Solution

Document Processing

Invoices, contracts, IDs, bills of lading, scans — extracted, classified, reconciled against your systems, and filed automatically when they match. Mismatches route to a human with the diff already highlighted.

Available as SaaS or on-premise. Documents stay where regulation requires.

~85%
Straight-through processing
10×
Faster cycle time
99.4%
Field-level accuracy
IDP Pipeline · SISAV

Invoice INV-7812

S
Signal

Email attachment · supplier portal · EDI

I
Interpretation

Classify, OCR, extract fields

S
Solution

3-way match · PO · GR · invoice

A
Action

Post to ERP / DMS or route to reviewer

V
Verification

STP rate · downstream rejections

Not “documents scanned”

“87% straight-through, cycle 6h → 22m”

The document drag

Manual keying and rule-based OCR are slow, brittle, and quietly wrong.

$15

average cost to manually process a single invoice (Aberdeen).

15%

of invoices contain a discrepancy that goes undetected without 3-way match.

5+ days

average AP cycle time at mid-market companies. Discounts get missed.

Templates

break the moment a supplier changes layout. Maintenance never ends.

How it works — SISAV cycle

Same five phases. Worked example: an inbound supplier invoice.

Signal

Inbound from anywhere

Documents arrive in many shapes. The pipeline ingests them all and normalizes to a single document object with original media attached.

  • AP inboxes, shared mailboxes, ticketing attachments
  • Supplier & customer portals, FTP/SFTP drops
  • EDI 810/850/855/856, X12, EDIFACT
  • Mobile uploads (driver photos, expense receipts)
  • Scanners & MFPs, cloud storage (S3, SharePoint, Drive)
📧

Email

🌐

Portal

🖨️

Scanner

📦

EDI

📱

Mobile

☁️

Cloud

All sources → unified document queue

Interpretation

Classify & extract, no templates

Vision-LLM extraction, not brittle templates. New supplier layouts are handled day one. Every field comes with a confidence score and bounding-box citation.

  • Doc type classification (invoice, contract, ID, BoL, receipt…)
  • Header + line-item extraction with confidence scores
  • Tables, signatures, stamps, handwriting
  • Multi-language and mixed-script docs
  • Bounding-box citations for every value

Extracted fields

SupplierAcme Logistics GmbH 99%
Invoice #INV-7812 99%
PO refPO-2026-0331 98%
Total (incl. VAT)€12,480.00 99%
Bank IBANDE89 3704… 82%

IBAN sub-threshold → flagged for review.

Solution

Reconcile against systems of record

Extracted values are matched against the truth source: PO, goods receipt, contract, registry, customer master. Tolerances and policy rules are configurable.

  • 3-way match: PO ↔ goods receipt ↔ invoice
  • Contract ↔ SOW ↔ deliverable cross-checks
  • ID ↔ sanctions/registry verification
  • Tolerance bands & policy-as-code
  • Duplicate detection across the open ledger

3-way match

PO total€12,480.00 ✓
Goods received100% ✓
Invoice total€12,480.00 ✓
Bank IBAN vs supplier mastermismatch ⚠︎

IBAN differs from supplier master → route to AP for verification.

Action

Auto-file or human review with diff

Clean matches post to your ERP/DMS automatically. Anything ambiguous lands in a review queue with the diff highlighted — no scrolling, no re-keying.

  • Post to SAP, Oracle, NetSuite, Microsoft Dynamics, Workday, custom
  • File to SharePoint, Box, S3, M-Files, Documentum
  • Reviewer queue with diff & bounding-box overlay
  • Reviewer corrections feed the model — same supplier won’t fail again

Match clean

Post to ERP

Invoice booked, payment scheduled.

Mismatch

Reviewer queue

Diff & original side-by-side.

Threshold & routing rules per doc type

Verification

STP rate, accuracy, and downstream signal

The cycle closes when the booking sticks. Rejections from the ERP, late corrections, and reviewer overrides feed back as quality signals.

  • Straight-through processing rate per doc type and supplier
  • Field-level accuracy from reviewer corrections
  • Downstream rejection & reopen tracking
  • Audit trail per document — who, what, when, why

Last 30 days · AP

Documents handled42,118
STP rate87.4%
Avg. cycle time22 min
Field accuracy99.4%
Cost / document−78%

Everything IDP needs

In one platform. Connectors to the systems you already run.

Template-free extraction

Vision-LLM models handle new supplier layouts on day one.

Confidence + citations

Every field has a score and a bounding box. Auditors love it.

3-way / 4-way match

PO, GR, invoice (and contract) reconciled with tolerance bands.

Duplicate detection

Across the open ledger — not just exact-match invoice numbers.

Reviewer cockpit

Diff highlights, side-by-side original, one-click approve.

Continuous learning

Reviewer corrections become training signal. Same supplier, fewer flags next time.

Multi-language & script

Latin, Cyrillic, CJK, Arabic, mixed-script docs.

PII masking

Redact before any external LLM call. Policy-as-code.

ERP-native

SAP, Oracle, NetSuite, Microsoft Dynamics, Workday, custom.

Deploy your way

Document Processing runs wherever the documents have to live.

S

SaaS

Hosted, encrypted at rest. Up in days. Region pinning available.

P

On-Premise

For regulated industries. Documents and embeddings never leave your perimeter. Air-gap compatible.

H

Hybrid

Sensitive fields redacted on-prem; LLM extraction in cloud. Policy per doc type.

Common doc cycles

Same SISAV pattern, different system of record.

Accounts payable

Invoices ↔ PO ↔ goods receipt. Capture early-pay discounts.

Contract intake

Counterparty, term, value, renewal — extracted, indexed, alerted on key dates.

KYC & onboarding docs

IDs, proofs of address, registry extracts — verified against sanctions and registries.

Logistics paperwork

BoLs, CMRs, customs declarations — reconciled against shipment manifests.

Insurance claims intake

FNOL forms, photos, repair estimates — triaged with confidence per claim.

Healthcare admin

Referrals, prior-auth forms, lab reports — HIPAA-ready on-prem deployment.

Stop re-keying documents

Tell us what comes through your inbox — we’ll quote the STP ceiling.

Powered by the Aglium platform:

Built on Agentrika

AI automation platform — describe it, deploy it.

Powered by AI Gateway

Sanctioned MCP gateway — governed AI access to your data.