The vendor sends a tax invoice, a delivery challan, an e-way bill, and a test certificate. Eight minutes of stores-team transcription. Six minutes of accounts-team typing. A meeting at month-end to argue over which line mismatched. The AI document engine reads all four documents at once, populates the GRN, populates the bill, runs the three-way match against the PO, and lands the bill in the right inbox — flagged as auto-cleared, or flagged with the exact field that mismatched. Available as a value-added module on top of P2P or Inventory.
Some vendors upload to the portal. Some email PDFs. Some send WhatsApp messages with snapshot attachments. Some hand a sheaf of papers to the truck driver and the stores person photographs them at the gate. The AI accepts all of them — single doc or multi-doc bundle — and threads the entire bundle to one PO. No "please upload as a single PDF" instructions to your suppliers. They send what they have.
Generic OCR reads pixels and gets you 80%. The accountant catches the 20% — and that's the whole problem. Farvision pairs layout-aware OCR with an LLM that knows what fields ought to be in what relationship. The total of line items must equal the subtotal. Subtotal plus tax must equal grand total. Item description on the invoice must match the item on the PO. The two engines cross-validate; low-confidence fields get flagged for review, not silently filled in. Per tenant, accuracy improves over months as the model learns your vendor formats. From day one: 98% field accuracy. By month six: 99.4%.
The stores person at site is the role that benefits most. Before: receive truck, count bags, pull out the paperwork, walk to the office cabin, type item description, type quantity, type batch number, type rate, save, print, file. Eight minutes per delivery, more if the handwriting is bad. Now: scan the bundle on the phone, the GRN form is already populated when they tap "Open GRN", they verify against the physical count, hit Confirm. The hardest part of the job — typing — is gone. The role becomes about verification and judgement, which is what it was supposed to be.
The accounts person at HQ has the complementary half of the same problem. Vendor invoices arrive in a hundred formats — some by email PDF, some scanned-and-emailed, some via portal. The team types every line item, every tax, every discount, every reference number. Then they reconcile against GRN, then against PO. By month-end the AP backlog is two weeks deep. Now: bill auto-populates from the same extracted data the GRN used. Tax breakdown extracted line-by-line. TDS/WHT calculated against the vendor's compliance record. The accountant reviews exceptions, posts the voucher, moves on.
This is the cleanest piece of finance automation in the building. The PO has line items at agreed rates. The GRN has line items at received quantities. The bill has line items at invoiced quantities and rates. Three-way match compares all three line by line: quantity (within tolerance), rate (within negotiated band), tax (correct rate per HSN/VAT category). If everything aligns within tolerance, the bill auto-clears for payment scheduling. If anything's outside tolerance, the bill lands in the discrepancy queue with the exact field flagged. 94% of bills clear automatically. The remaining 6% — where the value of human judgement actually is — get human attention.
The 6% of bills that don't auto-clear are precisely the ones that need attention. The discrepancy queue routes each by mismatch type to the team best placed to resolve it. Quantity-short goes to the stores team for re-count. Rate mismatch goes to procurement for negotiation reference. Tax incorrect goes to accounts. Vendor-compliance issues go to the vendor manager. Each comes pre-loaded with the exact field that mismatched, the evidence, and a suggested action. Resolution stops being archaeology and starts being decision-making.
We were sceptical about the premium fee. Two weeks in, the head of accounts came to me asking if we could roll it out to the other group companies. Half her team's day was gone — back. The audit team called it the cleanest documentation trail they'd ever seen at a developer.— CFO · Top-15 Indian developer
Send us five real vendor invoices — your most common formats. We'll run them through the AI engine on a free trial tenant and walk you through the extraction accuracy, the GRN auto-fill, the bill auto-fill, and the three-way match. 30 minutes. Bring your stores supervisor and your accounts head — they're the ones whose lives change.