By Luis Lopez, AI transportation consultant, CEO of Go Hub.io Holdings Corp and subsidiaries, and host of the Freight Guru Podcast
Freight companies hear the word “AI” constantly, usually attached to a product pitch. The trouble is that the label covers two very different kinds of technology. One writes and reads. The other forecasts and scores. If you buy one expecting the other, you will be disappointed.
As an AI transportation consultant, the first question I ask about any tool is simple: does it produce language, or does it produce a number? That one question sorts most of the market.
The short version
- Generative AI creates new content, mostly text, but also summaries, drafts, and structured data pulled from messy documents.
- Predictive AI learns patterns from historical data and estimates what is likely to happen next, such as a transit time, a price, or the odds that a load goes wrong.
Both are statistical systems. Neither understands your business the way a seasoned dispatcher or broker does. Both are only as good as the data and the process around them.
What generative AI is good at in freight
Freight runs on unstructured information: emails, rate confirmations, bills of lading, proof of delivery photos, carrier packets, and long email threads about a single shipment. Generative models are strong at turning that mess into something usable.
Reading and summarizing documents
A model can pull the pickup address, commodity, weight, and reference numbers out of a document that arrives in an inconsistent format. It can also condense a forty-message email thread into a short status note. If you want more detail on the document side, see this breakdown of how AI is changing freight document processing.
Drafting routine communication
Quote follow-ups, appointment requests, status updates, and customer replies follow patterns. A generative tool can draft them for a person to review and send. The time saved is real, but only when someone checks the draft before it goes out.
Answering questions from your own documents
Internal guides, carrier requirements, and customer routing instructions often sit in shared folders nobody reads. A generative assistant connected to those files can answer a question in plain language and point to the source.
Where generative AI goes wrong
Generative models can produce text that sounds confident and is simply wrong. In freight, that matters. A made-up detention rule or an invented reference number is worse than no answer, because it looks authoritative.
- They can state things that are not in the source document.
- They are poor at exact arithmetic unless connected to a calculator or a rating engine.
- They do not know your contract terms unless you give them the contract.
The fix is process, not hope. Require a human to approve anything that creates a financial or legal commitment, and make the tool cite the document it used.
What predictive AI is good at in freight
Predictive systems are built on numbers: past shipments, past prices, past transit times, past delays. They look for patterns and project them forward.
Estimated arrival and delay risk
Given enough history on a lane, a model can estimate when a truck is likely to arrive and flag shipments that appear to be slipping. This is the idea behind predictive ETA and freight visibility.
Pricing and demand estimates
Predictive models can estimate what a lane is likely to cost or how tight capacity may be. They are a guide for a pricing decision, not a replacement for judgment. The related topic of quoting is covered in AI freight quoting and pricing.
Matching, scoring, and anomaly detection
Scoring a carrier for risk, matching a load to the trucks most likely to cover it, or flagging an invoice that looks unusual compared with past ones are all predictive tasks. They produce a score or a ranked list, not a paragraph.
Where predictive AI goes wrong
A predictive model assumes the future looks like the past. When something breaks that assumption, the output degrades quietly.
- Thin history. A lane you run twice a year gives the model almost nothing to learn from.
- Dirty data. If appointment times, locations, or charge codes are entered inconsistently, the patterns the model finds are partly noise.
- Sudden change. A port disruption, a weather event, or a major shift in the market can make last quarter’s pattern a poor guide.
- False precision. A prediction shown as a single clean number hides the range of likely outcomes.
Data quality is the recurring theme. I would read what AI cannot do when your freight data is poor before signing anything.
A simple way to choose
Ask what the output of the work looks like today.
- If the output is a message, a summary, a form, or an answer to a question, start by looking at generative tools.
- If the output is a forecast, a score, a ranking, or a flag, look at predictive tools.
- If the work involves both, such as reading an invoice and deciding whether it looks wrong, expect to use both, with a person reviewing exceptions.
Questions to ask before you buy either
- What data does it use, and who owns that data after the contract ends?
- What happens when the tool is unsure? Does it say so, or does it guess?
- Can I see why it produced a given answer or score?
- Where does a human approve or override the result?
- How do I measure whether it is working, using my own numbers rather than the seller’s?
The last question matters most. A tool that cannot be measured against a baseline you already know is a tool you cannot manage.
Where the two work together
The most useful setups combine them. A predictive model flags a shipment that is likely to miss its delivery window. A generative tool drafts the customer notice and summarizes the reason. A dispatcher reads both, decides what to do, and sends the message. Each piece does what it is good at, and a person stays accountable for the decision.
That division of labor is the practical lesson. Use generative AI for language and documents, use predictive AI for estimates and scores, and keep people in charge of judgment calls. If you want a framework for vetting sellers, how to evaluate an AI vendor for trucking and logistics is a good next read.
If you found this useful, subscribe to the Freight Guru Podcast for more plain-language conversations about freight and AI.
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.
