GSTR-2B Matching in Seconds: What One CA-Built Excel Tool Claims
A CA's Excel tool says monthly GSTR-2B matching drops from hours to minutes. What the claim covers, what it leaves out, and how to test it on your books.
A case study on the ICAI's AI site says a client with 200 to 300 invoices takes a CA firm two to three hours of matching every month. For a firm with 30 clients, the same page puts it at 60 to 90 hours.
That is a lot of human time spent comparing two lists. And the cost of getting it wrong is not abstract: the page mentions interest under Section 50 and potential penalties when input tax credit (ITC) claims do not line up.
If you run a business, you may never see this work. Your CA or accounts team does it. But you pay for it, in fees or in salaries, and you feel it when a vendor's invoice is missing from the government portal and your credit is at risk.
The problem, in plain words
Every month, someone takes your purchase register (the list of bills you booked) and sets it beside GSTR-2B, the statement downloaded from the GST portal showing what your suppliers have reported. The job is to sort every invoice into one of four piles.
Matched. Matched, but the tax amount differs. In your books but missing from GSTR-2B, which puts your credit at risk. In GSTR-2B but not in your books, which may mean a credit you forgot to claim.
Today this is mostly Excel and patience. The author of the ICAI page says firms with 5 to 50 staff still do it by hand, because the paid tools it names (ClearTax, IRIS GST, Tally Prime integrations) cost Rs. 10,000 to 50,000 a year, need cloud access and some onboarding, and are aimed at larger enterprises.
What the sources show
The ICAI page describes a tool built in Excel VBA, the macro language inside Excel, by a chartered accountant. You load the two files and it sorts invoices into the four piles, colour-coded, with the amount difference shown per tax head (IGST, CGST, SGST, CESS).
A few design choices matter more than the headline. It matches on GSTIN first, and falls back to trade name and invoice number when the GSTIN is missing. It finds column headers by name, so a different file layout does not break it. It checks for duplicates in your own data before it reconciles. And it runs offline, so the data stays on the machine.
The page is candid about where the AI sits. Claude was used as a development partner to write and debug the code. The matching rules, the edge cases and the output design come from the author's GST practice. The reconciliation itself is ordinary, rule-based code, not a model guessing at your invoices.
The author's before-and-after table claims under 2 minutes per client and under 1 hour firm-wide. The page gives no test description, no sample size and no accuracy figure for those numbers. Treat them as a claim to check, not a result.
KPMG's India site tells a related story from the other end of the process. An unnamed enterprise used a GenAI-based invoice reader to extract vendor invoice data, then applied the company's own parameters to decide credit eligibility. KPMG describes the earlier process as manual, slow and dependent on individual judgement. The page gives no numbers and no date, so it shows the direction of travel rather than the size of the gain.
Think of it as a ledger clerk with two columns and a ruler. The tool is a very fast ruler. The clerk is still the one who decides what to do about a line that does not match.
How the workflow would run
- Export the purchase register from your accounting software and download GSTR-2B for the same month.
- Load both into the matching tool, which checks for duplicate rows first.
- The tool sorts every invoice into the four piles and shows tax differences by head.
- A person reviews the two exception piles and the mismatches: a call to a supplier, a correction in the books, a decision on the credit.
- Only after that review does anyone file or claim anything.
Step four is the one the sources say least about. Neither page describes a review or sign-off step. For any version you build or buy, put it in writing.
What to measure before and after
Pick one client, or one month of your own books, and write down four things before you automate. Minutes spent on matching. Number of invoices in each of the four piles. How many mismatches were found only after filing. Rupees of ITC that sat unclaimed or at risk for more than one cycle.
Then run the same month through the tool and compare. If the tool is faster but the piles differ, the piles are the story.
What is not known yet
The ICAI page states no accuracy testing and no validation on real data. It lists no behaviour for a wrong GSTIN or an unsupported file format. Its own roadmap admits two gaps: classifying eligible and ineligible ITC under Section 17(5), and invoices booked in one month that appear in a later month's GSTR-2B. Both are common in real books, and the tool does not handle them yet.
The page also does not say how the tool is sold or supported, and the licence uses machine-level fingerprinting. Ask who fixes it when the GST portal changes its file format.
Local and offline is a real privacy advantage over cloud upload. It is also a reason to ask where the files are kept and who can open them.
What to test first
Take last month's already-filed reconciliation for one client. Re-run it with the tool, or with any matching tool you are considering. Compare the four piles line by line against what your team found by hand.
If every difference can be explained, you have a tool worth a second month. If even one cannot, you have learned something cheaply.
What would your team need to see in that comparison before they trusted it with a live filing?
