How auto-categorization works in Books

Mercury Books categorizes transactions for you as they arrive, so most of your bookkeeping starts from a coded set of transactions rather than a blank column. This article explains what goes into that and how to steer it.

The three layers

Books works through three layers to decide which GL account a transaction belongs to.

1. Your rules. If a transaction matches a rule you've created, that rule decides the category. Rules take precedence over everything else, so they're the way to lock in a decision you don't want revisited. See our Managing categorization rules article.

2. Similarity matching. Books looks at how you've categorized similar transactions before — same merchant, same description pattern, same account — and applies the same treatment.

3. AI. For transactions that don't match a rule or resemble anything you've coded before, Books uses a model that reads the transaction's description and amount alongside your chart of accounts, and proposes the account it fits best.

If none of the three produces a confident answer, the transaction is left uncategorized for you to code.

How your corrections feed back in

Every time you recategorize a transaction in Books, that decision is learned. The next similar transaction is more likely to arrive coded the way you'd code it.

This works in both directions with your banking dashboard. Categorization is shared across Mercury, so the work you do in Books informs how your Mercury transactions are categorized, and vice versa.

Tip: Correcting the first few of a recurring transaction pays off quickly. It's usually faster than waiting to fix them in bulk later.

Reviewing what was categorized for you

Auto-categorization is a starting point, not a substitute for review — particularly for a backlog of historical transactions, where there's less signal to work from.

  1. Go to Transactions
  2. Filter to uncategorized transactions and code them
  3. Then scan the categorized ones, starting with your largest amounts and your recurring transactions
Screenshot 2026-09-15 at 5.41.04 PM.png

Tip: Sort by amount and check the top of the list first. A handful of large transactions in the wrong account will distort your reports more than a long tail of small ones.

Getting better results

  • Create rules for anything predictable — recurring vendors, subscriptions, transfers you make every month
  • Keep your chart of accounts tight. A shorter, clearly named list of accounts is easier to categorize against than a long one with overlapping names
  • Name accounts the way you'd describe them. Descriptive names give the model more to work with than generic ones
  • Correct as you go rather than in one pass at the end of the quarter


How your chart of accounts should be structured is an accounting question rather than a product one — your accountant or bookkeeper is the right person to ask. See our Understanding your chart of accounts article for how to make changes once you've decided.


FAQs

Can I turn auto-categorization off?

It runs as transactions arrive. If you'd rather categorize everything yourself, you can change any category Books assigns, and rules give you a way to control the outcome exactly.

Why was a transaction categorized incorrectly?

Usually because there wasn't much to go on — a vague bank description, an unfamiliar vendor, or an account name that could plausibly apply. Correct it, and create a rule if it's a transaction you'll see again.

Does auto-categorization change a transaction after I've set it?

No. Once you set a category, it stays. The one exception is transfers between your own accounts, which Books relinks when the other side of the transfer arrives.

Does it work on historical transactions?

Yes, though results are stronger for ongoing activity than for a large backfill of older transactions. Review a backfill before relying on the reports.

 

Mercury is a fintech company, not an FDIC-insured bank. Banking services provided through Choice Financial Group and Column N.A., Members FDIC.

AI-generated responses and suggested actions may vary and are not guaranteed. Please review outputs before taking action.

Did you find this article helpful?