Most AP automation vendors quote 95–99% accuracy measured on clean, structured PDFs. For Indian mid-market AP, the relevant question is not what accuracy rate a vendor claims but how their system handles GSTR-2B reconciliation, TDS category classification, and IRN validation failures, the failure modes that carry direct compliance and financial liability. Five architecture-level questions, asked during the vendor demo, separate vendors who have solved for Indian compliance from those who have not.
Two vendors are on your shortlist. Both quoted 99% accuracy and showed clean dashboards. Neither demo used an invoice that looked anything like the ones your AP team actually processes. You have no basis to distinguish them on compliance capability, and you are about to decide based on price and interface preference.
Most Indian mid-market AP automation evaluations go wrong here. The accuracy number is a benchmark figure measured on the vendor's test set. It tells you nothing about how the system performs on GSTR-2B mismatches, TDS section boundaries, IRN validation gaps, or multi-GSTIN routing, the failure modes that generate compliance liability in Indian AP.
Why Accuracy Percentages Don't Predict Compliance Performance in Indian AP
Vendor accuracy benchmarks are typically measured at the document level or on header fields such as vendor name, invoice number, invoice date, and total amount. These fields achieve high accuracy across all systems and carry the least compliance consequence.
The compliance-critical fields in Indian AP, including HSN/SAC codes and GST component splits at line-item level, GSTIN routing, IRN verification, and TDS applicability, are where accuracy drops and extraction errors produce compliance events rather than processing delays.
A 1% error rate on header fields is a minor inconvenience. At 5,000 invoices per month, a 1% error rate on GSTIN routing produces 50 invoices posted to the wrong entity, each requiring a correction filing; the same rate on TDS applicability flags produces 50 potential under-deductions per month, where liability under the Income Tax Act as typically applied sits with the payer.
What those fields are and how to test them on a sample invoice set is covered in AI Invoice Processing Accuracy in India: What CFOs Actually Get. This article covers the questions to ask before you run that test.
What Five Questions Should a CFO Ask Any AP Automation Vendor Before Shortlisting?
Ask these five questions in the vendor demo. The answers and deflections tell you more than any accuracy slide.
| Question | Strong answer | Deflection |
|---|---|---|
| How does your system reconcile extracted invoice data against GSTR-2B, and what happens when there is a mismatch? | Describes scheduled GSTN data pull or IRP integration; explains how ITC blocks are flagged and routed for exception handling | "We match against your ERP data" or no mention of GSTR-2B |
| How does your system handle TDS applicability when the vendor invoice does not state the section? | Describes vendor master flagging for TDS-applicable vendors; explains the human confirmation step for section classification (194C, 194J, 194H) | "We extract what's on the invoice": the system only captures what the vendor wrote, with no applicability logic |
| When your system routes an invoice to the wrong GSTIN, who identifies the error and how? | Describes multi-GSTIN routing logic with automated entity mismatch detection before posting | "The approver catches it": manual review is the only check |
| How do you handle IRN validation for e-invoice-applicable vendors — live IRP check or post-extraction match? | Clearly states whether validation is live against IRP or post-extraction match; explains what triggers an exception on IRN failure | "We capture the IRN from the PDF": extraction only, no clarity on whether the IRN is verified |
| If your system misses a TDS deduction or posts to the wrong GSTIN, what is your SLA for correction and where does liability sit? | Names a specific correction SLA; is clear about where liability sits between vendor and customer | Deflects to accuracy statistics or "that's handled in your ERP" |
For context on TDS deduction mechanics in AP automation, including how vendor classification and section mapping work in practice, see TDS in Accounts Payable: Automating Deduction and Compliance in India.
A manufacturing company in the ₹200–300 Cr revenue range identified a GSTIN routing gap and a TDS applicability gap during evaluation using questions 3 and 2 respectively, gaps the vendor demo had not surfaced (per IQInvoice deployment data).
What Should You Do If a Vendor Cannot Answer These Questions?
A vendor who deflects on GSTR-2B reconciliation, TDS classification, or IRN handling is not having a communication problem. They are revealing an architecture gap. The compliance work their system does not handle will land on your team as manual exception volume, and the liability for missed deductions or wrong-entity postings will sit with you, not with them.
The questions above also function as a scoring tool when comparing two shortlisted vendors. A vendor who answers questions 1, 3, and 5 but deflects on 2 and 4 has a specific, nameable gap. Their TDS logic relies on vendor-stated sections, and their IRN handling is extraction-only with no validation step. That is a quantifiable risk for your invoice mix, not a general concern about accuracy.
Before shortlisting, ask each vendor to process three to five invoices from your own AP inbox during the demo. Include one with multiple HSN/SAC codes, one routed to a secondary GSTIN, and one from a TDS-applicable vendor that does not state the section on the invoice. Their handling of those three invoices will answer questions 1 through 4 faster than any benchmark slide.
For the full set of India-specific requirements to include in an AP automation evaluation, see AP Automation Evaluation Checklist for Indian CFOs: What Global Guides Miss.
To see how IQInvoice handles these questions with your own invoice sample, request a demo.
Key observations
- Vendor accuracy benchmarks are measured on header fields and clean PDF sets and do not reflect performance on the compliance-critical fields where Indian AP automation fails.
- Five architecture-level questions, asked in the vendor demo, surface GSTR-2B reconciliation gaps, TDS classification logic, IRN validation approach, GSTIN routing checks, and liability boundaries that accuracy percentages do not reveal.
- A vendor who deflects on any of the five questions is indicating that exception handling for that failure mode will fall to your team.
- The evaluation test that matters is not a benchmark comparison: it is three to five invoices from your own AP inbox, covering multi-HSN, multi-GSTIN, and TDS-applicable formats, processed through the vendor's system in the demo.
- Scoring two shortlisted vendors against these five questions converts a subjective evaluation into a specific, nameable capability comparison.