Expense categories

Does AI Categorisation Send Your Transaction Data? A UK Walkthrough

Follow one £12.99 subscription through an expense app's categorisation steps to see what an AI model might receive, and which settings keep you in control.

Editorial illustration for Does AI Categorisation Send Your Transaction Data? A UK Walkthrough

Whether the data is sent depends on the app, and the honest way to find out is to follow a single transaction through it. Some apps only ever show an AI model a short merchant name and a price band. Others pass much more. Here is how to trace it, using one made-up line: a £12.99 payment to a streaming service on a shared household account.

Step one: where does the line start?

The raw line on your statement might read something like a card payment, a reference code and a date. Before any category is chosen, a sensible app tidies the description: it strips codes, dates and branch numbers, leaving just a merchant name. Ask whether the app does this on your device or on its servers. Either can be reasonable, but you should be told.

Step two: what happens before AI is involved?

A good order of operations looks like this:

  1. Your own rules: you once said "anything from this merchant is Subscriptions", so it is Subscriptions.
  2. Your history: you have labelled this merchant before, or one very like it.
  3. Only after both fail, and only if AI is switched on: a model is asked for a best guess.

Our £12.99 line is caught at step one or two for most people, so no AI call is needed at all. The more rules and corrections you build up, the less ever leaves the app.

Step three: what does the model see?

This is the part to pin down. For the streaming line, a careful app would send the cleaned merchant name and an amount band, perhaps "£10 to £25". It would not send your name, the sort code and account number of the household account, the balance after the payment, or the other fifty lines of the statement.

SpendixAI, which is built around India and INR rather than the UK, describes its own approach in exactly these terms: normalised merchant name and an amount band, not your name, account numbers, full statement text or balances. Use that as a benchmark for what a clear answer sounds like, not as a suggestion to sign up from here.

What about a household where two people share one account?

Shared accounts raise the stakes slightly, because a partner's spending is also in the file. That is a good reason to prefer an app that sends less, and to check who in the workspace can see what. Ask whether access is role-based and whether the account owner can export or delete records.

QuestionA reassuring answerA reason to pause
What leaves the app?Merchant name plus a price band"The transaction data"
Can AI be off?Yes, entirely, or per featureOnly on a paid plan, or not at all
Unmatched lines?Held for you to reviewSilently guessed
Reading a statement layout?Optional, clearly labelledAlways on, not mentioned
Sign-in?No request for your online-banking loginAsks for banking credentials

Run a small trial. Import a file with a dozen invented lines, switch AI off, and note how many get a category. Switch it on for one feature and compare. If switching AI off leaves unplaced lines waiting for you, the control is real.

Before you upload a real statement

  • Write down what you want from AI: faster labelling, or nothing at all.
  • Remove any month you would rather not share, such as one with a sensitive payment.
  • Check how to delete a transaction and how long a recycle bin keeps it.
  • Never give your online-banking details to any tracker; a statement file is enough.

Data-protection rules in the UK are outside this article's scope. If you need to know your rights, use the official source.

SPENDIXAI JOURNAL

Useful ideas start with a clearer picture of your own spending.

Explore your reports