QuickBooks Online Adds Confidence Badges to AI Match Suggestions
QuickBooks Online now shows confidence badges on AI match suggestions; this report covers how to accept them, override them, and avoid common pitfalls.
QuickBooks Online’s bank transactions page has been rebuilt around AI, and every customer now has access to it. The software proposes a match or a category for each imported transaction, grades each suggested match with a confidence badge, and pre-fills the likeliest category when it is confident. Features can vary by plan. We walked through the behaviour, the badges, and the points where you still need to supervise.
Matching has grown broader than exact amounts
Older versions of QuickBooks Online matched only what was almost a mirror image. Your record and the bank record needed near-identical amounts and close dates. Exact matches still surface first, but the difference now lies in what happens when nothing lines up perfectly.
The AI spots partial matches, where your record covers part of a bank amount, and combined matches, where several of your records together explain one bank line. That helps with card payouts that arrive net of fees, or one transfer covering two invoices. It also creates room for error, because a plausible grouping is not always the right one.
Category suggestions lean on your history
When no match exists, the software tries to categorise instead. It reads the full description your bank sends and weighs it against your own history. If it recognises a customer or supplier name in that description, it proposes the categories you normally use for that name, along with a snapshot of your history together.
If names are missing from your past transactions, it falls back to how you categorised similar items before. Failing that, it recommends categories generally associated with the name it found. An icon marks rows carrying an AI suggestion. Where one category stands clearly above the rest, the table arrives pre-filled and waits for your confirmation.
Reading a confidence badge
Every suggested match now carries a badge showing how much trust it deserves. Treat it as a triage signal rather than a verdict. A high-confidence badge suggests the pairing is safe to accept, even in bulk. A lower badge means open the row and check the details yourself: amount, date, and payee. The badge saves time by telling you where to look, not by removing the looking.
How do you accept or override a suggestion?
Both actions start from the same place. Open the Transactions menu, go to Bank transactions, and choose the account. Rows carrying an AI suggestion are marked with the icon described above.
To accept, select a row to expand it. Confirm the match or category on display, then accept or add the transaction. For speed, tick the checkboxes on several high-confidence rows and use the batch accept action to clear them together.
Overriding starts in the same expanded view. Clear the suggested match or category, choose the category you actually want, set the payee if needed, and add the transaction on your terms. Nothing is locked in, and a correction feeds back into future suggestions.
Mistakes are reversible. Find the item among the account’s reviewed transactions, undo it there, and it returns to the pending list for a rematch.
When should you wait instead of accepting?
Some transactions arrive with a message asking you to wait for a match. This usually means the related record has not settled yet, as with a payment still clearing through a processor. Forcing a category onto it invites a duplicate the moment the real record lands. Leave it pending for a day or two. The suggestion comes back, and the match is usually cleaner.
Where do the suggestions go wrong?
Thin history is the most common cause of weak suggestions. A new file, or a newly connected account, gives the system little to learn from. Build history deliberately: categorise manually for a while and set rules for recurring items, so repeats stop needing your attention.
Combined matches deserve a second look before you accept them. Expand the row and confirm the amounts genuinely sum to the bank figure. Watch for processing fees hiding inside the difference.
A pre-filled category is a guess, not a decision. If you catch a wrong guess before adding, change it in the expanded row. If you catch it after, undo and redo. Accepting suggestions without reading them teaches the system your habits, including the sloppy ones.
The short version
The AI layer saves real time on the review page, and the badges make its certainty legible. Accept in bulk where confidence is high, read the low-confidence rows yourself, and verify combined matches line by line. Let ambiguous transactions sit until the real match arrives. Your corrections are part of the training loop, so spend them wisely.