Walk into the finance department of any mid-sized Indian company between the 1st and the 11th of the month. You will find the same scene: spreadsheets open across three monitors, a pile of supplier invoices, a printout of GSTR-2B, and someone — usually the most experienced accountant in the room — matching line items by hand.
This is the GST reconciliation tax. It is the single most expensive recurring workflow in Indian B2B operations, and almost no one has automated it properly.
Why reconciliation is so painful
On paper the workflow is simple: match what your suppliers reported on the GSTN portal against what you actually purchased. In practice it is a mess of edge cases:
- Suppliers file late — your purchase from March shows up in May's 2B.
- Invoice numbers differ between your books and theirs — extra hyphens, leading zeros, prefix changes.
- Multiple invoices share the same number across financial years.
- Tax amounts are off by paise because of rounding rules that nobody applies consistently.
- Vendors mistype your GSTIN and the invoice never reaches you at all.
A senior accountant can recognise these patterns intuitively. But intuition does not scale. A 50-crore turnover company can easily process 8,000–15,000 purchase invoices a month. That is several days of full-time human work, every single month, just to ensure your input tax credit is intact.
What AI changes
The reconciliation problem has the exact shape that modern AI is good at: fuzzy matching with structured outputs, where most cases are obvious and the long tail benefits from learned heuristics.
Inside Novrex ERP, the reconciliation agent does three things every Indian finance team currently does manually:
- Pulls the latest GSTR-2B and matches each line against your books using vendor GSTIN, invoice value, tax components, and date — with tolerance bands you control.
- For every mismatch, it classifies the cause: late filing, invoice number variant, value mismatch, missing in books, missing on portal. It does not just flag — it diagnoses.
- It drafts the resolution: a follow-up email to the vendor for missing entries, an entry in your books for those that need to be created, a note for the auditor where neither side will move.
Why this is not just "automation"
We have had GST tools for years that could pull 2B and produce a reconciliation report. The reason teams still do it manually is because the report is just a starting point. Someone still has to call the vendor, draft the entry, decide whether to claim or hold the credit. That is the work.
AI is valuable here precisely because it can take the next step. It does not stop at "here are 320 mismatches." It produces "here are 287 you can resolve automatically with the actions I drafted, and 33 that genuinely need a human."
The ITC question
There is a subtler reason this matters. The 2024–2026 GST changes have made input tax credit much more dependent on real-time supplier compliance. If your supplier does not file on time, you cannot claim. Manual reconciliation typically catches these issues weeks after the fact, when it is too late to chase the supplier without paying out of pocket.
An AI agent that runs reconciliation continuously — not once a month — surfaces non-compliant suppliers in days, not weeks. That is direct cash impact. We have customers where this single capability has paid for the platform within a quarter.
What this looks like next
Reconciliation is one workflow. The same pattern — agent observes, classifies, drafts the next action — applies to TDS reconciliation, vendor onboarding, GST notice responses, and credit note matching. The Indian finance function as we know it is going to look very different by 2027.
The teams that move first will spend their CA capacity on advisory work, not data entry. That is the right way to use a CA.