By Luis Lopez, AI transportation consultant, CEO of Go Hub.io Holdings Corp and subsidiaries, and host of the Freight Guru Podcast
Most writing about artificial intelligence in freight describes what the technology can do. That is useful, but it is only half of the picture. As an AI transportation consultant, the first thing I look at is the other half: what the tool cannot do, and what it needs from the business before it can do anything at all.
This is not an argument against AI. It is an argument for buying it with open eyes. A carrier, broker or shipper who understands the limits will get far more out of the technology than one who expects it to fix a messy operation on its own.
AI Is Only as Good as the Data Behind It
Every AI system in logistics, whether it quotes freight, matches loads, predicts arrival times or reads documents, learns from or acts on data. If that data is incomplete, inconsistent or wrong, the output will be too. The model will simply be wrong faster and with more confidence.
Freight data is harder to keep clean than most people assume. Common problems include:
- Duplicate records. The same customer or facility entered five different ways, each with its own history.
- Free-text fields. Critical details such as appointment requirements or accessorial notes typed into a comments box where no system can reliably use them.
- Missing timestamps. Arrival and departure times that were never captured, or were entered hours later from memory.
- Stale reference data. Facility hours, contact names and lane rates that were accurate two years ago.
- Inconsistent units and codes. Weights in pounds on one record and kilograms on another, or equipment types described differently by each dispatcher.
If your transportation management system has these problems, an AI layer on top of it will inherit all of them. The plain, unglamorous work of cleaning and standardizing records usually produces more value than the model itself. If you are still choosing a core system, the basics in what to look for in TMS software matter more than any AI feature bolted onto it.
What AI Cannot Do Well Today
1. It cannot know what was never recorded
A great deal of freight knowledge lives in people’s heads. The receiver that turns trucks away after 2 p.m. even though the appointment system says 4. The dock that cannot take a 53-foot trailer. The customer who always approves detention if you call but never if you email. None of that is in a database, so no model can learn it. Until that knowledge is written down in a structured way, AI will keep making decisions a veteran dispatcher would never make.
2. It cannot take responsibility
An AI tool can draft a rate, suggest a carrier or flag a document, but it cannot be accountable for the result. When a load is tendered to the wrong carrier or a quote is sent below cost, the business owns the outcome. That is why every automated decision with real financial or legal consequences needs a named human who reviews it or at least reviews the exceptions. A signed rate confirmation binds the company whether a person or a program generated it.
3. It cannot reliably handle the true exception
Models perform best on situations that look like the data they were built on. Freight produces a steady stream of situations that look like nothing before: a port closure, a sudden regulatory change, a customer bankruptcy, a storm that reroutes an entire region. In those moments, pattern-based predictions can become actively misleading because the pattern has broken. Human judgment is what carries an operation through.
4. It cannot build or repair a relationship
Freight still runs on trust. A shipper gives the difficult load to the broker who answered the phone at midnight last time. A driver takes the bad lane for the dispatcher who has treated them fairly. AI can handle routine updates and free up time for those conversations, but it cannot have them for you. Operations that automate away every human touchpoint tend to find out that the touchpoints were the product.
5. It cannot guarantee it is telling the truth
Language-based AI systems can produce answers that are fluent, specific and wrong. In freight that might be an invented regulation, a misread weight on a bill of lading, or a confident summary of a contract clause that does not say what the summary claims. Any workflow that relies on generated text or extracted data needs verification steps, especially for compliance, claims and anything a customer will rely on.
6. It cannot fix a broken process
If check calls are skipped, paperwork arrives late and billing happens whenever someone gets to it, automation will not create discipline. It will automate the chaos. The operations that benefit most from AI are the ones that already had a defined process and want it to run faster.
Where the Limits Show Up in Practice
These limits are not abstract. They show up in specific, predictable places:
- Pricing tools that quote confidently on lanes where the company has almost no history, because nothing told the model its data was thin.
- Predicted arrival times that ignore a facility’s real unloading behavior because dwell time was never captured.
- Document readers that handle clean, typed forms well and struggle with handwritten notations, stamps and exceptions written on a delivery receipt, which are often the details that matter most in a claim.
- Carrier-matching tools that recommend a carrier based on price and location without knowing about a service failure that was discussed by phone and never logged.
I covered the positive side of several of these use cases in the top 10 ways AI is changing LTL and truckload shipping. Both views are true at once: the capability is real, and so is the dependency on good inputs.
A Realistic Way to Set Expectations
Before adopting any AI tool, work through a short list of questions:
- What specific decision or task will this handle? “Make us more efficient” is not an answer. “Draft quotes for standard lanes for a person to approve” is.
- What data does it need, and do we actually have it? Pull a sample of real records and look at them before you assume.
- What happens when it is wrong? Identify the cost of an error and who catches it.
- How will we measure the result? Decide on the measurement before you start, using your own numbers from before the change.
- Who owns it? A tool with no internal owner drifts out of use within months.
Start with a narrow task where mistakes are cheap and easy to spot. Expand only after the results hold up for a meaningful period.
The Bottom Line
AI in freight is a real and useful set of tools, not a replacement for clean data, sound process and experienced people. The companies that get value from it treat it as an assistant that needs supervision and good information. The ones that are disappointed usually expected it to supply the discipline their operation was missing.
For more plain-spoken analysis of freight and technology, 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.
