AI freight quoting: why your rate and load history is the core input

What does an AI freight quoting system actually predict?#

An AI freight quoting system predicts two numbers for each request: the buy rate, meaning what a capable carrier will accept to haul the load, and the sell rate a customer is likely to accept. The margin, and the choice to quote, pass or price high, come from the gap between them.

Better systems also predict risk around those numbers: the chance a load falls off after booking, the chance of detention or a truck-ordered-not-used charge, and how long coverage will take. Each prediction needs examples where the outcome is known, which is why a broker's own history matters so much.

Speed is the visible benefit, since an agent can draft a reply to an emailed spot request without a rep looking up lanes by hand. The less visible benefit is consistency: the same lane, lead time and customer get priced the same way regardless of which rep opens the email.

Quote inputs and why each matters#

Quote inputs fall into three groups: what the load is, who is asking, and what happened after the quote. Most brokers capture the first group well, the second partly and the third rarely, yet the third is where a model learns.

The table lists the inputs a quoting model draws on and where they usually live in a brokerage's systems.

Market rate data versus your own history#

Market rate data and a broker's own history answer different questions, and good quoting systems use both. Subscription market rate services show the going rate on a lane across many participants; your history shows how your customers, carriers and facilities behave relative to that rate.

A broker that relies only on market data prices like everyone else. The edge comes from knowing that one shipper awards at a premium for reliability, or that one receiver's dock generates detention on most deliveries.

Why lost quotes matter as much as booked loads#

Lost quotes matter as much as booked loads because they show where the price was too high, too slow or wrong for the customer. A model trained only on booked loads sees prices customers accepted but never the prices they rejected, so it cannot learn where the line between winning and losing sits on a lane.

Most brokerages keep booked loads in the TMS, but quotes often live in email threads, spreadsheets or a rep's memory. Closing that gap is usually the most valuable data change a brokerage can make before buying or building a quoting tool.

• 1Every quote request with a timestamp, lane, equipment, dates and the requesting company.
• 2The quoted rate, any revisions and who quoted it.
• 3The outcome: won, lost, no response or pulled by the customer.
• 4A short loss reason where one is known, such as price, timing or capacity.
• 5The link from a won quote to its TMS load number and the final carrier settlement.

Data problems that quietly mislead a quoting model#

The data problems that most often mislead a quoting model are inconsistent lane definitions, missing accessorials and rates recorded without context. Each one makes the history look cleaner than the business really was.

Lanes keyed by city in one year and by zip prefix the next look like different lanes. Accessorials billed later on a separate invoice never reach the load record, so margins look healthier than they were. Contract loads mixed with spot loads, without a flag, teach the wrong price for a spot request. Fixing these in the export, rather than rewriting the source system, is usually enough.

Illustrative: a brokerage connects quotes to settlements#

Illustrative: a fictional mid-size freight brokerage runs a broker TMS for loads and carrier settlements, but its reps quote spot freight from Outlook and track contract bids in spreadsheets. Market rate subscriptions sit in a browser tab beside the inbox.

The president wants an AI quoting tool, and the vendor asks for history. The team finds that booked loads go back many years, but older quotes are scattered across mailboxes. They build a quote log, back-fill it from email for the busiest lanes, and link each won quote to its load number and carrier cost.

The quoting tool now learns from wins and losses rather than bookings alone, and reps see a suggested buy and sell rate with the history behind it. The president also asks whether the linked history could be licensed, which starts a rights review of customer and carrier agreements.

Can rate and load history be licensed to AI developers?#

Rate and load history can often be licensed to AI developers once customer and carrier confidentiality is handled, because linked quotes, awards, costs and service outcomes are what freight and pricing agents need. The broker keeps ownership; the history is licensed, not sold outright.

Customer contracts and carrier agreements frequently treat rates as confidential, and some prohibit disclosure in any form. Preparation typically tokenizes customers and carriers, generalizes facility locations to markets and may band rates or express them as changes rather than absolute values. Figures taken from market rate subscriptions stay out, because those subscriptions may restrict redistribution. Counsel should review the agreements before any rate data is scoped.

Within the SourceX five-step transaction, a brokerage starts with a fit check built on metadata alone: which TMS holds the loads, how many years are accessible and whether quotes were captured. Rights review then works through customer and carrier terms, the broker approves the final scope, and the SourceX Evidence Packet captures each record's source, the rights relied on, the permitted use, the privacy steps taken and the release sign-off.