AI will revolutionise your pricing – or expose internal problems

It takes an agent about four seconds to do what used to take a pricing analyst hours. It reads the incoming request, checks the request against a rate card, applies the markup, and drafts the response before most inboxes have refreshed. Forwarders are already running this on inbound RFQs. Trucking fleets use it on spot capacity. Airline cargo teams, whose pricing runs on a different logic, are catching up too. 

When the pricing underneath is solid, this is a genuine step change. Quotes go out in seconds instead of hours, while margins hold without discretionary discounts. Sales teams get their time back for the accounts that need a human conversation. 

However, when it is not solid, the agent applies your flaws faster and more often than any analyst ever could. It has no instinct for a surcharge that has not been updated since spring, or a markup set by someone (often by gut feeling) who left the company two years ago. It simply executes what it is given. So, one question to think about before you read on: If a pricing error went out at scale tomorrow, would anyone in your organisation notice before your customers do? 

A market that shows its hand 

Ocean and air freight pricing is unusually visible for a B2B market. Base rates are benchmarked daily through Xeneta and the Freightos indices for ocean, or the TAC Index for air, paired with surcharges that follow conventions the whole industry recognises. A large shipper does not need to trust your number, as they can solicit three competing quotes on the same lane in the same week and know within the hour whether you overpriced it.

Rather than fixing their pricing structures, the industry's long-standing answer has been to out-relationship it. A trusted account manager who secures priority space in a capacity crunch, or picks up the phone when something goes wrong, can make a client forgive an inconsistent quote that transparent benchmarking would otherwise expose. Personal judgement has been compensating for fragmented pricing for years, leading to a rigid, outdated infrastructure underneath. This is the legacy arrangement agentic AI now disrupts. Precisely because an agent has no relationship to lean on and no judgement to fall back on, it exposes exactly how much of your pricing quality was being carried by people compensating for the system, rather than by the system itself. In our work across T&L, that is the first thing automation reveals, regularly surprising leadership teams. 

Get the pricing backbone in place before automating 

Automated quoting processes change what a pricing team needs to get right. AI system integration inherently reduces the number of stakeholders who are able to catch what the system gets wrong. As a result, once an agent is generating quotes at scale, the quality of every quote depends entirely on the quality of the data and logic sitting behind it. 

Removing human safeguards introduces challenges, as freight prices are rarely a single number. A typical ocean quote stacks a base rate negotiated with the carrier, a forwarder margin on top, and a series of surcharges like fuel indices, bunker adjustments, terminal handling, peak season where relevant, and currency adjustment factors. Each component moves on its own update cycle, and in a well-run setup, each has a named owner. 

In most companies, however, that logic lives in spreadsheets and in people's heads. Two analysts or sales reps apply the same rate card slightly differently, and both believe they are right. But because a human is in the loop, they can discuss it and reach a final decision, which keeps the inconsistency small enough to work with – often unnoticeable even to management. An agent, in comparison, cannot do the same. It needs a clear pricing convention specified in writing before it can quote at all. 

In practice, agents require dedicated, structured inputs to arrive at a transparent decision. Machine-readable pricing starts at one rate card structure instead of five interpretations of it, a named owner, a validity period for every charge item, and a markup logic that exists somewhere other than in the minds of the people applying it. Further, companies need to construct guidelines on how exceptions are logged and valued. 

The same discipline applies to the data underneath. A human analyst who sees a customer spelled in two different ways across the ERP and the TMS reconciles it without thinking. Conversely, an agent treats them as two different customers and repeats that error on every inquiry it touches. 

Fixing such problems is unglamorous work and requires reconciling customer and lane identifiers, standardising the surcharge taxonomy, or validating the rate data the agent reads from. None of it is difficult per se, but it must all happen before the agent goes live for the simple reason that an agent cannot correct its own inconsistencies the way a person can.