By Luis Lopez, AI transportation consultant, CEO of Go Hub.io Holdings Corp and subsidiaries, and host of the Freight Guru Podcast
Every freight brokerage knows its gross margin. Fewer know which loads actually make money once everything that happens after the rate confirmation is counted. A load can show a healthy spread between the customer rate and the carrier pay and still lose money after a detention bill that never got rebilled, a claim, three hours of rework and a customer who pays late while the carrier was paid early.
Load profitability scoring uses data and AI to estimate the true contribution of each load, and then rolls those results up by customer and lane. The goal is not a prettier report. It is finding the specific relationships and lanes where the business is quietly working for free, and deciding what to do about them.
Gross margin is not true profitability
Gross margin is the customer rate minus the carrier cost. It is easy to calculate and it is the number most brokers manage to. The problem is everything it leaves out:
- Accessorials not rebilled. Detention, layover, lumper, driver assist and redelivery charges paid to the carrier but never billed to the customer, or billed and then disputed and written off. See truckload accessorial charges explained for the common ones.
- TONUs. A truck order not used is commonly billed when a load is canceled after a truck is dispatched or arrives. If you pay the carrier but cannot recover it from the shipper, it comes straight out of margin. Background is in what is a TONU in trucking.
- Claims. Cargo claims, short-pays tied to damage allegations, and the staff time spent documenting and resolving them.
- Rework. Loads that get re-covered after a carrier falls off, rescheduled appointments, corrected paperwork and rebilled invoices.
- Payment terms and quick-pay. When a customer pays well after you pay the carrier, you are financing the gap. Quick-pay programs change that math in both directions. The basics are covered in freight payment terms, quick pay and net 30.
- Staff time. Some customers need constant check calls, custom reports and appointment juggling. Others tender clean loads and stay out of the way. Both show up as the same margin on paper.
None of these costs is exotic. They are simply scattered across systems, so they rarely get tied back to the load that caused them.
Combining the data
True load profitability requires stitching together at least three sources.
The TMS
The transportation management system holds the load record: customer, lane, equipment, dates, customer rate, carrier pay, accessorials entered, appointment changes and status events. It is the backbone, and the load ID is the key that everything else should join to.
Accounting
Accounting holds what actually happened to the money: invoice amounts, short-pays, write-offs, credit memos, days to pay, carrier payment dates and quick-pay fees. Often the TMS shows one number and accounting shows another. The difference is usually where the profit leaked.
Email and messages
This is where AI earns its keep. Detention approvals, rate disputes, claim notices and “can you move the appointment again” requests live in email threads. Language models can read those threads, link them to load IDs and tag events such as detention requested, accessorial approved, dispute raised or appointment rescheduled. Counting messages and touches per load also gives a rough proxy for staff time without asking anyone to fill out timesheets.
Scoring loads, customers and lanes
Once the data is joined, the scoring itself is straightforward. Each load gets an estimated net contribution: gross margin, minus unrecovered accessorials and TONUs, minus claim costs, minus an allocated cost of staff time, minus the financing cost of the gap between paying the carrier and getting paid. Then roll it up.
- By customer: average net contribution per load, variance, share of loads that lost money, and the main driver of losses.
- By lane: which origin and destination pairs consistently need re-covers, generate detention or sit at facilities that run late.
- By customer and lane together: a customer can be profitable overall while one of their lanes drains money every week.
AI is useful in two specific places here. First, it can classify the reason behind each loss, so the output is not just “this customer is unprofitable” but “this customer loses money mostly on unrecovered detention at two receivers.” Second, it can predict the likely net contribution of a new load at the quoting stage, based on the history of similar loads, so the pricing desk sees the risk before it books. Keep that prediction inside sensible pricing guardrails so a model does not quietly reprice your book.
Acting on the results
As an AI transportation consultant, I find that the score is the easy part. What matters is what the team does with it.
- Fix your own process first. If accessorials are not being rebilled because nobody captures the approval, that is an internal problem. Tighten the capture and billing steps before blaming the customer.
- Reprice where the data supports it. If a lane reliably needs extra effort or generates detention, the rate should reflect it. Show the customer the pattern rather than just sending a higher number.
- Change terms where needed. Clear accessorial and cancellation language in rate confirmations and customer agreements reduces disputes later.
- Have honest conversations. Many shippers do not know their facilities hold trucks for hours or that their tenders change repeatedly. Bring specific load examples and ask for operational changes, not just more money.
- Decide deliberately. Some low-margin customers are worth keeping for volume, network balance or strategic reasons. The point is to make that choice on purpose, with the real numbers in front of you.
Data-quality limits
A profitability score is only as good as the joins underneath it. Watch for these problems:
- Missing load IDs on invoices, payments and emails, which makes matching unreliable.
- Accessorials paid off-system or lumped into a single carrier payment line.
- Staff time proxies that treat every email as equal effort. A one-line confirmation is not a two-hour dispute.
- Small samples on new lanes or customers, where one bad week can make a good account look terrible.
- Timing: claims and short-pays can surface weeks after delivery, so recent loads look better than they will end up.
- Misread emails: language models can mislabel a message or link it to the wrong load. Spot-check the tagging regularly.
Show confidence alongside every score, keep the underlying load list one click away, and treat the output as a starting point for a conversation rather than a verdict.
The bottom line
Gross margin tells you what you priced. Load profitability scoring tells you what you kept. By combining TMS, accounting and email data, AI can surface the customers and lanes that lose money and explain why. The decisions that follow, repricing, process fixes and direct conversations with customers, still belong to people who understand the relationships.
For more conversations on freight and AI, subscribe to the Freight Guru Podcast.
About the author: Luis Lopez is a Miami-based AI transportation consultant and logistics entrepreneur, the CEO of Go Hub.io Holdings Corp and subsidiaries, and host of the Freight Guru Podcast.