AI Hallucinations in Fleet Management: Why Verification Still Matters

I've been thinking a lot about AI lately.

Not because it's the newest technology. Not because everyone seems to be talking about it.

Because I think we're at an interesting point.

I've said before that I believe AI is going to have one of the biggest impacts our industry has ever seen. I really do. I think it's going to make data more accessible, help fleet managers answer questions they never had the time or resources to answer before, and remove a lot of the administrative work that keeps people from focusing on leadership.

But something happened recently that reminded me of the other side of that conversation.

During a live recording of the Fleet Success Show at GFX, Facundo Tassara and I asked Claude what we thought was a pretty simple question: Had any fleet ever won both NAFA's 100 Best Fleets award and Government Fleet's Leading Fleets award in the same year?

The AI answered immediately.

It sounded confident.

It even sounded convincing.

The only problem was...it was wrong.

The fleet it named had won both awards. Just not in the same year.

That distinction was the entire question.

Now, fortunately, we caught it right there on camera. Nobody made a decision based on it. Nobody got hurt.

But I've been thinking about what would have happened if that same type of mistake showed up somewhere else.

What if it was a maintenance recommendation? A replacement analysis? A budget justification? A compliance report?

Fleet isn't a profession where many of our decisions stay inside the fleet office. They eventually end up in front of finance, leadership, auditors, elected officials, or sometimes even attorneys.

That's one of the reasons I don't think the conversation should be about whether AI is good or bad.

I think that's the wrong conversation.

The better question is this: How do we use AI responsibly?

In my experience, the answer hasn't really changed from how we've always approached good fleet management.

We trust good information. But we verify it.

That's true whether the recommendation comes from a vendor, a consultant, another fleet manager, or now...an AI model.

One thing that's encouraging is that the technology is getting better quickly.

When AI is connected to verified information instead of relying only on what it learned during training, hallucinations drop significantly. More and more organizations are also building human review into their AI workflows because they recognize that important decisions still deserve another set of eyes.

That makes a lot of sense to me.

In fact, it's very similar to the way we've been thinking about AI inside RTA.

Rather than assuming one AI response is automatically correct, we've been building systems that validate one process against another before the answer ever reaches the fleet manager.

The objective isn't to slow anyone down.

It's actually the opposite.

It's to remove as much uncertainty as possible before someone has to make a decision.

Because I don't think fleet managers should spend their day wondering whether the software got something right.

They should be thinking about how to serve their organization better.

So what does that mean for those of us using AI today?

I've found myself coming back to three pretty simple habits.

First, I treat AI like an outstanding first draft.

Not the final answer.

Second, I always want to know where the information came from.

An answer grounded in my own fleet data is very different than an answer generated from general knowledge.

And finally (and this one might be the most important), I verify anything that's easy to verify.

Names.

Dates.

Dollar amounts.

Historical facts.

Interestingly, those are often the places where AI sounds the most confident while occasionally being the most wrong.

None of this makes me less excited about AI.

Actually...

It makes me more excited.

Because I don't think AI is replacing fleet professionals.

I think it's giving good fleet professionals more leverage than they've ever had before.

I've said before that AI still needs your expertise. It still needs your judgment. Fleet is a specialized profession, and context matters. That's true today, and I suspect it'll still be true for quite some time.

The tools will continue getting better.

Our responsibility doesn't change.

As fleet professionals, we've always been stewards of information before we were stewards of assets.

AI doesn't change that.

If anything...

I think it makes that responsibility even more important.

This article was inspired by a recent episode of our podcast. Check out the full episode for even more tips and tricks: