The Bottleneck / 21 Jun 2026

How to automate a slow quoting process with AI

A customer emails a freight forwarder asking what it costs to move a container from Shenzhen to Sydney. It is a simple question. The honest answer is that the quote will take two days to come back, and by then the customer has asked three other forwarders the same thing and will go with whoever replies first with a sane number. The desk is not slow because anyone is lazy. It is slow because the answer is scattered across a rate sheet, an inbox, a carrier contract and one senior broker's memory, and nobody has ever put those in the same place.

This is the first in a series I want to keep writing. Take one chokepoint from a real industry, look at why it stays stuck, and sketch the system I would build to clear it. Not the industry I work in. The shape repeats everywhere, and the freight quote desk is a clean place to start.

Why a simple question takes two days

A freight quote looks like one number. Underneath it is a small pile of decisions.

• Which lane, which mode, which carrier, which container type.
• The base rate, which moves with the market and the season.
• The surcharges. Fuel, currency, port fees, peak season, handling, customs.
• The commodity. Some goods are restricted, hazardous or priced differently.
• The terms. Who pays for what, and where the handover happens.

None of that is hard on its own. The problem is where it all lives.

• The base rates are in a spreadsheet, or three spreadsheets, none of them dated.
• The carrier contracts are in an inbox.
• The surcharge schedules are a PDF that changed last month.
• The real pricing logic, the part that wins or loses the deal, is in one senior broker's head.

So the quote cannot be made by most of the desk. It has to wait for the one or two people who can hold all of that at once. They are busy, because everything waits for them. That is the bottleneck, and it is made of people, not software.

Software was supposed to fix this

Freight has plenty of software. Rate management tools, transport management systems, portals. They store the data. They do not reason over the mess.

A tool can hold a thousand rate lines and still leave the actual pricing to a human, because the judgement that turns those lines into a quote was never written down. It lives in the broker's experience of which lane is really cheaper once you count the delays, which customer gets the sharp price, when to route around a congested port.

The quote desk is a legibility problem before it is an AI problem. You cannot automate a decision nobody has ever made explicit. Making the work queryable comes first. The model is the easy part. The hard part is dragging the knowledge into the open and cleaning it so there is something solid to reason over.

The factory I would build

The unit of work here is a quote. So I would not build a chatbot that sits beside the desk. I would build the factory that makes quotes.

It comes in two stages, and the first one is unglamorous on purpose.

Stage one. Make it queryable. Pull the rate sheets, the surcharge schedules, the carrier contracts and the last year of sent quotes into one structured, cleaned base. Sit with the senior broker and get the tacit rules out of their head and onto the page. This is the digital brain for the quote desk. It is most of the work, and it is worth doing even if you stop there.

Stage two. Draft the quote. Now the request comes in and an agent does the grind:

• Reads the request and pulls the relevant lanes and carriers.
• Applies the current surcharges and checks the commodity rules.
• Drafts the quote, with the rate, the terms and the timeline.
• Shows its sources. Which rate sheet, which date, which contract clause.

Then it stops. It does not send. A broker reads the draft, checks the number, adjusts what only a person can judge, and presses send. Read-only by default, because a wrong quote that goes out under your name is a real liability, and the cost of a bad send is far higher than the cost of a human glance.

What changes is the clock. A two-day quote becomes a five-minute one. The junior on the desk can now quote, because the system holds what used to live only in the senior's head. And you start winning the deals that always went to whoever replied first.

What stays human

The system does the eighty percent that is grind. It hands the broker the twenty percent that is judgement.

• The relationship price. The decision to go sharp to win a key account.
• The strange commodity, the unusual route, the thing the rules do not cover.
• The call to walk away from a quote that is not worth winning.

A model can draft a hundred routine quotes before lunch. It cannot decide that this customer is worth protecting at a thin margin. That is the broker's job, and it is a better job once the grind is gone.

The rule

The bottleneck is never the quote. It is that the knowledge to make one lives in too many places and too few heads.

Make it legible. Let the machine draft. Keep the human on send. Then go and find the next desk, because every industry has one.