AI Assistant
How does Finance AI categorize transactions?
Learn how the assistant chooses categories, asks clarifying questions, and remembers patterns.
Quick answer
Sav combines Plaid transaction details, your category list, your corrections, saved memory, and repeatable rules to categorize transactions. Sav asks you only when the answer is genuinely ambiguous.
Steps
- 1Connect at least one account so Finance AI can sync transactions.
- 2Choose or customize your categories during onboarding.
- 3Let Sav work through uncategorized transactions.
- 4If a transaction says AI will categorize tonight, open its detail screen and tap Categorize pending transactions now to start immediately.
- 5Answer any clarification cards when the assistant cannot confidently infer the right category.
- 6Correct categories when needed so future transactions improve.
What the assistant looks at
Sav uses the transaction name, merchant, Plaid category, account type and subtype, amount direction, similar transactions, existing rules, saved memory, and anything you told Finance AI during onboarding.
Account type and subtype help the assistant distinguish transfers from spending, income, loan payments, credit card payments, and account movement between checking, savings, brokerage, and credit accounts.
For peer-to-peer apps like Venmo, Zelle, Cash App, and PayPal, the assistant tries to categorize by the memo or counterparty instead of treating every payment as a transfer.
Rules and memory
When a pattern is repeatable, Finance AI can create a categorization rule so future matching transactions are handled automatically. For broader preferences that are not tied to a merchant rule, the assistant can save memory that helps with future answers.
On mobile onboarding, the start-categorizing screen shows an animated preview before AI training begins and notes that the process can take 5-10 minutes, so users should start it when they have enough time to finish.
If you tap Back during the mobile onboarding training chat, Finance AI saves the current transcript before returning to the intro screen so completed categorization work and the visible chat stay aligned.
If mobile onboarding is reopened during AI training, Finance AI asks the server to resume the training agent before showing the saved chat, so interrupted categorization can continue from the existing onboarding transcript.
During background categorization, Finance AI applies saved rules before asking Sav to make remaining category decisions. Sav also uses your existing rules and saved memory as guidance for new transactions. The Finances screen shows a progress notice while Sav is working, and you can dismiss that notice without stopping categorization.
Large background backlogs keep only the current bounded transaction packet in the assistant's working context, so completed packet history is not repeatedly processed. If the nightly work limit is reached, any rows still uncategorized remain queued for a later background pass.
When you do not want to wait for the nightly pass, a pending-category transaction can open a chat that asks the assistant to start categorizing right away.
- Rules are best for merchant names, memos, and amount patterns.
- Memory is best for durable preferences, like how you think about reimbursements or side income.
- You can override the assistant at any time by changing a transaction's category.
Frequently asked questions
How does Finance AI categorize transactions?
Sav combines Plaid transaction details, your category list, your corrections, saved memory, and repeatable rules to categorize transactions. Sav asks you only when the answer is genuinely ambiguous.
What the assistant looks at?
Sav uses the transaction name, merchant, Plaid category, account type and subtype, amount direction, similar transactions, existing rules, saved memory, and anything you told Finance AI during onboarding. Account type and subtype help the assistant distinguish transfers from spending, income, loan payments, credit card payments, and account movement between checking, savings, brokerage, and credit accounts.
What should I know about rules and memory?
When a pattern is repeatable, Finance AI can create a categorization rule so future matching transactions are handled automatically. For broader preferences that are not tied to a merchant rule, the assistant can save memory that helps with future answers. On mobile onboarding, the start-categorizing screen shows an animated preview before AI training begins and notes that the process can take 5-10 minutes, so users should start it when they have enough time to finish. Rules are best for merchant names, memos, and amount patterns. Memory is best for durable preferences, like how you think about reimbursements or side income. You can override the assistant at any time by changing a transaction's category.