The hype says AI is ready to take over whole business processes end to end. The reality is easy to see. AI isn't ready to run your dealership. How do we know? If it could, somebody you know would already be doing it. They'd have fired all their staff, turned on the robots, sold more cars, and bought a condo in Miami with the extra money.
I'm not an AI doomer. I think it changes most of how we work for the better. The point is that you are not late to the party if you're just starting to figure out where AI fits in your store. The technology is moving at breakneck speed. Picking which parts to adopt can feel like betting on the ponies without a tip sheet. It doesn't help that "AI" means something different depending on who's talking.
Most of the time people mean a large language model, the thing behind Claude, ChatGPT and Grok. But voice recognition and speech synthesis, image recognition, image generation and a newer class of decision models that work on confidence scores are all called AI too, and they're different tools. Naming them separately is the first step toward deciding which ones belong in your store.
Instead of treating AI like an advancing cloud of uncertainty, understanding the pieces lets you decide which ones to bring in to help the team you already have. The work is getting done today with or without AI. If you choose the pieces deliberately you keep control of the business and you can measure whether you are getting the results you want.
Think of it as a dial
A store might start with an AI phone line that only takes messages after hours. Once the BDC trusts what shows up in the morning, turn it up: let it book service appointments. Turn it up again and it can take a full sales inquiry and hand it to a person with the summary attached. Nobody turns it all the way to "sell the car", and nobody should. That's how I built Sundial, and it's how I'd tell any store to approach this regardless of what they buy.Where AI is helping today
Voice is getting really good. Remember the first AI images you saw, with six-fingered hands and words in Klingon? Now it's hard to tell them from a photograph. Voice is on the same curve. The newest models from Google and OpenAI are full-duplex, so the agent listens and speaks at the same time instead of waiting for silence. Paired with a quick reasoning model, they hold a real conversation. On a dealer line that means the 7pm caller gets a full intake instead of voicemail, and the BDC gets a summary and a recording instead of a missed-call notification.Prospect summaries. An LLM can read and classify a huge amount of content instantly. Instead of a sales meeting where people hem and haw about where each deal stands, you get a summary of every sales call this week, ranked, with the ones where a manager should step in flagged. Coaching becomes continuous instead of being a thing you do when you have time.
Scheduling and automation. Letting an agent touch your systems of record still requires some care. Guardrails in the prompt help, but an LLM is a probability machine, so anything sensitive should go through a deterministic layer: real tools with real rules, not instructions the model is supposed to follow. The agent can book a service appointment because there's a booking tool behind it that checks the calendar. The list of things your agent is not allowed to say is more important than the list of things it is allowed to do.
AI is here to stay. The frontier labs and the existential debates will grab the headlines, but for people trying to get real work done there is plenty to put to work today. A good AI strategy is mostly a matter of getting past the parlor tricks and focusing on where your actual team and your actual process can use a boost.
Andrew Smith builds Sundial, a RevOps and UCaaS platform for dealerships: sundial.click
