Freight Quote Automation: What the Agent Does, and Where It Stops

Freight quote automation means a system that takes an inbound shipment request, sends rate requests out to your panel of forwarders, reads the replies, and hands back a structured comparison, with nobody chasing anyone by email. It is not a rate database. The prices still come from your forwarders at request time, because no published tariff survives contact with a real booking. We built one for our own export desk, and the part worth writing down is where it refuses to act.

What the agent actually does

The flow starts one of two ways. Either a request arrives by email carrying a recognised subject line, or another internal system hands the shipment details over directly: destination country and port, mode of transport, container or cargo type, cargo description, weight.

From there the agent fans the request out to a configured panel of forwarders. Every outbound rate request carries a unique reference code, and that small detail is what makes the rest possible. When a forwarder replies three days later on a thread they have renamed twice, the reference is what matches their number back to the right shipment.

Replies get parsed into fields: rate and currency, carrier or vessel, transit time, free time before demurrage starts, delivery order fee, the validity window, and whatever surcharges arrived as prose in the middle of a paragraph.

A request is marked quoted once a minimum number of forwarder replies have landed. That threshold is a setting, not a line of code, because a business changes its mind about how many quotes it wants before deciding.

One design choice is worth copying if you build your own. Not every inbound reply is a quote; plenty are questions. So every message goes through a cheap keyword check first, and only the ambiguous ones get escalated to a language model. Running the expensive step on all traffic gives the same answer for several times the money, which is the arithmetic in AI agent cost per task.

A slow quote is a pricing problem, not a service problem

Drewry’s World Container Index, the benchmark most procurement teams reference, was assessed at $4,473 per 40ft container on 27 August 2026, down 1% on the week before. A 1% week sounds calm. It is not the shape of the risk.

The risk looks like October 2025. Xeneta analyst Emily Stausbøll reported on 29 October 2025 that average spot rates from the Far East to the US West Coast had climbed 38% in a month, to USD 2,138 per FEU, with Far East to North Europe up 18% to USD 1,968. The same analysis put Far East to US West Coast spot rates 60% below where they sat a year earlier.

Hold both facts at once. Rates fell 60% across the year and rose 38% in a single month inside that fall. A quote your desk started assembling on Tuesday and sent on Friday was not slow. It was wrong.

The desk on the other end is manual too

Descartes and SAPIO Research surveyed 616 shippers and logistics service providers across the United States, Canada and Western Europe, publishing on 15 September 2025. Only 17% described themselves as fully automated. Over a third were heavily or mostly reliant on manual processes.

The performance split is the number to keep. Among companies the study classed as industry-leading financial performers, 51% had fully automated processes. Among below-average performers, 5% did.

A second Descartes study, published on 4 November 2025, covered 434 freight forwarders and customs brokers across North America, EMEA, Asia Pacific and Latin America, recruited through seven trade associations. It found 25% naming manual workflows as the top inhibitor to their growth, and a firm-size gap underneath that: 60% of large companies called technology fundamental to growth, against 27% of small ones.

Read the two together and the picture is plain. Your quote is slow because two manual desks are emailing each other.

Step	Manual desk	Agent-run desk
Sending rate requests	One email per forwarder, details retyped per lane	One fan-out, each message tagged with a reference code
Matching replies	A person reads the thread and remembers	The reference code matches the reply to its request
Reading a reply	Fields copied into a spreadsheet by hand	Rate, transit, free time, fees and validity extracted into fields
A forwarder asks a question	Sits in an inbox until someone notices	Routed to operations, the answer relayed back
A request goes quiet	Discovered when the customer chases	A timed check escalates it to a named person
Deciding it is done	When somebody feels there are enough quotes	When the configured minimum number of quotes has arrived

The rule that keeps it honest: it does not answer for you

When a forwarder replies with a question instead of a price, the agent does not answer. The question goes to an internal operations inbox, a person answers it, and the agent relays that answer back to the forwarder under the same reference.

That limit is deliberate. A model guessing at “can you confirm the cargo is stackable” produces a rate quoted against the wrong assumption, and you find out at the port. A delay you can see costs less than a price you cannot honour.

A separate scheduled check watches for requests that have gone quiet too long and escalates them to a named person. Automation that stalls in silence is worse than no automation, because the desk stops watching for the thing it believes is handled.

What makes this harder than the category admits

Freight quoting resists automation for five specific reasons. A vendor who cannot name them has not built one.

Forwarder emails are unstructured and inconsistent. The same forwarder will format two replies differently in the same week.

Telling a quote from a question in free text is a genuine classification problem, not a keyword match.

Dates, durations and amounts all look alike in prose. “14 days free time”, “valid to 14 October” and “USD 14 per container” share a shape, and mistaking a validity date for a free time allowance is the error that costs real money.

Matching replies to requests across an asynchronous, multi-party thread needs the reference code discipline in place from the very first message out. Retrofit it later and you spend a month reconciling.

Knowing when a request has stalled requires a view of what normal looks like on that lane, because forwarders reply on their own schedule and not on yours.

What it costs to install

This is a standard AI Workforce Sprint, from $9,500 fixed, four to six weeks, built on your accounts and handed over with documentation. You own the mailbox, the code, the settings and the forwarder panel at the end of it, because a quoting desk you cannot change yourself is not an asset.

What you get is one tomte, our word for one production agent with one defined job, turning an inbound request into a compared set of forwarder quotes. It runs under the same rules as the rest of our agent fleet: registered before it runs, fenced to what it may touch, logging every action it takes. As of July 2026 that fleet covered 41 registered agents and 5,450 runs with zero failures across the 13 days measured.

Do the cheap test before scoping anything. Take your last twenty quote requests and count the hours between the customer asking and your reply going out. Then count how many of those hours were a person waiting on an email. The second number is the size of the job, and what to automate first covers how to rank it against the rest of your list.

Send us your last five quote requests, your forwarder list, and the format your customers expect a comparison in. You get a written fixed-price scope, a delivery window, and a plain statement of what the agent will and will not decide on its own, inside one business day. No meeting. Start async.