How Tally codes your bank feed

What Tally researches, codes, posts, and stages on its own each day, and which rows it deliberately leaves for you.

The Order Work Happens In

Most bank-feed rows are coded before you open them. Tally researches the merchants, codes what it can prove, posts what is safe, and leaves the rest with a reason.

A row is handled by the first thing that recognizes it:

A bank rule codes it, and an active rule posts it.
Category research proposes an account for rows no rule matched. These are suggestions only — research never posts.
Exception coding takes what is still left, researches the merchant properly, and codes, posts, stages, or asks.

What survives all three is genuinely uncertain, and that is what reaches you.

Exception coding posts a merchant group when the research is conclusive. Its report says so plainly — for example, "I researched, coded, and posted the $248.00 Verizon charges (3 rows) to Telephone & Internet."

When it posts a group, it also saves a rule so the same merchant codes itself next time.

Everything else is held back deliberately:

Staged for one-click posting — coded and ready, waiting on your click.
Client questions — Tally drafted a question to the company owner.
Special items — fixed asset, debt, tax, owner, or payroll activity. These are never posted automatically; they are staged for your call.
Needs you — Tally could not resolve it and says why.
Locked-period rows — reported, not touched.

Special-screen items are staged, not skipped. Fixed asset, debt, tax, owner, and payroll rows sit waiting for a person indefinitely if nobody looks at them.

Reading The Row Labels

In Bank Feeds and Fast Coding, a coded row tells you where its account came from:

Rule coded — an active bank rule applied it.
Suggested — category research proposed it. Change it freely; a human choice always wins.
Tally — Tally coded it from researched evidence.

A label tells you the mechanism, not that the answer is right. Review the payee, amount, and account before posting anything you did not personally research.

When Research Does Not Run

Category research reports why it produced nothing:

AI did not find any uncoded rows to research — nothing was eligible.
AI did not find a confident suggestion — it declined rather than guess. This is the intended behavior; an uncoded row is better than a wrong category.
AI suggestion budget is used up for this company this month — the company's bank-feed AI budget is exhausted for the month.

You can always code and post those rows by hand. See Code one bank transaction.

Automatic Posting

AI auto-post is a separate firm setting shown in the Bank Feeds header. When it is on, high-confidence research can post through the normal accounting path; when it is off, those rows wait for you. Active bank rules post independently of that switch.

Check The Work

Open Fast Coding > Activity or the account register to see what posted and from which source. See Review what posted and who posted it.

If something posted incorrectly, use Undo a posted bank transaction and then fix the cause — a posting that came from a rule will repeat until the rule changes. See Stop an overbroad bank rule.

If activity appeared that you cannot account for at all, report an unexpected change to the books before correcting anything.