Learn · Automation

How AI can turn freight booking emails into TMS orders

Q: Can it reply to customers automatically?

For standard questions, yes. Questions about ETAs, tracking status and documentation requirements are answered directly using live data from the TMS, and a booking confirmation follows the shipment being created. Emails that require human attention, such as complaints, non-conforming shipments and exceptional requests, are not answered automatically; they are forwarded to the right employee with a summary. If you want outbound mail reviewed before it leaves during a pilot phase, raise that explicitly with any vendor and ask how the threshold is controlled.

Q: Do we have to change our email setup?

No. The automation connects to what forwarders already use: Gmail, Microsoft Outlook and shared mailboxes (Logentic also supports WhatsApp Business), and one agent can manage multiple mailboxes simultaneously.

Q: Does it handle bookings that arrive as attachments?

Yes. Data is extracted from both the email body and attachments. On the document side, the extraction pipeline handles PDFs, scans, photographs and text embedded in the email body, at a published 99.2% extraction accuracy.

The operational mailbox is the real front door of a forwarding business. Booking requests, rate requests, pre-alerts, status questions and documents all arrive as unstructured email — and at most forwarders, every single one is opened, read, interpreted and re-keyed by an operator. Booking email automation replaces that chain with software that classifies the email, extracts the booking data from body and attachments, validates it, creates the shipment in the TMS, and hands only the exceptions to a human.

Why the mailbox is the bottleneck

Three things concentrate the workload of a forwarding operation in its inbox:

Volume. A typical logistics operator receives between 100 and 300 emails per day. Bookings sit in that stream mixed with pre-alerts, status enquiries, B/L instructions, complaints and internal mail — so before anyone processes a booking, someone has to find it.
Nothing is structured. A booking request might be three lines of text, a filled-in Excel sheet, a customer's own PDF form, or a forwarded thread where the actual request is four replies deep. The data an operator needs — parties, references, container numbers, vessel, ETA, Incoterms — is scattered across body text and attachments in no fixed place.
Every email is touched by a person. Even the routine ones. Reading, deciding what it is, re-keying the fields into the TMS, and typing a confirmation is manual work on every message, which means throughput scales only with headcount. Manual document handling alone averages 30 to 60 minutes per shipment.

The result is familiar: experienced operators spend hours per day on triage and data entry, response times stretch to hours during peaks, and the mailbox — not the trucks, vessels or TMS — sets the pace of the business.

What booking email automation actually means

"Email automation" gets used for anything from an out-of-office rule to a full AI pipeline. A complete booking automation pipeline has five stages, and skipping any of them means the manual work moves rather than disappears:

Classify intent — the system recognizes what each incoming email is: a booking request, a pre-alert, a status question, a B/L instruction, a complaint or internal mail — and routes it accordingly, without a person pre-sorting the mailbox.
Extract the booking data — reference numbers, container numbers, vessel names, ETAs, Incoterms and party details are read from both the email body and its attachments, whatever the format.
Validate — extracted data is checked before anything enters the TMS: are all mandatory fields present, do the parties match your customer records, is the request consistent with the applicable rate agreement? An incomplete booking becomes an exception, not a half-filled shipment file.
Create the shipment in the TMS — validated bookings are written directly into the TMS (CargoWise, Softpak, Descartes, Exact Online or similar) as an order, with an audit trail.
Confirm — and route the exceptions to a human. A confirmation goes back to the customer, and every email the system cannot handle confidently — a complaint, a non-conforming shipment, an unusual request, missing mandatory information — is forwarded to the right employee with a summary and an urgency indicator.

Most "email automation" tools do step 1 at best (rules-based sorting) or step 2 in isolation (extraction with no validation and no TMS entry). Steps 3 to 5 are where the time savings actually live, because unvalidated extraction still requires a person to check every field before the shipment exists.

Your four options compared

Approach	How it works	Strengths	Weaknesses
Manual triage	Operators read, sort and re-key everything	No setup; full human judgment on every mail	Throughput scales only with headcount; slow during peaks; re-keying errors
Shared-mailbox rules & templates	Sender/subject rules sort mail into folders; canned replies for common questions	Cheap; quick to set up	Sorts but doesn't process — data entry stays manual; rules break on forwarded mail and new senders; templates can't use live shipment data
RPA (screen robots)	Scripts replay clicks and keystrokes into the TMS	Automates the entry step for one fixed input format	Doesn't read unstructured email; needs a structured input to start from; brittle when the TMS UI changes
AI agents	Language-model-based system performs all five steps	Reads free text and attachments regardless of layout or language; validates before TMS entry; exception-based workflow	Requires TMS integration; per-email cost higher than mailbox rules

The practical difference shows up on the emails that matter. A mailbox rule can file a message with "booking" in the subject line, but it cannot notice that the container number is in the PDF while the vessel is in the body text, or that the "booking" is actually a cancellation. An AI agent reads the email the way an operator would — and then does the entry work an operator would have done next.

