"AI" now appears on every freight software homepage. On most of them it means a chat window bolted to a system of record. This guide is about the other kind — software where the AI does the reading, pricing and drafting before an operator touches the job — and how to tell the two apart in a demo.
It covers the four jobs AI does on a forwarding desk, why a freight-native model beats a general one with a prompt, what the AI gives a manager that a report builder cannot, what the first 30 days look like, and how to evaluate the options — including Zavin, which we build.
Key takeaways
- AI freight forwarding software does four jobs: it reads email and attachments, prices from rates on file or partner replies, drafts replies, quotes and documents, and acts on routine requests — with the operator approving anything expensive.
- Freight-native is the test that matters: a 40HC is not a 40GP, an Incoterm changes what a quote covers, and a house bill is not a master. A general model with a prompt gets these wrong; freight-trained software does not.
- The manager's payoff is intelligence that arrives written: scheduled reports with AI analysis, a copilot that answers from every record and can edit it, and partner turnaround data that decides who gets the next tender.
- Expect the first 30 days to run in review-and-send: inbox connected, rates loaded, every draft approved by a human while the system tunes. Go-live is under 14 days; trust is earned after.
- It runs alongside your system of record. Teams doing this have cut RFQ processing time by up to 85% and handle around 50% of routine email end to end.
What "AI" means in freight forwarding software
Four verbs. If a product cannot show you all four on your own emails, it has a chatbot, not AI freight forwarding software.
Read. Emails and their attachments — RFQs, rate sheets, booking confirmations, packing lists, bills of lading — turned into structured records: ports, equipment, commodity, Incoterm, weight, dates, parties.
Price. Rates on file applied through tariff templates and margin rules; where none exist, structured tenders to partners and replies parsed onto one cost sheet.
Draft. Acknowledgements, clarification questions, partner tenders, customer quotes, status replies and freight documents — written from the record, in the company's voice, translated where needed.
Act. Routine inbound work handled end to end, with an approval queue for anything that sends, creates or files.
The four jobs, on a real desk
Read: the RFQ buried in a fourteen-message thread
A customer forwards a thread. Somewhere in it is a packing list PDF, an "actually make it 3×40HC" and a ready date. The AI reads the thread and the attachment — not the filename — extracts the shipment, links it to the existing account, and flags the one fact still missing with a drafted clarification instead of guessing. The same reading turns a carrier's Excel tariff into rate lines in the company repository and a booking confirmation in Chinese into a shipment draft in English.
Price: rates on file, or the tender goes out
If the lane is covered, the quote is priced from the company's own rate repository through the margin rule for that account and branch — ready for approval in under 30 seconds. If it is not, structured rate requests go to the right pricing contact at each partner office, replies land on one cost sheet as they arrive, and the system records how long each partner took. The AI proposes; the pricing rule decides; the operator approves.
Draft: the reply that already knows the file
The acknowledgement goes out in the company's voice. The partner tender carries the exact cargo details. The customer's "can you confirm free time at destination?" is answered from the quote record, not from memory. A reply in Spanish is drafted in Spanish. Every draft is edited or sent by the operator — none of them is written from a blank screen.
Act: the routine work that never reaches the desk
Tracking queries, document change requests, status questions and acknowledgements are handled from the inbox end to end once the system is tuned — around 50% of routine email. Anything expensive — sending a quote, creating a shipment, filing a document — waits in an approval queue that shows what was extracted, what was drafted, and why the system is confident.
Freight-native AI versus a general model with a prompt
This is the section to test in a demo, because it is where generic AI fails quietly.
- Equipment. A 40HC and a 40GP are different boxes with different rates. A general model will happily treat them as the same "40ft container".
- Incoterm scope. FOB and DAP change which charges belong on the quote. Freight-native software applies the right charge scope; a prompt does not know there is one.
- Chargeable weight. Air freight prices on the greater of gross and volumetric weight. Software trained on forwarding computes it; a general model quotes the number it sees.
- House and master. A house bill is not a master bill, and a consolidation has one of the second and many of the first. The shipment model has to know.
- Thread to case. Six people reply to one RFQ. Freight-native software keeps it one case; a mailbox assistant opens six.
