Purchase invoice reconciliation against GSTR-2B

This is a guide to the matching work, not to the tax treatment. What you may claim, when, and what to do about a supplier who has not filed are questions for your CA, and the answers change with amendments and circulars. Nothing here should be read as tax advice.

GSTR-2B is the auto-drafted statement of input tax credit generated from what your suppliers have filed. Your purchase register is what your own team recorded. In principle these describe the same transactions. In practice they disagree every month, and the gap has to be explained before a return is filed.

The reconciliation itself is largely mechanical: line up two lists, find what matches, investigate what does not. What makes it expensive is that the two lists are keyed differently, arrive at different times, and are wrong in different ways.

Why the two lists disagree

It is worth separating the causes, because they have completely different remedies and lumping them together is what makes the task feel endless.

The fourth class is the one worth attacking first. It is not a real disagreement at all: both sides have the same invoice and the strings do not match. Purely mechanical, purely repetitive, and frequently the largest single group by count.

It is a matching problem before it is a tax problem

Invoice numbers are the worst possible join key and they are the one everybody uses. A supplier who writes INV/2026-27/0412 on the invoice may file it as 412, or INV-2026-27-412, or with a branch prefix your copy does not carry. Your team typed what was on the paper. Both are faithful records of the same transaction and a string comparison will never join them.

Which means the reconciliation should not lead with invoice number. A more forgiving approach matches on the combination of supplier GSTIN, taxable value and date proximity first, and treats the invoice number as confirming evidence rather than as the key. That single change usually moves a large share of the monthly exception list into the matched pile.

• Normalise before comparing: strip punctuation and leading zeros, fold case, and compare what is left.
• Match on GSTIN plus taxable value plus a date window, rather than on the invoice string.
• Treat a value agreement to the rupee with a near-miss invoice number as a match needing confirmation, not as a mismatch.
• Keep the original strings on both sides. A normalised key is for matching; the record must still show what each party actually wrote.

What remains after that is the part that genuinely needs judgement: suppliers who have not filed, invoices recorded in the wrong period, and real value differences. That is a much shorter list, and it is a list where a person's time is actually worth spending.

The part software cannot fix

Where a supplier has not filed, no amount of automation closes the gap. The credit is not available because the counterparty has not done something, and the only remedies are commercial: chase them, escalate, or make it a condition of payment.

What software does help with is making that chase systematic rather than heroic. A list of non-filing suppliers, ranked by the value at stake, sent on the same day each month, with the specific invoices attached, is a different activity from remembering to follow up. It is also the point where reconciliation stops being a compliance chore and starts being a working capital conversation.

Whether to withhold payment over an unfiled invoice is a commercial and contractual decision with real supplier-relationship consequences, and in some cases a legal one. That belongs with your CA and your commercial judgement, not with a matching rule.

What automation realistically changes

We are not going to put a percentage on this, because we have not measured it in your business and a number invented here would be worth less than the conversation it replaced. What we can describe is the shape of the change.

1. 1The mechanical classes — identifier differences and straightforward value matches — stop consuming attention, because they are resolved before anyone looks.
2. 2The remaining exceptions arrive already sorted by cause, so the work becomes a set of similar decisions rather than a pile of unlike ones.
3. 3The supplier chase becomes a standing monthly output rather than something that happens when there is time.
4. 4The reconciliation becomes repeatable month to month, which matters more than the hours saved, because it stops being dependent on one person's memory of last month's workarounds.

The honest caveat: this works when your purchase register is reasonably disciplined. If invoices are recorded weeks late or against inconsistent supplier records, reconciliation will surface that rather than solve it. That is genuinely useful information, but it is not the outcome people expect on day one.