Importing Transactions

Bliss supports CSV and XLSX imports with automatic format detection, AI classification, and deduplication.

The Smart Import flow

1. Upload — Drag a CSV/XLSX file into the import page and select a destination account.
2. Processing — The backend detects the file format, deduplicates against existing transactions (date-range-scoped SHA-256 hash), and classifies each row with the 4-tier AI pipeline.
3. Review — Inspect classifications, correct any mistakes, and approve rows.
4. Commit — Confirmed rows become transactions. Corrections feed back into the AI model.

Tip: For your first import, start with a small batch (a single month or a hundred rows) and review the classifications carefully. Each correction trains the AI on your specific merchants and habits, so subsequent larger imports arrive with far higher auto-classification rates.

File format detection

Bliss matches your file’s column headers against known adapter signatures. If a match is found, it auto-maps columns. If not, you create a custom adapter.

Headers must be on the first row. Bliss reads column headers from the first row of the file (or of the sheet, for Excel files). Many bank exports start with a few lines above the headers — account name, statement period, balance. If your file shows as Unknown format even though your bank is supported, open it, delete the lines above the header row, save, and upload it again.

Preconfigured bank adapters

Bliss ships with 30+ preconfigured adapters that automatically recognize CSV exports from major banks worldwide. Just upload your file and Bliss will detect the format:

Two generic fallback adapters (Date/Description/Amount and Date/Description/Debit/Credit) cover banks not listed above, as long as their CSV uses standard column names.

Bank not listed? Create a custom adapter in seconds — see below.

Creating a custom adapter

When Bliss can’t recognise your file’s format, it shows an “Unknown Format” alert along with a preview table of up to 3 rows from your file — so you can see exactly what Bliss is reading before you configure anything.

Click “Create Adapter for this Format” (or Import Adapters → New Adapter) to open the adapter form. Bliss pre-fills the form using the headers it detected, so you mostly just pick values from dropdowns rather than typing column names by hand.

Match Headers

Detected column names appear as removable chips. You can click × on any chip to remove it, or type a new name and press + to add one. These are the headers Bliss will look for in future uploads to identify this file format automatically.

Column mapping

Each field (Date, Description, Amount, etc.) shows a dropdown pre-populated with the headers from your file. Just pick the right column for each field.

Amount strategy

Choose how amounts are encoded in your file:

Date format

Pick from common presets (e.g. DD/MM/YYYY, YYYY-MM-DD, MM/DD/YYYY) or choose Auto-detect to let Bliss figure it out. You can also enter a custom format string.

Row preview

As you adjust settings, the Row Preview card at the bottom of the form updates in real time to show how the first row of your file will be parsed — date, description, amount, and currency. Use this to catch mapping mistakes before saving.

Default currency

Select the currency to apply when your file doesn’t include a currency column. Leave blank to inherit the destination account’s currency.

Adapters are saved per-tenant and reused automatically on future imports.

Bliss Native CSV format

For maximum control, prepare your data in the Bliss Native format. This bypasses adapter detection and supports all fields including investments and tags.

Required columns:

Optional columns:

Example:

transactiondate,description,debit,credit,account,category,currency,ticker,assetquantity,assetprice,tags
2024-06-15,AAPL Purchase,5000,,Schwab,Stocks,USD,AAPL,25,200,
2024-06-15,Salary,,8500,Revolut Daniel,Salary,EUR,,,,"income,june"
2024-06-16,Grocery Store,45.80,,Nubank,Groceries,BRL,,,,

With Native CSV, rows skip AI classification entirely — accounts and categories are resolved by name.

Investment rows

Buying or selling a stock, ETF or crypto asset needs a ticker, an assetquantity and an assetprice. Without them Bliss can’t price the holding or work out cost basis, so the row can’t be approved or saved until they’re filled in. The same rule applies when you add a transaction by hand or approve one from bank sync.

Two exceptions are worth knowing:

• Funds (the built-in Funds category) can omit the ticker, as long as they have a quantity and a price. A ticker-less fund is tracked by its description and isn’t priced from market data, so update its value from Manage Assets.
• ETFs always need a ticker, even though they share the Funds pricing path.

Assets in manually valued categories (real estate, private equity and similar) don’t need any of these.

Review and commit

After processing, review the staged rows. The AI assigns a category and confidence score to each row. You can:

• Accept individual rows or in bulk
• Override the category with a correction (this trains the AI for future imports)
• Skip duplicates or irrelevant rows
• Tag a row directly in the review drawer — add existing tags or create new ones on the fly, no need to wait until after commit

Rows with confidence above your autoPromoteThreshold (default 0.90) are auto-confirmed.

Why some rows weren’t committed

When you commit, Bliss creates a transaction for every confirmed row it can. A row it can’t create is held back: it stays in Transaction Review with a badge that says why, and the commit toast counts them by reason — for example “12 created. 25 held back: 18 missing price, 4 missing account, 3 duplicates.” Use Held back only in Transaction Review to list just those rows.

Commit again once you’ve fixed the rows: rows that commit lose their badge. “Still to review” in the toast counts rows you haven’t confirmed yet.

AI agents see the same information through the MCP server: in list_imports, each held-back row has error: "HELD_BACK:<codes>" and import.lastCommit.heldBack has the counts by reason.

Next steps

• Investment portfolios — how investment transactions are processed after import
• AI classification — understand the 4-tier pipeline