AI agents · no hype
AI agents for freight forwarders: what they do and what a person decides
· 7 min read · Flettia team
In logistics, 'AI' is used to name three different things: a chatbot that answers questions, a tool that reads documents, and an agent that does work. This article is about the third, and about the part almost nobody explains: what stays in a person's hands when the agent works well.
What an agent is here (and what it isn't)
An agent, at a freight forwarder, is a program that reads what comes in (an email with a request, a supplier reply, a document), decides within fixed rules what to do with it, executes the action (creates the quote, sends the rate request, records the rate that arrived) and writes down what it did, step by step, in the file.
It isn't a chatbot: it doesn't wait for someone to ask. It isn't a document extractor: reading is the first step, not the job. And it isn't fully autonomous: there are decisions it doesn't make, not because it can't, but because it shouldn't.
What information it works with
- The forwarder's email: that's where customer requests and supplier replies arrive. No inbox access, no agent.
- Own tariffs and contracts: the first thing it checks before asking anyone.
- The supplier directory by category and lane: who gets asked for ocean, air or road rates.
- The house catalogue of destination surcharges: they come from there, not from the supplier.
- The house rules: what is assumed when data is missing, what is never assumed, who signs each step. Learned during induction from the real operation, not from a form.
What the agent decides and what a person decides
The agent decides how to classify an email, what data that request's Incoterm requires, what to assume and how to declare it, which suppliers to ask and what to record when they reply. A person decides the margin and final price, approves everything going to the customer, and resolves the exceptions: a DDP delivery (duties must be pre-calculated), loose cargo without dimensions, an ambiguous origin under EXW or FCA.
That boundary is not a technical limitation that will go away with time. How much to charge a customer in a negotiation is a commercial decision; automating it doesn't improve it, it hides it.
The six controls you should demand
- Nothing goes out without approval: everything addressed to a customer stays a draft until someone on your team approves it.
- The customer only sees what they should: never the cost, the profit, the supplier's name or the internal routing.
- No number is made up: prices come from your suppliers and your tariffs and are transcribed; the agent doesn't estimate or interpolate.
- Declared assumptions: if a field the house usually assumes is missing, it assumes it and writes it down where you can see it.
- The margin is yours: setting the final price is a commercial decision and a person makes it.
- Every action is logged: the agent's runs live in the file and nobody edits them; they are the record.
Where it works: in your systems or in ours
The right question isn't 'do I have to change systems?' but 'where does my operation live today?'. If it lives in email and an in-house system, the agent works there: today Vera, Flettia's quoting employee, connects to Gmail, Zoho and the forwarder's system database; other connections are evaluated during induction. If there is no system, or you want one, Flettia's TMS is included when you hire employees.
How to try it without betting the operation
A paid two-month pilot. The first two weeks are induction: the employee learns from the real operation (how you quote, whom you ask for rates, what your house assumes) and needs access to email, tariffs and databases. For the rest of the pilot it works on real requests while a person on your team reviews what it produces, a few minutes a day. After two months you decide; if it didn't earn its place, 100% of what you paid for it is refunded.
What gets measured during the pilot is not a promise: processing time, supplier wait, minutes of human review, corrections, volume handled and total time to send, with the same definition as the manual process and split by mode, direction and business hours.
What to ask any agent vendor
- Where do the prices it enters come from? If the answer includes 'estimates', keep asking.
- What does it do when data is missing? It should separate what it assumes and declares from what blocks and gets asked.
- Who approves what goes to the customer, and where is the log of each action?
- What does it connect to today, by name, and what gets evaluated later? Distrust logo walls.
- How is it charged: per user, per transaction or by volume? And what happens to my data if I leave?
- What does it measure during the pilot, and against which baseline?
Frequently asked questions
Does an AI agent replace my pricing team?
No. It does the repetitive part (read, extract, ask, record, build) and leaves people the part that decides the business: the margin, the approval and the exceptions.
Do I need to change my email or my system?
No. The employee works on the email and systems you already use (today Gmail, Zoho and your system's database). Flettia's TMS is an included option, not a requirement.
What about my rates and my customers?
Each forwarder's data is isolated from every other's, the customer never receives internal information, and every agent action is logged for audit.
Meet Vera with one of your requests
Book 30 minutes and watch the quoting employee work on a real RFQ from your operation.