Security

What data is sent to an AI model when an app categorises expenses?

What data an expense app sends to an AI model when it categorises transactions, how a merchant string differs from a whole statement, and what to ask any app.

Editorial illustration for What data is sent to an AI model when an app categorises expenses?

It depends on the app, and the honest answer is to find out before you switch AI on. The safest designs send only a short merchant string and a rough amount band, not your name, account numbers, full statement or balances. A few optional features send the whole uploaded document. Ask any app which of these it does, and whether you can turn each off.

Is sending a merchant name different from sending a statement?

Very. A merchant string such as a shop name and a city says little about you on its own. A whole statement carries your name, address, account number, balances, salary credits and a year of habits. The first is a small question to a model; the second is handing over your file. When an app says it uses AI, look for which of the two it means.

How does SpendixAI handle this?

SpendixAI categorises in stages. First it applies your own rules. Then it uses your history and similar merchants. Only if AI is switched on does an AI model handle what is left, and it is given the normalised merchant name and an amount band, not your name, account numbers, full statement text or balances. You can switch AI off per feature, set a monthly AI budget or turn it off entirely, in which case unmatched items are held for review. Some optional features, such as reading an unfamiliar statement layout, do process the uploaded document, so read what each switch does.

You can also connect Claude or ChatGPT through MCP, where you choose a scope, can start read-only and can disconnect at any time. That is separate from categorisation and is worth treating as its own decision.

What should you ask any expense app?

  1. What exactly is sent to the AI model for a categorisation: a merchant string, a transaction row or the whole file?
  2. Does it include my name, account numbers, balances or notes?
  3. Is the AI step optional, and can I turn it off for each feature?
  4. Which provider receives the data, and in which country is it processed?
  5. How long is it kept by the app and by the provider?
  6. Is my data used to train any model? Is that stated in writing?
  7. Can I limit the cost or the volume of AI calls?

If an app cannot answer these in plain words, treat that as an answer. Be especially careful with questions about retention and training; an app's own description is where you find the commitment, so read it rather than assuming.

How do you read the answer?

What the app saysWhat it suggestsWhat to do
Only normalised merchant name and amount band are sentA narrow, low-exposure designCheck the retention and training statements as well
Transaction rows are sentMore exposure, depending on the columns includedAsk whether names, notes or balances are removed
The whole statement is processedHighest exposureUse only for features you need, and confirm retention
No details givenUnknownDo not enable it until you know
AI can be switched off per featureYou have controlSwitch off whatever you do not need

What can you do yourself to reduce exposure?

  • Keep sensitive details out of notes and merchant aliases.
  • Use rules for your regular merchants so the AI sees fewer items.
  • Review corrections; each one teaches the tool and reduces the need for AI later.
  • Set a monthly AI budget if one is offered, as a ceiling on how much is used.
  • Delete transactions or documents you no longer need, according to the app's retention options.

A closing thought

AI categorisation is a convenience, not a requirement. A tool that works well with your own rules and history and treats AI as an optional last step gives you the benefit without much exposure. Whatever you use, write down what you chose and when, so you can review it after a few months.

SPENDIXAI JOURNAL

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

Explore your reports