Resolving delivery notes in the back office faster

The problem

The company operates more than 1,800 trucks and handles national and international transport. The back office checks around 26,000 delivery notes a month, roughly 1,450 documents per working day. For each delivery note, eight clerks check quantities, damage, additional services and signatures, compare the details with the transport management system and decide on credit notes or additional charges. Each case takes eight minutes on average, tying up around 3,500 hours a month. Delivery notes arrive as scans, PDFs from customer portals or faxes, often with handwritten additions and inconsistent fields.

The delivery notes vary widely in content and format, and many contain handwritten notes on damage, waiting times or extra stops. Clerks have to interpret these notes, compare them with trip data in the TMS and decide whether a credit note, an additional charge or internal clarification is needed. Relevant information is often scattered across the document, for example in the header, in tables or in the margin. Queries to dispatch or customer service are frequent because details are missing or contradictory. As a result, open cases pile up, credit notes are delayed and demurrage charges sometimes only become visible weeks later, making it harder to manage quality and costs.

Why it was left unsolved

Until now, the problem was seen as necessary grunt work in the back office. The processes are fragmented, involve many different document layouts and seem too operational for a dedicated IT project. Classic OCR solutions were considered too rigid because they expect fixed fields and cannot reliably capture handwritten notes or free text. At the same time, the volume is too high to keep ignoring: the eight clerks are regularly at full capacity, and covering for vacation or sickness immediately leads to backlogs.

The solution automatically analyzes incoming delivery notes, extracts the relevant details and provides the back office with structured suggestions. Documents from scans, PDFs or faxes are first standardized; then quantities, damage notes, waiting times and additional services are interpreted in the context of the trip data. For each delivery note, the system creates a case with a suggested category, the affected line items and a recommendation on whether a credit note, an additional charge or internal clarification is needed. Clerks review these suggestions in an interface, make adjustments and approve the decision, which is then processed further in the TMS or ERP.

Where generative AI comes in

Generative AI understands free text and handwritten notes on delivery notes, maps them to trip data and line items and formulates structured decision suggestions. Unlike rule-based recognition, it handles different layouts, languages and phrasings and interprets incomplete information sensibly; the final decision still rests with the back office.