50 Delivery Notes, One Receiving Log:
How to Skip Manual Data Entry
A French warehouse receiving 50 deliveries a day collects 50 delivery notes (bons de livraison) — one from Geodis on a printed Sage report, one from DSV as a digital platform PDF, one scribbled on carbon paper by a local carrier. Each is a legally required record under Code de commerce Article L123-22, demanding 10-year retention. Yet none share a common format, and no WMS can extract their data by itself. The receiving team spends hours typing line items into Sage X3 Geode or Cegid — and the three-way match stalls before it starts because the one document that proves what was actually delivered has never been structured.
Key Takeaways
- France's 2026 e-invoicing mandate skipped delivery notes entirely — the one document that proves what actually arrived on the truck stays a paper free-for-all while every invoice goes structured.
- Your WMS tracks pallets with RFID and barcodes, but the delivery note data that feeds your ERP's three-way match still enters the system through a keyboard — 50 deliveries and 2.5 hours of typing per day.
- One column definition reads every supplier's delivery note by what fields mean, not where they sit on the page — turning a 2.5-hour typing shift into a five-minute review in ImageToTable.ai.
Why the Receiving Log Is the Bottleneck French WMS Software Can't Solve on Its Own
French warehouse management systems — Sage X3 Warehousing (Geode), Mecalux Easy WMS, Hardis Reflex, Cegid, LogiX — are excellent at what they do. They track inventory locations, orchestrate picking, manage lot and batch traceability. But every one of them shares the same dependency: they need structured data to begin. A WMS can match a received pallet to a purchase order (bon de commande), update stock levels, and flag discrepancies — but only after someone enters the delivery note (bon de livraison) line items into the system.
That entry step is where the bottleneck lives. A receiving clerk at a French logistics hub might handle 30 to 50 deliveries in a single shift. Each carrier — Geodis, DSV, DB Schenker, a local transporteur — hands over a different document. Some are crisp PDFs printed from the supplier's ERP. Some are handwritten on carbon-copy pads in the cab of a truck. Some arrive as JPEGs photographed on a warehouse phone and forwarded through email. The WMS sees none of them until a human types the data in.
French warehouses reporting automated receiving workflows typically deploy barcode scanning and RFID at the pallet level. But the delivery note itself — the document that answers "what was on this truck, from which supplier, in what quantity" — still gets processed through keyboards and spreadsheets. The automation stops at the physical goods. It doesn't touch the paper.
The receiving log (carnet de réception) — the daily record of all goods received — is the downstream artifact of this data entry. It's what the accounting team uses for three-way matching, what the inventory controller checks against stock movements, and what the auditor requests under Article L123-22 of the Code de commerce. When it's maintained manually from dozens of inconsistent source documents, errors compound. One mistyped quantity. One SKU that doesn't match the purchase order. One delivery note that got filed under the wrong date. The three-way match — bon de commande (purchase order) → bon de livraison (delivery note) → facture (invoice) — can't proceed reliably when the middle document's data has never left the paper it arrived on.
Fifty Suppliers, Fifty Formats: The Real Source of French Delivery Note Chaos
French invoices are converging. The 2026 electronic invoicing mandate (Ordonnance 2021-1190) will require every French business to receive invoices through a certified platform, with Factur-X and Chorus Pro driving format standardization across the entire facture ecosystem. Delivery notes (bons de livraison) were left out of the mandate entirely. No standard format, no mandatory fields, no structured data requirement — because under French commercial law, the bon de livraison is not even a legally required document.
The result is complete format anarchy. In a single morning at a French warehouse, the receiving team might encounter:
Geodis or DB Schenker printed PDF. A clean, multi-column layout with shipment reference, supplier name, line items with quantities and units of measure, and a signature field for proof of delivery (bon de réception). Looks structured — but the field positions, column headings, and even the language of labels vary by supplier. One supplier labels the item reference as "Réf. article," another as "Code SKU," another prints only the EAN barcode.
DSV digital platform export. Generated through myDSV or Schenker Connect — a web-based format with tracking codes, pallet counts, and handling unit IDs. Contains the delivery data but organized around logistics metadata (loading meters, temperature zones for STEF cold chain) rather than the purchasing department's line-item table.
Local carrier handwritten bon de livraison. Small transport companies — the ones handling regional deliveries for agricultural suppliers or construction material distributors — still use carbon-copy books. The driver fills in the delivery date, a product description, and a box count by hand. Handwriting quality varies from legible to barely decipherable, and the document may arrive smudged or creased from a day in the truck cab.
A warehouse that sources from 40 suppliers in a given week will see 40 completely different delivery note formats. Every new supplier adds another layout to learn, another set of field positions to memorize, another Excel template to maintain. This is not a problem that goes away with a better WMS. A WMS stores data. It does not extract it from unstructured documents — and unlike invoices, delivery notes have no Factur-X to force them into a common schema.
The irony is that the data fields on delivery notes are relatively consistent across formats — supplier name, delivery date, PO reference, item codes, quantities — they just appear in different places, with different labels, in different languages. The challenge is not that the information is complex. It's that it's scattered. And the traditional answer — open each PDF, copy each value, paste into Excel — scales linearly with volume. At 50 deliveries a day, that's 250 line items to transcribe. At 99% accuracy, that's still 2-3 errors per day, compounded across the month.
How Batch Extraction Turns a Day of Manual Entry Into Five Minutes of Review
Batch processing changes the equation by decoupling extraction from format. Instead of opening each delivery note individually and copying values from wherever they happen to sit on the page, you define the columns you want once — Supplier Name, Delivery Date, PO Reference, Item Code, Quantity Delivered, Receipt Notes — and the extraction engine reads every document in the batch against that column list, locating each value by what it means rather than where it sits on the page.
