How to import any bank statement into a budget app (using the AI you already have)

2026-07-19 · 6 min read

Every budget app faces the same cold-start problem: your financial life already exists, in a bank, as months of transactions — and typing it in by hand is where most people give up. The industry's standard answer is “connect your bank,” which quietly means handing your online-banking credentials to a third-party aggregator and letting your transaction history flow through servers you've never heard of.

There's a better way, and it became possible only recently: the AI assistant you already use is extremely good at reading bank statements.

The idea

Any bank in the world can export a statement — PDF, CSV, sometimes Excel. Those files are messy and every bank formats them differently, which is exactly the kind of unstructured-to-structured problem modern AI models solve in seconds. So instead of building fragile bank integrations, Gilt does something simpler:

• Gilt gives you a prepared prompt that describes precisely the file format it can verify and import — columns, dates, currencies, signs.
• You hand that prompt, together with your statement, to your own AI — Claude, ChatGPT, whichever assistant you already have, in their app, under your account.
• The AI returns a structured file. Gilt validates every row — dates, amounts, duplicates, balances — and shows you the result for review before a single transaction is saved.

Why this is more private than “connecting your bank”

Bank-linking flows route your data through an aggregator's infrastructure and often require your actual banking credentials. With the bring-your-own-AI approach, exactly two parties see your statement: your bank, which already has it, and the AI provider you chose and already trust with your questions. Gilt's own app makes zero network calls to any AI service — it just verifies a file that never leaves your phone.

Why it's more universal

Aggregators cover a few thousand banks, mostly in the US and Western Europe. Statements exist everywhere. A small bank in Armenia, a neobank in Brazil, a credit union with a website from 2009 — if it can produce a PDF, this works. That's not a corner case; for most of the world, it's the only way any budget app will ever import their data.

What review-before-save looks like

AI models are good, but “good” isn't “trusted with your ledger.” That's why the import screen shows every parsed transaction — amount, date, category suggestion — and nothing is written until you confirm. Bad row? Fix or skip it. The AI does the tedious part; you keep the authority.

Six statements, six accounts, one evening

In practice, importing a year of history across several accounts takes an evening of coffee-length sessions: export statement, paste prompt, drop the file back into Gilt, review, done. Compare that to retyping a thousand transactions — or to giving a data broker your bank password.