AI automation · Logistics Logistics AI: automating transport orders, CMR documents and claims, and what stays with dispatchers

Typical workflows in logistics

These are typical processes, not client projects. In each, the AI prepares and matches, and a dispatcher or clerk decides. For the warehouse and shipping in your own company, see AI in logistics and warehouse.

1. Transport orders into the TMS

Orders arrive by email, PDF, customer portal or EDI, each customer with their own layout and abbreviations.

• Input: the order by email, PDF or portal export.
• AI prepares: reads pickup and delivery addresses, time windows, goods, weights, pallet numbers and special requirements, matches the customer and the agreed rate, and flags what is unclear or unusual, such as a dangerous goods indicator. It drafts the order in the transport management system.
• Dispatcher checks: every flagged item, and decides on subcontracting or loading. Dangerous goods are always handled by a person with the required qualification.
• Output: an order draft in the TMS and a confirmation draft for the customer.

2. Consignment notes and proof of delivery

Signed CMR notes, delivery receipts and photos come back from drivers and consignees in all conditions.

• Input: scans and photos of CMR notes, delivery notes and receipts.
• AI prepares: reads the consignment number, date, signatures and remarks, matches each document to its order, and highlights reservations such as damage or shortage noted by the consignee. It marks missing signatures and missing pages.
• Clerk checks: every highlighted reservation and every incomplete document.
• Output: the order marked complete, ready for invoicing, and a short list of documents that need follow-up.

3. Freight invoice checks

Carrier and subcontractor invoices are compared with the agreed rate, the order and the proof of delivery.

• Input: the invoice, the rate agreement, the order and the proof of delivery.
• AI prepares: compares price, route, weight, surcharges and waiting time, and lists each difference with the lines it comes from.
• Clerk checks: each difference and decides whether to pay, query or dispute.
• Output: a short summary per invoice and the approved invoices passed to accounting.

4. Shipment enquiries and damage claims

Customers ask where their goods are. When something goes wrong, the file for the claim has to be assembled quickly.

• Input: the enquiry, the status data in the TMS or tracking system, and for a claim the CMR note, photos, correspondence and delivery records.
• AI prepares: a reply with the current status taken from the system, and for claims a file with all documents sorted by date, the reservations made on delivery and the deadlines that apply.
• Claims handler checks: the reply where something has gone wrong, and decides on liability, settlement and any filing.
• Output: a reply draft or a status answer, and a complete claims file.

What stays with people, and what we do not do

• Dispatching and subcontracting: planning, assigning loads and deciding who carries what stay with dispatchers.
• Customs and dangerous goods: declarations are the declarant’s responsibility, and dangerous goods are handled by people with the right qualification. The AI reads and prepares, it never submits.
• Drivers and staff: we do not build systems that allocate tasks to individuals based on behaviour, monitor them or evaluate their performance. These are high-risk under Annex III of the AI Act, see risk categories.
• Claims decisions: liability, settlement and filing are decisions for the claims handler.
• Customer chat: gives status answers from live data, says clearly that it is an AI and hands over exceptions and claims to a person.

Rules that matter

The table lists the main sources as an orientation. Which rules apply depends on the mode of transport, the route and the contract.

For people: where AI is used to manage workers, Annex III of the AI Act lists task allocation and the monitoring and evaluation of performance as high-risk. In Germany the works council also has a right of co-determination on technical systems that can monitor behaviour or performance (§ 87 BetrVG). On top of this, the GDPR applies to driver and customer data: a data processing agreement with every provider (Art. 28) and processing in the EU or UK. Art. 4 of the AI Act asks for sufficient AI literacy among staff, and Art. 50 requires AI chatbots to identify themselves. This overview is general information, not legal advice. Your legal adviser and your trade association can confirm what applies.

Systems you probably use

Forwarders and hauliers typically work with a transport management system such as SAP TM, CargoWise or a TMS from a specialist vendor, often with a freight exchange, a visibility platform such as project44 or Transporeon and accounting software alongside. We connect through the interfaces these systems offer (API, EDI, export, email or a watched folder) and work alongside them where none exists. The names are for orientation only and do not imply any partnership.

How AI assistants find logistics providers

Shippers ask ChatGPT, Gemini or Google’s AI Overviews for “a haulier for pallet freight from Leeds to Rotterdam” or “a forwarder for temperature-controlled goods”. The answer draws on the company’s own pages, trade association and certification listings, the Google Business Profile and reviews.

• A page per lane, service or goods type: routes, transit times, vehicle types, handling and certifications in plain words.
• Verifiable credentials: licences, memberships such as FIATA, ISO certificates and customs status listed the same way everywhere.
• Consistent company data: name, address, depots and contact details identical on the website, in directories and on the Google Business Profile.
• Structured data: Schema.org Organization with the services and areas served.

Is it worth automating? Eight questions

1. Does someone re-key transport orders from emails and PDFs into the TMS?
2. Do signed CMR notes and receipts wait in a pile before they are matched to orders?
3. Are carrier invoices checked against rates only in part?
4. Do customers call to ask where their goods are?
5. Does assembling a claims file take days of searching in mailboxes?
6. Are deadlines for reservations and claims tracked by memory?
7. Is there a written rule on what may go into AI tools and who checks the result?
8. Can you name one task where a draft that needs checking would already save time?

Three or more yes answers usually point to a workflow worth testing. Question 7 is the prerequisite.