A new survey report from LotLinx asserts that in its current form, “AI is broken for dealers.”

The company contends the era of “one-size-fits-all artificial intelligence” is over for auto dealers. The recent survey of 215 executives from both franchised and independent dealerships shows that 84% of dealers say they “often” or “almost always” fail to get what they need from generative AI tools.

And over half of survey responders say their most critical need is “better understanding of data and inventory risk” to maximize ROI.

Prescott Tweedy, senior vice president of product at LotLinx, talked with Auto Remarketing in late April about the survey results and the implications for the future of AI at dealerships.

As for its current use, Tweedy said it’s “a bit like using a supercomputer to do a fraction.” For example, using AI, dealers can feed their Toyota Camry descriptions into a generative AI tool and ask it to optimize them for consumer engagement.

“They’re (dealer AI tools on the market) really doing these surface-level tasks. And the future of AI is really on the agentic side,” said Tweedy.

Instead of just asking the AI tool a question, which it can certainly respond to, Tweedy contends these tools should be able to run prompts and tell dealers how they are bleeding margin, whether their markdowns are working for them or against them, and more.

“Because there’s a disconnect in the data between most LLMs (Large Language Models) and chat interfaces and what’s actually on your lot, it tends to give you these generative responses,” said Tweedy. “And so that’s why people tend to stay in that generative ecosystem, because they’re used to getting generative responses, and they don’t necessarily understand the power if you start to expand usage.”

That’s one of the challenges, he said, as there needs to be a little bit more handholding.

“We can’t expect dealers to be hyper-technical engineers,” Tweedy said.

A market waiting to be served

Tweedy said there is a market waiting to be served, rather than a market opportunity.

“When I think about market opportunity, you’re really trying to go out there and convince people that this is a thing. And we’ve had LLMs for two and a half to three years now, since OpenAI opened our eyes to these things. And then Gemini came around,” Tweedy said. “So I think it’s less and less now about the market opportunity. People know these things exist. You don’t have to convince them to use it.”

Now, it’s about servicing the market in a way that actually helps dealers run their businesses. Dealers know the tools exist; according to Tweedy, they are just waiting for businesses to help them identify how these tools can solve some of their biggest challenges.

This includes a better understanding of demand, inventory, and all the components that go into managing a dealership day-to-day. However, currently, Tweedy said most AI offerings for dealers are what LotLinx calls “inventory blind,” meaning they cannot understand which vehicles matter to specific markets and dealerships.

“LLMs are trained on a huge basis of knowledge. And so when they see the words Toyota Camry, they think of it more as a noun instead of how a dealer thinks about it, which is really an asset. So they can tell you a bunch of things about, oh, this Toyota Camry, it happens to be red, etc.,” said Tweedy. But that’s where it ends.

Because it doesn’t have connections into, for example, a dealer’s customer relationship management system or learning management system, the tool lacks some of that fundamental knowledge that a dealer really understands about what a Toyota Camry means to them and their market.

“Without connecting those data points, the LLM that they’re using is probably not going to be effective,” said Tweedy.

In other words, an LLM is only as powerful as the data you give it.

“So there hasn’t been something that could, for example, take all of the little things that make the Toyota important, the deal, all the little CPO facts, all the little things that make that car that dealer’s car, that can’t be translated,” he said.

That’s where the AI auto vendor market is going, Tweedy said: focusing on how to transform huge amounts of data specific to the dealership, to the market, and other factors.

“It doesn’t help to say, ‘Hey, you’re priced 120% to market when you’re looking at a national average,’” said Tweedy. “We need to get hyperspecific into your local geo. That 120% national is likely spot-on at 99% in your market. And so I think that’s where we tend to see that there’s some blind spots.”

If you’re not giving the LLM the structure and the knowledge base and feeding it the data for it to understand your business and market, it’s going to continue to give you “quasi-correct answers, something that sounds like it feels correct, but isn’t,” said Tweedy.

This comes down to an architectural mismatch between what generative AI was built to do and what dealers need it to do.

When OpenAI started talking about ChatGPT, it was that GPT part that got everybody really enthusiastic. It’s truly like a next-word predictor and does a good job of “sounding human.”

“When we think about architectural mismatches, it’s really about getting away from being a next-word predictor and really trying to orchestrate now into being a next-action predictor,” said Tweedy.

Tweedy said this is one of the next steps in the evolution of AI: How do you take something that can give you really good responses and turn that into problem-solving?

“That’s the break in the chain that we’re trying to solve for … taking something good at responses and then turning it into something that can actually take action on those responses,” said Tweedy. “So, the next thing is creating the agentic flows that then say, ‘I can do this work for you to give you back the time to focus on the customer experience inside of your dealership, to focus on all those other things that drive value inside of your own system.’”

Chatbots are passive tools, waiting for you to move and take action.

“And what we want is to move into more proactive approaches, like ‘Hey, I’ve given you my data, tell me when I need to take action,” said Tweedy.

That means moving away from a passive logic loop toward tools that analyze inventory in real time and provide actionable intelligence.

“What we’re trying to figure out, and what everybody is trying to figure out is how do you take all these disparate data sources, clean them, get them into one sort of functional system that now knows all of the connective glue at your dealership and can highlight risk in front of you instead of waiting for you to say the ship is on fire,” said Tweedy. “The hope is that we can prevent that ship from ever catching fire.”

The agentic path forward

Tweedy thinks it’s just a matter of time, but that timeframe could range from a few months to up to five years before the AI market meets what dealers really need.

“I think you’re going to start to see a lot of players come into the space that claim to do the things that people want them to do, and it’s going to be proving that out,” said Tweedy. “I think the market is ready to bring some things to market, and we have to prove that these things do the things that we tell people they can do.”

It’s no longer enough to spend a bunch of tokens on chats, prompts, and responses; it’s actually about taking action inside of the flows and things dealers want these systems to do for them to give them back the time they need.

The catch is that these vendors need to be able to bring these systems to dealers in a cost framework that works for their business.

Are dealerships ready themselves for this agentic path forward?

“I think a lot are not. I think that’s up to us as partners with these dealers of getting away a little bit from being a service provider and getting more towards being a trusted partner that helps you to understand how these systems can work to your benefit,” said Tweedy.

LotLinx has been developing its own offering, LotGPT, to “check all these boxes.” The tool is designed to analyze dealers’ live inventory alongside real-time market supply and shopper demand.

“We’ve trained our model to be dealer thought-focused. And so now it’s just about using the data that you give it. It’s training against your own data to give you responses,” said Tweedy.

For example, LotGPT will optimize vehicle descriptions and immediately import them into vAuto for participating dealers.

“And now in 30 seconds or less, all of your descriptions go from this really generic SEO, or sometimes it’s even just a bunch of words, into a little bit more consumer-facing and a little bit more engaging in that way,” said Tweedy of the tool that is currently being piloted.

And if the dealer gives LotLinx access to its Google AdWords platform, LotGPT adjusts bidding strategies in real time, adjusts keywords, creates campaigns, and can add or drop vehicles from campaigns.

“It’s about creating efficiencies in some of these marketplaces that for a long time have been set and forget,” said Tweedy. “And so that’s where we’re really starting to focus our efforts, which is being on the agentic frontier, trying to be first to market and proving that all these things can actually do what we tell you they can do.”