no single transaction would reveal on its own.

Stage two: AI sorts surprises from normal life

Statistics are blunt. Plenty of things that look unusual are perfectly expected: an annual insurance premium that only hits once a year, a seasonal heating bill, a holiday spending bump. If WIMM emailed you about every one of those, you would stop reading the emails.

So only the charges flagged in stage one move on to stage two, where WIMM sends them to Claude (the Sonnet model) to judge. The AI reads each flagged charge in the context of your patterns and decides whether it is a genuine surprise or just normal variation, reviewing up to 20 in a batch. Because only the already-flagged transactions reach this stage, the AI is doing focused work on a short list, not scanning your whole history.

The point of stage two is restraint. It is what keeps your annual premium from being treated like a mystery, and it is why the alerts you do get are worth your attention.

What actually reaches your inbox

Only high-severity anomalies become an email. WIMM calls it your Unusual Transactions email, and it is sent after a sync, so it arrives alongside your fresh data rather than at some random hour.

You are in control of whether it sends at all. The notification is a toggle in your preferences, so if you would rather check anomalies in the app instead of by email, you can turn the email off and nothing breaks.

Be clear about what this is

Photo by Annie Spratt on Unsplash

It is worth saying plainly: this is a heads-up about your own spending patterns, not bank fraud detection. WIMM is not claiming a charge is fraudulent. It is telling you a charge does not match how you normally spend, and then leaving the judgment to you.

That framing is the honest one. You still review the flagged charge and decide what it means. Maybe it is a legitimate big purchase you forgot about, maybe it is a subscription that quietly doubled, maybe it is something you need to dispute. WIMM's job is to put it in front of you. Your job is to decide.

How this differs from the recurring-charge trackers

A lot of money apps focus their attention on recurring charges. Rocket Money, for instance, is built largely around finding and cancelling subscriptions, and that is genuinely useful for the bills that repeat every month.

WIMM's anomaly detection is aimed at a different target: the one-off weird charge. Not the subscription you forgot you had, but the single transaction that is bigger, doubled, brand new, or out of pattern, the kind of thing that hides precisely because it only happens once and so never shows up on a list of recurring items. The two approaches cover different blind spots, and the one-off is the one most tools leave uncovered.

The honest tier note, and where to look

Anomaly detection is a Premium feature. Both stages, the statistical pass and the AI judgment, run on it, along with the Unusual Transactions email. That is the plain answer: this is part of what Premium pays for.

If you want to see how flagged transactions surface before deciding, the demo drops you straight into the transactions view with a sample household loaded, so you can explore the spending data the detection reads from: app.wimm.money/demo?mode=transactions.