- When it is unsure, it asks. The most important behaviour is the one that looks like a limitation: a missing ready date produces a question, not an assumption.
It learns your vocabulary, too
Freight-native AI also learns the company's own vocabulary — charge names, partner offices, house accounts — so the same charge is the same charge from cost sheet to quote to management report.
What the AI gives a manager that a report builder cannot
The freight forwarding software guide covers dashboards. This is about the layer above them: the parts only a language model can do.
The analyst that reads every report
Scheduled reports arrive with the analysis already written — what moved, what stalled, where margin leaked — in the voice the reader needs: executive, sales, pricing, operations or finance. Data-quality warnings travel with the numbers ("these quotes have no owner", "this filter is empty"), so a clean-looking chart does not hide a dirty record.
Ask, then act
The copilot answers from every record and file the user is allowed to see — "which open quotes over $20k are older than a week?", "what did we last quote this account on SHA–LAX?" — with the record attached. Asked to, it makes the change: adjusts the quote, updates the shipment. Access is field-level and branch-aware, so it cannot see what the user cannot.
Who answers fastest
Because every tender is timed, the manager knows which partner offices reply in an hour and which in three days. That becomes tender routing, and it becomes evidence in the next commission conversation.
The first 30 days
Honest expectations, because the AI is only as good as the tuning.
Days 1–3: connect and load
Inbox connected — Gmail, Microsoft 365 or IMAP. Rate sheets and tariffs loaded into the repository; partner directory seeded and then grown automatically from inbound mail. CargoWise connection tested with a preflight check.
Week 1: review everything
Every classification, every draft, every quote passes through the approval queue. The desk sees exactly what the system extracted and rejects what it got wrong. This is not a limitation of the pilot; it is how the system learns the company's lanes, charge names and tone.
Weeks 2–4: tune and expand
Confidence rises on the lanes and message types it has seen. Routine acknowledgements and status replies start going out with lighter review. Quoting on repeat lanes is approve-and-send. Go-live in review-and-send mode is under 14 days; the share of work handled without edits grows from there.
How to evaluate AI freight forwarding software
Five tests, different from the ones for general forwarding software. Bring the ugliest RFQ you received this month.
- Does it read the attachment, or the filename? Send a packing list PDF with a misleading name and check what it extracted.
- Does it price by rule, or by prompt? Ask which margin rule produced the sell rate and whether a floor would have stopped it.
- What does it do when it is unsure? A missing ready date should produce a question. If it produces a quote, it is guessing on your name.
- Who approves what, and is there a trail? Ask to see the queue and the log of who sent the last ten quotes.
- Can it translate, summarise and reply in the thread's language? Test with a genuine partner email, not an English one.
AI-native or legacy-plus-AI
The landscape splits by where the AI sits, not by whether the word appears on the homepage.
AI-native — built around the email. Zavin (inbox, quoting, CRM, rates, shipments, documents and reporting on one data spine, alongside the system of record). Wisor (quoting and rate procurement). Raft (customs, accounts payable and back-office documents). Sedna (email management for freight teams). Freightmate (document and RFQ extraction).
Legacy-plus-AI — built around the record. CargoWise, Magaya, GoFreight and Descartes are systems of record adding AI features to the job file. Strong at storing and complying; the AI arrives after an operator has already read the email.
Neither category is wrong. The question is which end of the job you want automated first.
Where Zavin fits
Disclosure: Zavin is our product; read this as a vendor's own scoping.
Zavin is AI-native freight forwarding software for the commercial desk. It reads every inbound email and attachment, parses tariffs into a company-wide rate repository, prices quotes from tariff templates and partner replies, drafts replies and documents in the company's voice and language, and handles routine requests end to end under approval control. Scheduled reports arrive with AI analysis; a copilot answers from — and can edit — every record, under field-level access by branch and country.
It runs alongside CargoWise as the system of record, goes live in review-and-send in under 14 days, and is priced publicly at $125 per seat per month plus usage-based AI processing, every module included. Pilot teams have seen quote win rates improve by six percentage points.
What it is not: a customs-filing engine, an accounting ledger, or a replacement for the TMS you already run.
For the broader category, see freight forwarding software; for the industry picture, AI for freight forwarders. Or book a walkthrough on your own RFQs.