Where does AI get partner and carrier rates from?

Because the answer decides whether the product works on your business or somebody else’s.

Quoting looks like one job. It is really two. One is arithmetic: take a rate you have, apply your rules, produce a number. The other is diplomacy: work out who might price this lane, ask them, wait, read what comes back, and ask again.

Software has been good at the first for years. The second has resisted everybody, because it is not a data problem. It is a relationship, conducted by email, with no obligation on anyone to reply.

The checklist puts it plainly: nobody has fully solved agent-network rate sourcing, so the honest answer is a map, not a claim. That is why B1 has two halves. “Where do you get rates from” invites a list of logos. “What happens when the rate is not there” tells you what the software does on a Tuesday afternoon when your customer wants a number for Dar es Salaam.

Before you can test a vendor’s answer, you need your own. Most forwarders have never written it down.

Sort your lanes into four buckets. Not by mode or customer, but by where the number comes from.

Now count. Put a bucket against each of your top ten lanes. Most forwarders who do this are surprised twice: by how much sits in bucket four, and by how much of bucket one is out of date.

That count is your evaluation. A vendor who covers buckets one and two brilliantly is a strong buy for a forwarder whose business is mostly bucket one, and close to useless for one whose business is cross-trade through agents. Neither comes out of a demo.

Six reasons, and none is about how clever the software is.

• The rate does not exist until somebody makes it. An agent’s price for a shipment is a decision, made when asked, based on their space and how much they want your business. There is nothing to fetch.
• The reply is not a rate, it is a paragraph. It arrives with inclusions that differ by agent: terminal handling at both ends, documentation, ISPS, chargeable weight rounding, free time. Two prices for the same lane are often not comparable, and the gap is where the margin goes.
• There is no standard to plug into. DCSA’s published standards list names bill of lading, booking, track and trace, VGM, arrival notice, port call and vessel schedules, described as common data models for digital interoperability between carriers, ports, terminals, shippers and freight forwarders. Pricing is not among them. In air, IATA’s ONE Record is described as a standard for data sharing that creates a single record view of the shipment. Pricing is not in that scope either.
• Published references cover very few lanes. Drewry’s World Container Index is a weekly spot assessment across eight major East-West routes, widely used as an independent reference for index-linked contracts. Your enquiry is door to door, on a lane nobody indexes.
• Validity is short and the clock is unforgiving. A partner quote good for seven days is a different object from a contract good for six months. Software that stores it as a rate will quote it three weeks later, which is worse than no rate at all.
• It is reciprocal. You quote for their inbound, they quote for your outbound. A tool that blasts a thousand requests at your network damages the network.

The second column is not hopeless. It is a communication problem rather than a lookup problem, and the two need different promises from a vendor.

A map, not a claim. A map of your lanes rather than theirs.

A strong answer has four parts, and you can ask for each by name:

• 01What is covered, by bucket. Which of the four they can price without a person, in your buckets rather than in general.
• 02The number on your top ten lanes. The checklist’s own follow-up: how many of my top ten would you have a rate for on day one? Any answer other than a number is not an answer.
• 03What happens when the rate is not there. Does a person get a task, does the agent get an email, does the enquiry just sit.
• 04What it costs you to improve the coverage. Rate cards to upload, credentials to hand over, an agent list to clean up. Ask how many hours from your side.

Four, and the first is the one the checklist warns about.

Vague talk about integrations

A long list of connections is not a coverage answer. Ask what each one returns: a rate, a schedule, a booking, or a page a person still has to read. A login and a screen is fine, and it is not a rate you can quote from.

A carrier count instead of a lane count

A large number of carriers means little if none of them price the lane you sell. Bring your ten lanes and ask about those.

An index mistaken for a rate

Market references tell you where a lane sits. They do not tell you what you can sell it at.

“A person handles it” described as a feature

A fallback where a person does the work is not wrong, and for bucket four it may be the only truthful answer. What matters is whether that person is yours or theirs, and whether the chase happens without anybody remembering.

One more, which is an asymmetry rather than a trap: a demo runs on a lane with excellent coverage. The twenty enquiries you choose under question A1 should include the three lanes that give your team the most trouble.

Nobody can conjure a rate that has not been made yet. What a product can do is take the asking off your team.

That is the shape of an honest answer to the second half of B1, from any supplier. As an example, here is ours, in the wording we publish elsewhere on this site.

• Emails all your agents for a lane in one go, with the route and charge lines filled in, then chases the ones who do not reply.
• Where a lane has to be priced by another agent or over the phone, the wait belongs to whoever is answering — not to the software. FreighAI still writes the RFQ, chases the reply, and turns whatever comes back into a priced, margin-checked quote the moment it lands.

The shape matters more than the vendor. It does not claim to know the rate. It claims to remove the asking, the waiting and the re-typing, and it says out loud that the delay belongs to the third party. That is the most anyone can honestly promise on bucket four.

Hold us to it in the same words. If any vendor, us included, describes bucket four as solved, the checklist has told you something.

One page. Four rows, one per bucket, one column per vendor, filled in the same day.

Then decide with a rule you set in advance. Bucket four is never a reason to reject a vendor. A claim to have solved it always is.

Two things belong in the pilot rather than the meeting. Put your three worst lanes in it deliberately. And agree in writing what “covered” means before anybody counts: a lane is covered when a quote can go out without a person leaving the product to find something.

Finally, run B1 on yourself. How long does an agent rate take today, who chases it, and how many enquiries die in that gap. That is what the software is competing against, and it is usually worse than anybody believes.