Use case · Transport & logistics
Delivery note OCR: automatic data entry into your TMS
The AI reads supplier delivery notes — even scanned or in another language — and extracts products, batches, weights and quantities. You validate the flagged uncertain cases; line-by-line entry disappears.
In brief
Keying delivery notes line by line — products, batches, weights, quantities — is slow, repetitive and error-prone, especially with scanned or foreign-language documents. The AI reads each note, extracts the structured data and flags the uncertain cases so all you do is validate.
In detail
Every delivery note retyped by hand into the TMS or ERP means lost time, and errors you pay for at invoicing. The AIR.TEK automatic reading engine — in production on 195,000 real delivery notes, 99.3% of documents validated — extracts the data, matches it against the order and the invoice, and only flags what deserves a human eye.
The delivery note, an invisible bottleneck
Delivery notes arrive from everywhere — PDFs by email, scans, a photo taken on the dock — and in every format. Rekeying them into the ERP is repetitive, time-consuming and error-prone: one quantity mistyped, one reference missed, and the discrepancy ends up on the invoice.
At high volume, this work takes hours every week and delays accounting reconciliation. It is exactly the kind of task AI absorbs, leaving your teams to check rather than type.
What the AI extracts from a delivery note
From a PDF, a scan or a photo, OCR and AI extract the key information: supplier, date, delivery note number, related order number, references and descriptions, quantities delivered, units, lot numbers and expiry dates where relevant.
The data is structured and normalized — even when every supplier has its own layout. That is where AI goes beyond generic OCR: it understands what each field is, not just the raw text.
Matching: delivery note ↔ order ↔ invoice
The real value is not the data entry, it is the matching. The AI checks the delivery note against the original order and the invoice: quantities as ordered, references as expected, prices consistent. This “three-way matching” (order, receipt, invoice) is what stops you paying for goods that were never delivered.
Discrepancies — a missing quantity, a reference that was never ordered, a price that does not match — are detected and raised, instead of going unnoticed until the books are closed.
The flow, from dock to ERP
Capture: the delivery note arrives by email, upload or photo. Reading: OCR, then structured field extraction. Matching: comparison with the order and the invoice. Control: compliant cases go through; discrepancies and doubtful cases are flagged for human approval. Integration: validated data feeds your ERP or your accounting system.
The AI does the data entry and the first check; your teams decide on whatever falls outside the rules. It saves time; it is not blind automation.
Discrepancies and edge cases: people stay in control
An unreadable scan, a handwritten note, a quantity mismatch: rather than guess, the AI marks the document “to review” and presents it to an operator with the details. You stay in control of anything that commits the company, while automating the large majority of delivery notes without friction.
The engine: measured on 195,000 real delivery notes
Behind this automation is an extraction engine built for transport: an OCR that copes with scans and photos, followed by a specialized language model, in production at a customer on a flow of 8,000 delivery notes a day — 195,000 real documents processed in six months, 2,365 layouts. It reads a delivery note the way a transport planner would — which issuer, which order, which lines — instead of looking for zones in a template. It also recognizes 36 transport document types, in 14 families, and extracts their type, number, date and issuer; the delivery note is the one on which extraction is complete.
These are production figures: 99.3% of documents validated — 97.9% with no human touch, 1.4% taught to the system by a few corrections — and the rate rises with use. Against human-verified ground truth, on documents validated automatically, the issuer is correct 99.4% of the time, the date 98.9%, the number 94.4%. 100% of documents are handled, up to 10,000 an hour on the base deployment. The measurement method is published, and it is replayed on your own documents during the pilot.
The engine runs on a dedicated server, on your premises or hosted, and is sold under an annual license, with unlimited volume within that server’s capacity: no per-page or per-credit billing. At several thousand documents a day, the cost is known in advance.
Confidentiality & integration
Your delivery notes, your suppliers and your prices train only your model: extraction and matching happen where you decide, even on your SABER server, from which nothing is sent to a third-party service. The AI connects to your ERP, your accounting and your email, where the documents already flow.
Comparison
Delivery note entry: three approaches
| Manual entry | Generic OCR | AIR.TEK | |
|---|---|---|---|
| Field understanding | Depends on the operator | Raw text, little structure | Understands each piece of data and structures it |
| Order/invoice matching | Manual and slow | Rarely included | Automatic, discrepancies flagged |
| Doubtful cases | Errors go unseen | Not handled | Flagged for human approval |
| Confidentiality | In-house but slow | Often a third-party cloud | Hosting of your choice, even on your premises |
How it works
The AI does the work. You approve.
- 1Connect to your tools
The AI plugs into your existing data and documents, without replacing your business software.
- 2The AI does the work
It reads, processes and proposes continuously — no re-keying, 24/7.
- 3Humans validate
Your teams only settle the uncertain cases; the result flows back into your workflow.
Questions
Your questions, answered.
Which delivery note formats are accepted?
PDFs, scans and photos — including layouts that differ from one supplier to the next. The AI structures the data whatever the original format.
Does the AI detect discrepancies with the order or the invoice?
Yes: it matches the delivery note against the order and the invoice and flags quantity, reference or price discrepancies instead of letting them through.
Does the data integrate with my ERP?
Yes. Validated data feeds your ERP or your accounting system via API; the AI plugs into your existing tools.
Do my delivery notes and my prices stay confidential?
Yes. OCR and matching run where you decide; with a SABER server, your documents do not leave your premises.
What happens with an unreadable document?
It is marked “to review” and presented to an operator with the details. A person decides; the large majority of compliant delivery notes go through with no intervention.
What are your figures worth?
They come from production at a customer: 195,000 real delivery notes, 99.3% validated (97.9% with no human touch), issuer correct 99.4% of the time and date 98.9%, against human-verified ground truth on documents validated automatically. The full method is on the benchmark page; on your documents, we measure during the pilot.
How is it billed?
Through an annual license with unlimited volume within your server’s capacity — not per page or per credit. The amount is quoted after a free 30-minute first call.
Is there funding to help pay for the project?
Yes, in France. Our training courses can be funded by your OPCO (the French skills operator). AIR.TEK is also referenced by Bpifrance: if your AI project meets the eligibility criteria, it may qualify for a Bpifrance Diag Innov grant of up to €6,500 excl. VAT. We point you to the right scheme.
How much does AI automation with AIR.TEK cost?
No €12,000 development paid up front. You subscribe to your AI team: one all-inclusive monthly subscription, an operating expense (opex) rather than a heavy investment. The amount depends on your scope and is set after a first free discovery call (30 min, remote).