What AI logistics means for a forwarder
A forwarder's time goes into the inbox. A request arrives with a packing list attached, and someone retypes the cargo, looks up the rates and writes the offer. Later the client asks "any update?", and someone searches the thread, the carrier site and accounting before answering.
Language models are good at exactly the first part of that work: reading unstructured text and turning it into structured data. They are not a source of facts about your business. A useful AI agent keeps those two apart.
Where AI helps, and where it should not decide
| Job | What AI can do | What stays with your systems and team |
|---|---|---|
| Quote requests | Read the e-mail and attachments and extract cargo, route and Incoterm | The prices come from your rate engine. Your operator checks and sends |
| Order questions | Gather the facts from each step of the order and draft a reply | The operator picks the order, chooses a draft and sends it |
| Translation | Draft the reply in the client's language | The operator reads it before it goes out |
| Tracking | Rarely needed: checking a carrier on a schedule is ordinary automation | The rule for when a status e-mail goes out |
The facts inside an order, step by step
"Gather the facts" sounds vague, but an order holds different facts at each step of its flow, and each kind of question has its answer in its own part. This is how the Logistics Order agent in Logistics ERP splits them.
- Shipment request
- Offers
- Booking and payment
- Services
| Step | The client asks | The facts the agent reads |
|---|---|---|
| Shipment request | "What exactly did you book?" | Cargo lines, containers, pickup and delivery, options, and what is still missing |
| Offers | "What was the agreed price?" | The agreed sell price per service, as it stands on the order. No new prices and no discounts |
| Booking and payment | "Is it booked? Is it paid? Is the B/L ready?" | Booking number, vessel and voyage, and the cargo, VGM, document and S/I cut-offs. Every invoice with its amount, open amount and due date. Each B/L, HAWB or MAWB with its status |
| Services | "Where is my shipment?" | Each service and its handlings in journey order, from pickup, loading, export entry and storage to delivery, with planned or actual times, and the carrier's Track & Trace for departure and arrival |
The agent also keeps to a few rules when it writes from those facts:
- Planned is not done. A planned time is written as planned or expected, and the carrier's tracking, when present, beats the plan for departure and arrival.
- Draft is not issued. A draft bill of lading is called a draft; only a final one is called issued.
- Overdue is said plainly. Open and overdue invoices are named.
- Missing is said too. What the order does not hold yet is reported as not available, not filled in.
- Your margin stays inside. Buying prices, vendor rates and margins never appear in a reply; prices are the agreed sell side, and only when the question is about price.
The rule that matters: AI reads, your systems price
A language model can write a rate that looks right and does not exist. In a quote, that is a commitment to a client. So the model should only turn the request into shipment data; every amount in the offer should come from your own rates, margins and charges, the same ones your team quotes with by hand.
The same goes for status. A reply written from a long e-mail thread can repeat a date that has since changed. A reply drafted from the live order states what the order says now.
Questions to ask before you trust an AI agent
- Where do the prices come from? They should come from your rate engine, never from the model.
- Can it send on its own? Ask what reaches a client without a person reading it first.
- Can you see what it read? Every extracted field should be visible and correctable before anything is priced.
- What does it read? It should work from the live record, and it should not guess which order an e-mail is about.
- What does it cost, and what limits the cost? Ask how use is measured and whether you can set a monthly limit.
- Where does your data go? Ask which AI provider receives it, and whether it is used to train models.
How Logistics ERP does it
Two AI agents work inside the e-mail window your team already uses. The Quick Quote agent reads a request and its attachments and, with your rate engine, turns it into priced offers in about a minute. The Logistics Order agent drafts three replies from the live order. Neither sends anything: your operator checks, edits if needed and presses Send, from their own mailbox.
AI Credits are an add-on per company, € 50 per month, with a monthly limit your administrator sets. The agents use Claude, a model made by Anthropic; a company without the plan sends it nothing. Security & trust says what is sent.