By Luis Lopez, AI transportation consultant, CEO of Go Hub.io Holdings Corp and subsidiaries, and host of the Freight Guru Podcast
AI can do useful work in freight. It can read documents, draft emails, predict arrival times and flag risky loads. It can also be oversold. The gap between a polished demonstration and a tool that works on your freight, with your data, on a busy Tuesday, can be wide.
The ten signs below are patterns to watch for during a sales process. None of them proves a product is bad. Each one is a reason to slow down and ask a harder question.
1. It Claims to Replace Your Whole Team
Freight operations involve judgment, relationships and constant exceptions. Tools that take over specific, repeatable tasks tend to deliver. Tools pitched as a replacement for dispatchers, brokers or customer service staff as a group are describing an outcome, not a product. Ask which exact tasks the tool performs from start to finish without a person, and which ones it only assists with.
2. The Demo Uses Perfect Data
A demonstration built on clean sample data shows the best case. Your data has missing fields, duplicate customer records, and rate confirmations in a dozen formats. Ask to see the tool run on a sample of your own documents and shipments, including the messy ones. If the vendor resists, that tells you something. Our guide to the data fields to clean before buying AI explains what to prepare.
3. Accuracy Is Quoted as a Single Number With No Context
An accuracy figure means little unless you know what was measured, on which data, and how an error was defined. A document reader that reads the large majority of fields correctly can still produce an unusable result if the one field it misses is the rate or the container number. Ask how accuracy is measured, on what kind of documents, and what the error rate looks like on the fields that matter to you.
4. Nobody Can Explain What Happens When It Is Wrong
Every AI system makes mistakes. A serious vendor has thought about this and can show you confidence scores, review queues, and how an error is caught before it reaches a customer or an invoice. If the answer to “what happens when it gets one wrong” is that it rarely happens, the failure handling has probably not been built.
5. Integration Is Described as Easy
Connecting a new tool to a transportation management system, an accounting package, telematics feeds and customer portals is often the hardest part of a project. It is where timelines slip. Ask which systems the tool already connects to in production, whether the connection reads and writes or only reads, and who does the integration work. Our list of questions to ask about AI features in your TMS covers this in more detail.
6. Return on Investment Is Guaranteed Up Front
Results depend on your volume, your current process, your data and how well your team adopts the tool. A vendor who has not studied any of that cannot know what you will save. Be cautious with savings figures presented before discovery. A better sign is a vendor who proposes measuring a baseline first and then defines success in terms you both agree on.
7. The Word “AI” Covers Everything
Rules engines, statistical forecasts, optimization solvers and language models are different technologies with different strengths. Plenty of valuable freight software is rules-based, and there is nothing wrong with that. The concern is vagueness. If a vendor cannot say which technique handles which task, they may not understand their own product’s limits. Our explainer on generative AI versus predictive AI in freight is a useful primer before those conversations.
8. There Is No Human Review Step for High-Stakes Actions
Some actions are low risk to automate, such as sending a routine status update. Others carry real consequences: booking a carrier, quoting a customer, submitting a customs entry, or issuing an invoice. A tool that performs high-stakes actions with no approval step, no audit trail and no way to set limits is asking you to accept risk the vendor will not carry. Ask which actions can be gated behind human approval and how every automated action is logged.
9. Questions About Your Data Get Vague Answers
Your rates, customer lists and shipment history are commercially sensitive. You should get clear written answers to basic questions:
- Where is the data stored and who can access it?
- Is it used to train models that serve other customers?
- Can you export it in a usable format if you leave?
- How long is it kept after the contract ends?
Hesitation on any of these is a reason to pause until the answers are in the contract.
10. The Contract Is Long but the Pilot Is Not Offered
A vendor confident in the product is usually willing to prove it on a limited scope before asking for a multi-year commitment. Be wary when the only option is a long contract with large setup fees and no defined exit. Ask for a time-limited pilot on one lane, one customer or one document type, with written success criteria and a clear right to walk away.
What a Realistic Pitch Sounds Like
It is easier to spot overpromising when you know what the opposite looks like. A credible vendor tends to:
- Name the specific task the tool performs and the tasks it does not
- Ask detailed questions about your data and systems before quoting results
- Show the tool failing gracefully, not only succeeding
- Describe the work your team will have to do during setup
- Offer reference conversations with operations people, not only executives
- Agree to measure results against a baseline
As an AI transportation consultant, the first thing I look at in any AI proposal is how the vendor talks about limits. A vendor who can describe where the product struggles usually understands it well enough to support it.
A Simple Way to Test the Claims
- Pick one narrow problem. Choose a task that is frequent, measurable and annoying, such as entering rate confirmations or answering tracking requests.
- Measure it today. Record how long it takes and how often errors occur.
- Run the tool on real work for a fixed period. Keep a person checking the output.
- Compare. Look at time saved, errors introduced and how much review was needed.
- Decide with your own numbers. If the tool performs, expand. If it does not, you have lost little.
For a fuller checklist, see our guide on how to evaluate an AI vendor for trucking and logistics.
The Takeaway
Overpromising is not unique to AI, but the technology makes it easier because demonstrations can look impressive on a narrow sample. Healthy skepticism does not mean avoiding AI. It means asking for evidence on your own freight, keeping people in the loop where the stakes are high, and committing in steps instead of all at once.
For more straightforward analysis of AI in freight and logistics, 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.