● Agentic · documents

Documents posted into the ledger.Classified, extracted, and checked before they post.

Document AI reads GST invoices, purchase orders, GRNs, and bank statements. It checks vendor, GSTIN, line items, and tax, then posts into eManage or holds the document when the totals do not match.

eManage Document AI classifies inbound PDFs and photos, selects the right OCR model, cross-checks GSTIN, amounts, and PO lines, then posts GRNs and vouchers or flags anomalies before a clerk touches the packet.

Types

GST invoice

PO and GRN too

Action

Post or hold

Anomaly before voucher

Formats

PDF · image

Phone or scanner

Loop

Closed

ERP master write-back

eManage AI · Document AI

Document AI classifying a GST tax invoice and extracting vendor, GSTIN, lines, and tax into ERP

PDF · PNG · JPG · GSTIN CHECK

The problem

The packet arrives. The clerk retypes it.

Every GRN, invoice, and e-way document is a second entry of a fact that already existed on paper. Errors show up at GST, not at the gate.

Latency

Same-day inward, next-day books

Invoices pile until accounts has a quiet afternoon.

Mismatch

PO lines vs what was billed

Quantity, HSN, and tax splits are checked by eye.

KYC

Onboarding is a folder

PAN and GST certificates still get keyed into party masters.

On the device

Approve the extract

Invoice fields, GSTIN, and PO lines reviewed on a 16 Pro Max — hold or post without a weak crop of the desktop UI.

9:09

Document AI

Northloom Textiles

NLTD
Extract review · 98.6%RK_INV_1842.pdf

Tax invoice

Rivet Knitworks

Bill to Northloom Textiles

21 Sep 2026

Plant 04

₹ 2,00,600

Payable · matches PO + GST split

  • Vendor

    Rivet Knitworks

    ✓
  • GSTIN

    03AABCR0000R1Z8

    ✓
  • Invoice

    RK/2026-27/1842

    ✓
  • PO match

    NL-PO-4418 · 3 / 3 lines

    ✓
  • Taxable

    ₹ 1,70,000

    ✓
  • GST 18%

    ₹ 30,600

    ✓
HoldApprove & post

How it works

01

Classify

The agent decides invoice, PO, bank statement, KYC, or unknown — and refuses to force a wrong template.

02

Extract

It selects the trained model for that class. Fields come out structured, not as a blob of text.

03

Validate

GSTIN checksums, PO balances, and amount totals are cross-checked before anyone files.

04

Post

Clean packets become ERP transactions. Exceptions land in a hold queue with the original image.

Pipeline

Classify → extract → post

  1. 01

    Classify

    Invoice, PO, KYC, bank — or unknown hold.

  2. 02

    Extract

    Model chosen per class, fields structured.

  3. 03

    Validate

    GSTIN, totals, PO balance.

  4. 04

    Post

    Voucher and image — or the hold queue.

Write-back is the product

Extraction that dies in a CSV is a demo. eManage posts to party masters, GRN, purchase voucher, and the document archive attached to the transaction.

  • Hold on GSTIN, HSN, or amount mismatch — never silent overwrite
  • Original image stored against the voucher for audit
  • Works from a phone photo at the gate or a desktop scan in accounts

Questions plants actually ask

Invoices, purchase orders, e-way / transporter docs, bank statements, and KYC packs (PAN, GST certificate). Unknown types go to a review queue instead of a wrong template.

The agent posts nothing. It flags quantity, rate, tax, or HSN deltas and keeps the image on the hold queue for accounts.

Yes. OCR models can sit on plant or HO edge hardware so trade documents do not leave your network.

It removes retyping. Clerks work exceptions and approvals — the same controls they already have in eManage.

Next