DOCUMENT TYPES

Delivery note extraction

By Sygnet Research. Written by Sygnet, sourced, checked before publication.

A delivery note (also called a delivery slip or packing list) is the document a supplier's warehouse issues when goods leave for a customer, confirming what was shipped, in what quantity, and when. Warehouse staff, logistics coordinators, and receiving clerks on the buyer's side all handle it, usually at the exact moment they're busiest: during unloading. Finance and procurement teams then use it downstream to reconcile against purchase orders and invoices. Automated extraction turns a stack of these into structured records the moment goods arrive, so discrepancies surface before payment rather than during a month-end audit.

Fields to extract

FieldFormat / exampleValidation check
Delivery note numberDN-2024-08813Unique, sequential per supplier
Delivery date14/03/2024Not in the future; matches carrier timestamp
Supplier nameNordholt Freight & Supply Ltd.Matches known supplier registry
Supplier ID / SIREN552 100 554Checksum valid (see SIREN and SIRET)
Buyer / recipient siteWarehouse 4, Rue de la Gare, LilleMatches internal site list
Purchase order referencePO-88213Exists in purchase order system
Line items (SKU, description)SKU-4471, "Steel bracket 40mm"SKU exists in catalog
Quantity ordered vs delivered500 / 480Flag if delivered < ordered
Unit of measurepcs, kg, palletConsistent with catalog unit
Batch / lot numberLOT-22A410Format matches supplier convention
Carrier / transport companyMeridian LogisticsMatches approved carrier list
Signature / stamp presenceYes/NoDetected on page, not just expected
Total packages / pallets12Matches carrier manifest if available
Damage or discrepancy notes"2 units damaged"Free text flagged for review

Why this document is hard to process

Delivery notes are the least standardized document in the supply chain. Every supplier prints its own template, and many still use carbon-copy forms filled out by hand on a loading dock, in poor light, with a pen that's running out of ink. Quantities are sometimes corrected by hand after the fact, with a strike-through and an initial next to it. That's exactly the kind of edit a template-based system misses entirely.

Multi-page notes are common when an order has many line items, and page 2 rarely repeats the header information page 1 had, so a system extracting page by page loses context. Some suppliers issue notes in the local language of their warehouse, not the buyer's language, which becomes a problem the moment a company sources from multiple countries.

There's also a reconciliation dimension that other documents don't have to the same degree: a delivery note only means something when compared against a purchase order and, later, an invoice. Quantity mismatches, partial deliveries, and back-orders are the norm, not the exception, and a system that extracts fields without understanding the three-way match is only doing half the job. Damage claims and discrepancy notes are often scrawled in a margin, easy for OCR to skip. Getting these right requires reading the whole page as a document, not a grid of text.

How Sygnet processes it

Sygnet first classifies the incoming file as a delivery note, distinguishing it from invoices, purchase orders, and receipts that often arrive in the same inbox. It then infers the document's schema rather than relying on a fixed template, which matters given how much delivery note layouts vary supplier to supplier. Each extracted field carries a confidence score, and every value is linked back to its exact location on the page through field-level provenance, so a reviewer can check a quantity against the original scan in one click.

Validation rules run automatically: quantity ordered against quantity delivered, date logic, supplier ID checksums, and cross-checks against the linked purchase order or invoice already in the system. Handwritten corrections and damage notes get flagged rather than silently accepted or dropped. Only fields below a set confidence threshold, or ones that fail a validation rule, go to a human reviewer; everything else moves straight into structured output as JSON, ready for the ERP or WMS. This selective routing is the difference between a system that helps and one that just adds another queue. For a broader comparison of extraction approaches, see OCR vs VLM.

Validation rules worth applying

  • Delivered quantity should never exceed ordered quantity without an explicit backorder or over-delivery flag
  • Delivery date must fall after the purchase order date and before the associated invoice date
  • Supplier ID (SIREN/SIRET) must pass its checksum and match a registered supplier
  • Line item SKUs must exist in the product catalog, not just resemble a valid format
  • Sum of line item quantities should match the stated total packages or pallets, when both are present
  • Signature or stamp presence should be verified visually, not assumed from a template field
  • Batch/lot numbers should follow the supplier's known numbering convention, flagged if anomalous
  • Discrepancy or damage notes should trigger a hold on the related invoice until resolved

FAQ

Can Sygnet match delivery notes to purchase orders automatically?

Yes. Once a delivery note is extracted, Sygnet looks up the referenced purchase order number and compares line items, quantities, and dates. Mismatches (short deliveries, extra items, wrong site) get flagged before the invoice is approved for payment, which closes the loop that a three-way match is supposed to provide.

What happens when a delivery note is handwritten or poorly scanned?

Sygnet uses a multimodal approach rather than relying purely on OCR text extraction, which helps with handwriting, strike-throughs, and low-quality scans. Fields with low confidence are routed to a human reviewer along with the exact location on the page, rather than being guessed at silently. See OCR vs VLM for the underlying reasoning.

Does this replace our WMS or ERP system?

No. Sygnet extracts and validates the data, then delivers it as structured output into whatever system already holds your inventory and financial records. It's a layer that removes manual re-typing and reconciliation work, not a replacement for the systems that run your warehouse or your books.

NEXT STEP

See it on your own documents

One email when we publish something worth your time.