By Luis Lopez, AI transportation consultant, CEO of Go Hub.io Holdings Corp and subsidiaries, and host of the Freight Guru Podcast
Many transportation management systems now advertise AI features. The label appears on quoting screens, load matching, document capture, tracking and reporting. Some of these features are genuinely useful. Others are a rebranded rule or a lightly adjusted report. The difference is rarely visible from a product page, and it usually does not show up in a polished demo either.
As an AI transportation consultant, I suggest treating every AI feature in your TMS like a new hire on probation. You would ask what it does, what it is allowed to touch, how you will know when it is wrong, and who is accountable. These ten questions are the ones I would put to any provider, whether the feature is already in your system or arrives with the next update.
The ten questions
1. What exactly does this feature do, and what is it not meant to do?
Ask for a plain description of the task, the inputs it uses and the output it produces. A clear answer sounds like “it reads inbound rate confirmations and fills load fields.” A vague answer sounds like “it optimizes your operation.” Also ask for the limits. A provider who can describe where the feature does not work usually understands it better than one who cannot.
2. Is it actually AI, or is it automation with a new name?
Fixed rules, such as “send an alert if a load is two hours late,” are useful but are not machine learning. That is fine, as long as you know what you are paying for. Ask whether the feature learns from data, what kind of model sits behind it, and whether it behaves differently over time. If you need a refresher on the distinction, Generative AI vs. Predictive AI in Freight explains the two broad families.
3. What data does it learn from and use to make decisions?
Find out whether it uses only your data, pooled data from other customers, or outside sources. This affects both accuracy and confidentiality. A feature trained mostly on other companies’ lanes and customers may not reflect how you operate. Also ask what fields it depends on. If it needs clean equipment types and accurate location data and yours are inconsistent, the results will reflect that.
4. Who owns my data, and can it be used to train models for others?
Read the contract, not just the sales deck. Ask whether your shipment, pricing and customer data can be used to improve a shared model, whether that can be switched off, what happens to your data if you leave, and how it is protected in the meantime. Pricing and margin data deserve particular care.
5. How will I know when it is wrong?
Every AI feature makes mistakes. What matters is whether you can see them. Look for confidence indicators, audit trails showing what changed and why, and easy ways to flag errors. A feature that silently edits records is far riskier than one that proposes changes for approval.
6. Can a person review or override it, and is that the default?
Ask where the human sits in the process. For low-risk tasks such as sorting documents, automatic action may be acceptable. For anything involving money, customer commitments, carrier selection or compliance, the safer design has the system recommend and a person confirm. Check that overrides are simple and that they are recorded.
7. How is accuracy measured, and can I test it on my own freight?
Be skeptical of a single headline accuracy claim. Ask what was measured, on whose data and over what period. Better still, request a trial using your real loads and documents, and compare the results with what your team does today. Measure the cost of errors as well as their frequency, because a rare mistake on a high-value shipment can outweigh many small wins.
8. What happens when the information is incomplete or unusual?
Freight is full of exceptions: oversize loads, unusual accessorials, reconsignments, missing paperwork. Ask how the feature behaves when it meets something it has not seen. The right answer is that it flags the case and hands it to a person. The wrong answer is that it guesses.
9. What does it cost, and what does it replace?
AI features may be bundled, priced per use, or sold as add-ons. Clarify the full cost, including implementation, training time and any per-transaction charges that grow with volume. Then ask what work it is supposed to remove. If nobody can name the tasks it will eliminate or speed up, you are paying for a promise rather than a process change. For a wider look at selecting providers, see How to Evaluate an AI Vendor for Trucking and Logistics.
10. How does it fit with the rest of my systems and my team?
A feature that works in isolation but forces double entry elsewhere will not last. Ask how it connects to your accounting, tracking and customer systems, and what training your staff will need. Also consider who will own it internally. Someone should be responsible for reviewing its performance regularly, adjusting settings and deciding when to turn it off. If you are unsure whether your organization is ready, Top 10 Signs a Trucking Company Is Not Ready for AI is a useful self-check.
How to use these questions
You do not need to ask all ten about every feature. Match the depth of scrutiny to the risk. A feature that suggests a report layout deserves a quick look. A feature that sets customer rates, tenders loads, approves carriers or releases payments deserves the full list.
A practical way to apply them is to build a short scorecard for each AI feature. Record the answers, note who gave them, and revisit the scorecard after a month of real use. Compare what you were told with what you observed. If the provider’s answers and your experience diverge, that gap is the most important information you will collect.
Warning signs worth noting
- Answers that rely on buzzwords rather than describing a process.
- Reluctance to let you test with your own data.
- No way to see or export a log of what the feature did.
- Claims of improvement with no stated baseline or measurement method.
- Pressure to enable full automation immediately instead of phasing it in.
Start small and keep records
When you do switch on an AI feature, begin with one team, one customer group or one lane set. Keep a record of what it did and what your staff changed. After a few weeks you will have your own evidence, which is worth more than any benchmark a provider can show you. Expand when the numbers justify it, and be ready to roll back when they do not.
The aim is not to be skeptical of AI for its own sake. It is to make sure that each feature earns its place in your operation by doing something specific, measurably well, with clear accountability when it does not.
For more practical discussion on AI and freight, 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.