What makes freight email genuinely hard

Booking email is harder to automate than generic office email, and three properties explain most of the failures of simpler tools:

Mixed languages. A Dutch forwarder's mailbox routinely contains Dutch, English and German in the same thread — sometimes in the same email, when a customer forwards a carrier's English pre-alert with a Dutch instruction on top. Keyword rules and templates configured in one language miss the rest.
Forwarded chains. The actual request is often buried in a forwarded thread: the current ask sits on top, but half the data (references, container numbers, earlier agreements) lives further down — alongside outdated details from earlier legs of the conversation that must not be picked up.
Data split across body and attachments. A single booking commonly arrives as a short body text plus a PDF or Excel attachment, with fields divided between the two and sometimes conflicting. A system that reads only the body, or only the attachment, produces confidently incomplete bookings. (Attachment handling is its own discipline — see how B/L and CMR documents are processed and the Bill of Lading guide for what's inside those PDFs.)

Any vendor demo will work on a clean, single-language email with one tidy attachment. Judge the system on a forwarded, bilingual thread with a scanned PDF instead.

What results to expect

Published benchmarks from Logentic's email automation and document processing pages:

Automation rate: 80% of emails fully automated, with human attention only on routed exceptions.
Response time: under 30 seconds average, 24/7 — including evenings and weekends when no operator is watching the mailbox. The fastest path, a clean booking email, goes from mailbox to TMS entry in 8 seconds.
Operator time: around 3 hours per day saved per employee.
Document side: attachment extraction at 99.2% accuracy; shipment handling time from ~45 minutes down to ~2 minutes.

What automation does not remove: judgment on genuine exceptions. A complaint, a request that conflicts with the rate agreement, or a booking with missing mandatory information still lands with a person — the difference is that it arrives pre-summarized and prioritized, and it is the only category of email a person still opens.

Rate requests are the neighbouring workflow in the same mailbox: turning an incoming rate enquiry into a priced quotation follows the same classify–extract–validate pattern. See quote automation.

Six questions to ask any booking automation vendor

Does it read both the email body and the attachments — and reconcile them when they conflict?
How does it handle forwarded chains and long reply threads — does it isolate the current request from the history?
What does it validate before creating the shipment — mandatory fields, customer records, rate agreement — and what happens when validation fails?
Does it create the shipment directly in the TMS with an audit trail, or produce a file someone still has to import?
Can outbound replies start as drafts for human approval, and can the auto-send threshold be tuned per customer or per email type?
How are exceptions surfaced — with a summary and urgency indicator, or just as another unread email?

Frequently asked questions

Can it reply to customers automatically?

For standard questions, yes — questions about ETAs, tracking status and documentation requirements are answered directly using live data from the TMS, and a booking confirmation follows the shipment being created. Emails that require human attention — complaints, non-conforming shipments, exceptional requests — are not answered automatically; they are forwarded to the right employee with a summary. If you want outbound mail reviewed before it leaves during a pilot phase, raise that explicitly with any vendor and ask how the threshold is controlled.

Do we have to change our email setup?

No — the automation connects to what forwarders already use: Gmail, Microsoft Outlook and shared mailboxes (Logentic also supports WhatsApp Business), and one agent can manage multiple mailboxes simultaneously.

What happens when a booking is incomplete?

It becomes an exception, not a half-filled shipment. The system detects that mandatory information is missing and escalates to the right employee rather than creating an incomplete order in the TMS.

Does it handle bookings that arrive as attachments?

Yes — data is extracted from both the email body and attachments. On the document side, the extraction pipeline handles PDFs, scans, photographs and text embedded in the email body, at a published 99.2% extraction accuracy.

Is it safe to let software read our customer email?

Ask any vendor where processing happens and where content is stored. Logentic publishes that all email processing takes place within the EU in line with GDPR, that email content is not stored outside your own environment, and that it is an AEO-certified service provider.