Assisted transaction organization
Automatic expense categorization you can review and correct
Expense Atlas can suggest categories for imported transactions and receipt items using saved user rules, deterministic matching, cached results, merchant context, and AI assistance. Ambiguous or incorrect results remain editable.
How category suggestions move through the pipeline
Different inputs can take different paths. A known user rule can resolve an item directly, while a new or ambiguous merchant may need AI assistance or a Needs Review result.
- 1
Keep the source visible
Activity can come from a receipt, statement, CSV file, manual entry, or eligible Plaid connection. Source context helps users check duplicates and interpretation.
- 2
Normalize merchant text
Merchant cleanup and aliases can group variants such as a processor-prefixed description with a reviewed canonical merchant name.
- 3
Apply known rules first
Active user rules, cached decisions, and built-in patterns can suggest a category without asking a model to infer every transaction.
- 4
Use AI assistance when needed
When deterministic paths do not resolve the item, available merchant, memo, amount, and receipt context can support an AI-assisted category suggestion.
- 5
Review and correct
Low-confidence, invalid, or unresolved outputs can be routed to Needs Review or manual categorization. Users can correct categories and manage merchant aliases.
Synthetic demo data
Resolving an unfamiliar merchant description
A payment processor label can obscure the merchant that actually matters.
Imported description
The source text is preserved for review.
SQ *NORTHSTAR MKT
Normalized merchant
Cleanup removes the processor prefix.
Northstar Market
Suggested category
Based on the available merchant and transaction context.
Groceries
User correction
The user changes the category after checking the purchase.
Household Consumables
The correction improves the reviewed record and can support a user-controlled rule or alias. It does not prove that every future transaction from the merchant belongs in the same category.
CSS illustration with synthetic demo data. This is not a product interface or screenshot.
Categorization is assistance, not automatic perfection
- Merchant names can be cryptic, shared across purchase types, or missing. Amount and memo context can also be incomplete.
- Rules, caches, aliases, provider categories, and AI suggestions can all produce incorrect results.
- One merchant can legitimately span multiple categories, so a past correction should not be treated as universal truth.
- Transfers, refunds, split purchases, income, cash withdrawals, and receipt line items often need more context than one transaction description provides.
- Review uncategorized and surprising results before using category totals in a Personalized Budget or actual comparison.