This approach, called Custom Column Extraction, works differently from template-based OCR. A template tool needs you to draw a bounding box around the "supplier name" field on every layout variation — because it identifies data by position. When you have 40 supplier formats, you need 40 templates. When supplier 41 arrives, you need one more. By contrast, semantic extraction identifies data by context: it knows that "SAS Transports Durand" near the top of the page, associated with fields labeled "Expéditeur" or "Fournisseur" or "Supplier," is the supplier name — regardless of whether it appears in the header, in a sidebar, or inside a table. One column definition works across all formats.
Files are processed securely and not stored.
The workflow for a French warehouse receiving team is straightforward:
The key shift is this: instead of spending 3 minutes per delivery note transcribing data (which, at 50 deliveries a day, consumes over 2.5 hours), the team spends 5 minutes reviewing the extraction output. A single-page delivery note processes in 5–10 seconds — an 18× speed improvement over manual entry. The time saved doesn't just reduce cost. It eliminates the window during which a typing error — a transposed digit in an SKU, a wrong decimal in a quantity — propagates into the WMS and downstream into the three-way match.
At 50 deliveries per day with an average of 5 line items each, manual entry produces roughly 250 data points. Even at a conservative 1% error rate — one miskeyed character per 100 keystrokes — that's 2–3 errors per day and 50–75 per month. Each error triggers an exception in the three-way matching process: a quantity mismatch, a missing PO reference, a supplier name that doesn't link to the ERP vendor master. Batch extraction doesn't eliminate the need for verification — but it shifts the task from transcribing data to confirming data, which is orders of magnitude faster and less error-prone.
From Receiving Log to Three-Way Match: Closing the Loop the French Way
The ultimate destination for delivery note data is not the receiving log itself. It's the three-way verification that authorizes payment: bon de commande → bon de livraison → facture. Under standard French procurement practice, no invoice should be approved for payment until the quantities and references on the delivery note have been confirmed against both the purchase order and the invoice.
But the three-way match can only be automated if all three documents are structured. The purchase order lives in the ERP — structured by definition. The invoice is increasingly structured, pushed toward standardization by Factur-X and Chorus Pro. The delivery note is the missing link. When it remains unstructured — a scanned PDF, a photo, a handwritten slip — the three-way match cannot begin without manual intervention. Accounts payable either chases the warehouse for delivery confirmations, manually types line items from PDFs, or skips the delivery verification entirely, trusting that what was invoiced matches what was ordered. That last option is how French companies end up with an average of 5% double payments or overpayments on supplier invoices.
A structured receiving log — exported from batch extraction into Excel, CSV, or directly imported into the WMS — changes this dynamic. The log becomes the digital record that flows into the three-way matching pipeline:
For French accounting, the downstream benefits compound. The delivery note data feeds directly into the purchase ledger — PCG accounts 607 (achats de marchandises) and 401 (fournisseurs) — with VAT (TVA) reconciliation against compte 44566 handled at the invoice stage. The 10-year document retention requirement becomes a searchable digital archive instead of a storage room full of filing cabinets. And the monthly reconciliation that used to consume days of accounting time becomes a review of exceptions — not a reconstruction of what was received.
Frequently Asked Questions
Do I need to create templates for each supplier's delivery note format?
No. Semantic extraction reads delivery notes by understanding what each field means — not where it sits on the page. A column named "Supplier Name" will locate the supplier regardless of whether it's labeled "Fournisseur," "Expéditeur," or printed in the header without a label. This is the fundamental difference from template-based OCR, which requires a separate template for each layout variation. One column definition covers every supplier format, including new ones you haven't seen before.
Can it read handwritten French delivery notes?
Yes. The vision model processes handwritten text — including cursive French handwriting common on carbon-copy bons de livraison. Accuracy on handwriting is lower than on printed text, so handwritten fields benefit from a quick visual check during the review step. For the best results on handwritten delivery notes, see our guide on handwritten delivery note extraction.
What happens if a delivery note has a different column structure than the one I defined?
Semantic extraction handles missing fields gracefully. If a particular delivery note doesn't contain a field you defined — for example, some supplier formats don't include lot/batch numbers — that column simply remains empty for that row in the output. The extraction doesn't fail or produce garbage data because a field is absent. This is essential for batch scenarios where format variability means not every document has every field.
How does this integrate with our existing WMS (Sage, Cegid, etc.)?
The extraction output — an Excel file or CSV — can be imported directly into your WMS or ERP. Sage X3, Cegid, and most French WMS platforms support CSV or Excel imports for receiving records. If your WMS has an API for receiving transactions, the structured CSV output can be mapped to the API fields. The extraction step is separate from the import step — you control how and when the data enters your system. For more on French ERP integration patterns, see our guide on extracting French delivery note data into Excel.
How long do I need to keep the digital records?
Under Article L123-22 of the Code de commerce, French businesses must retain all commercial documents — including delivery notes (bons de livraison) and receiving records (bons de réception) — for 10 years from the close of the accounting year. Digital copies are legally valid substitutes for paper originals, provided they guarantee the integrity and readability of the document over the retention period.
Can I use this for supplier invoices at the same time?
Delivery notes and invoices serve different purposes and carry different data — delivery notes contain quantities and item references but no pricing; invoices contain pricing and tax breakdowns. They are processed through the same batch extraction workflow but should be handled as separate batches with different column definitions. For batch invoice processing in a French context, see our guide on batch-processing French supplier invoices.
The receiving log is the document that proves your warehouse received what your suppliers claim they delivered — and the document your accounting team needs to approve payment without overpaying. When it's built by hand from 50 different formats, it's a bottleneck. When it's built by extraction from a single column definition, it's a 5-minute review.