The verdict in three sentences
For a haulier processing 25,000 documents a month, AI extraction of delivery notes and CMRs usually pays back in under a year, provided you keep human validation of uncertain cases. Real accuracy on crumpled, handwritten paper sits between 92 and 97 % per field, which is enough if the system knows when it is unsure. The real gain is not only the data entry avoided: it is invoicing customers 5 days earlier, which means cash.
What an extraction project costs in 2026
Around Amsterdam, Schiphol and the Port of Rotterdam corridor, hauliers with 50 to 150 trucks juggle paper CMRs signed on delivery, customer delivery notes in every format, weighbridge tickets and international consignment notes. A typical project combines OCR, a language model to structure fields, business rules and a review interface.
| Item | 2026 range (excl. VAT) | Scope |
|---|---|---|
| Scoping and sampling | 3,000 to 6,000 EUR | Analysis of 500 to 1,000 real documents, field list |
| Extraction pipeline | 8,000 to 18,000 EUR | Capture, OCR, structuring, confidence score |
| Review interface | 4,000 to 10,000 EUR | Queue of uncertain cases, one-click correction |
| TMS and invoicing integration | 4,000 to 12,000 EUR | Matching with the transport order, accounting export |
| Acceptance and tuning | 1,000 to 4,000 EUR | Tests on 4 weeks of live flow |
| Total project | 20,000 to 50,000 EUR | Delivered in 8 to 14 weeks |
Processing cost ranges from 0.03 to 0.10 EUR per document depending on page count, scan quality and the model used. For 25,000 documents, the monthly bill runs between 750 and 2,500 EUR, hosting included. The low end covers typed, legible notes; the high end, multipage CMRs with handwritten remarks.
Accuracy by field type: where AI struggles
An overall rate means little. What matters is field-by-field accuracy and the ability to route uncertain items to a human.
| Field | Typical accuracy | Difficulty | Recommended handling |
|---|---|---|---|
| CMR or delivery note number | 97 to 99 % | Low | Automatic, checked against the TMS |
| Delivery date and place | 95 to 98 % | Medium | Automatic if consistent with the order |
| Weight and pallet count | 93 to 97 % | Medium | Review if gap above 5 % |
| Consignee signature and stamp | 92 to 96 % (presence) | Medium | Detection, not authenticity check |
| Handwritten reservations (box 24) | 85 to 92 % | High | Always human review |
| Trailer registration | 94 to 97 % | Medium | Matched against the fleet |
With a well-tuned confidence threshold, 10 to 20 % of documents go to human review, where an operator corrects in 20 to 40 seconds instead of 2 to 3 minutes of full keying. Reservations matter most: a missed reservation is a cargo claim you discover too late.
e-CMR: should you wait for digitalization?
The Netherlands was an early adopter of the e-CMR protocol and electronic consignment notes are growing, driven by the EU eFTI regulation that fully applies from 2027. Yet in 2026 most shippers and consignees still work on paper, especially SMEs and last-mile deliveries. AI extraction therefore stays useful for several years, and the same pipeline can ingest structured e-CMRs without OCR. Advice: require an eFTI-compatible output data format from day one.
Mini case study
Need a professional website?
Kolonell builds websites that attract clients, optimized for the Sénégalese market. Free quote in 2 minutes.
Sarah, finance and admin director of a 90-truck haulier near Schiphol, employs 4 clerks who key 25,000 documents a month, about 1,000 hours monthly. Invoices go out on average 9 days after delivery. Chosen project: 36,000 EUR, processing at 0.06 EUR per document, or 1,500 EUR a month. After rollout, 15 % of documents go to review, bringing the workload down to about 250 hours a month. Two roles are redeployed to collections and customer service, about 7,000 EUR a month of reallocated payroll. Invoicing moves to 4 days after delivery: on revenue of 1.2 million EUR a month, collecting 5 days earlier frees about 200,000 EUR of working capital. Payback on direct costs: about 7 months.
FAQ
What accuracy should I expect on handwritten CMRs?
Between 92 and 97 % per field for core information, and 85 to 92 % for handwritten reservations. A well-designed system automatically sends the 10 to 20 % of uncertain cases to an operator.
What is the cost per document?
From 0.03 to 0.10 EUR depending on page count and scan quality. For 25,000 documents a month, plan 750 to 2,500 EUR of monthly cost.
Do we need new scanners or a driver app?
Usually not: a 12-megapixel smartphone photo or a 300 dpi desktop scanner is enough. A driver app with guided framing improves accuracy by 2 to 3 points.
Is the data GDPR compliant?
CMRs contain names and signatures, so personal data. Processing must be hosted in the European Union, with a defined retention period, often 7 years for accounting records in the Netherlands.
Does it integrate with our TMS?
Yes, through an API or file export; integration accounts for 4,000 to 12,000 EUR of the budget. Matching against the transport order is what pushes useful accuracy above 98 %.
Let's scope your project. Send us 50 representative documents and your monthly volume: we measure real accuracy and price the project between 20,000 and 50,000 EUR, delivered in 8 to 14 weeks. Detailed quote within 48 h. WhatsApp +221 77 596 93 33.
Mohamed Bah
Fondateur, Kolonell
Passionate about digital and entrepreneurship in Africa, Mohamed has been helping Sénégalese businesses with their digital transformation since 2020. Founder of Kolonell, he believes every SME deserves a professional and accessible online présence.
