The AI-for-Dealers Field Guide

I have sat in more vendor demos than I care to count. The pitch is always the same: a slick dashboard, a few GPT-generated summaries, and a number that looks impressive until you ask where it came from. Most dealers walk out of those rooms more confused than when they walked in. This guide is the one I wish someone had handed me ten years ago.

What "AI in automotive" actually means right now

Strip away the marketing and AI in automotive today is four things.

Language models that read and write. Drafting lead responses, summarizing calls, answering after-hours questions. This is the most mature category and the first one worth your attention.

Transcription and conversation intelligence. Turning every call and meeting into searchable text. Quietly the highest-leverage tool in the building, because your store already generates thousands of recorded conversations that nobody reviews.

Prediction and scoring. Which lead closes, what the trade is worth, where the price should sit. Useful, older than the hype, and only as good as the data feeding it.

Agents. Software that takes multi-step actions on its own. Real, moving fast, and mostly not ready to run unsupervised in your store. Anyone selling you a fully autonomous BDC in 2026 is selling the demo, not the deployment.

That is the whole map. Every product you see this year is some mix of those four, plus a markup. The question is never whether it is AI. The question is which of the four it is, and whether the work it does is worth what they charge.

The vendor theater checklist: five questions that end the hype fast

Vendor theater has one tell: the demo is amazing and the specifics are fog. Five questions burn the fog off fast.

1. Where does this number on your dashboard come from? Make them walk the path from raw data to the figure on screen. Theater breaks here more often than anywhere else.
2. Which model runs this, and what is yours versus rented? There is no shame in building on a frontier model. There is shame in calling a thin wrapper proprietary AI.
3. What happens when it is wrong? Ask for the failure story and the recovery process. A vendor with no failure stories has either no customers or no honesty.
4. Can I talk to a store like mine that has run this for a year? Not a logo slide. A phone number.
5. What do we own when we leave? Data, transcripts, configurations, history. If the answer takes more than a sentence, you already have it.

You are not trying to embarrass anyone. You are checking whether the product survives contact with specifics. The good ones do, and the good ones respect you more for asking.

What actually moves revenue: the three use cases I trust

My standard is simple: I only recommend what I have built, run, or measured myself. Three use cases clear that bar.

Conversation intelligence. Your store records thousands of calls and meetings nobody reviews. A manager samples two or three. AI can review all of them. At Strolid I built an internal meeting-intelligence pipeline in TypeScript and Python that turned recorded conversations into searchable, accountable records, and I am steering that work toward open source as an AI safety, governance, and accountability tool. The pattern holds anywhere people talk for a living, and a dealership is exactly that.

Drafting and response consistency. Language models writing the first draft of every lead response, with a human editing and sending. I will not quote you an invented closing-rate lift. What I will say is that the speed and consistency problem in lead handling is twenty years old, I watched it cost deals back in my Lazare and Lia days, and drafting is the first version of that problem AI genuinely solves at scale.

Market intelligence. Knowing where your pricing, inventory, and visibility sit against the market this week, not last quarter. This is the problem I am building Vandoko MIP around at vandoko.ai, live in beta: a market intelligence platform combining Vandoko-AI, On-the-Map, and the Scrapey-Do engine.

The five plays I would run in your store this quarter

All five run this quarter, with off-the-shelf tools, for less than the cost of most single vendor contracts.

Play one: transcribe everything for thirty days. Every sales call, every BDC call, the Saturday meeting. Do nothing fancy with it yet. Just read it. What your customers actually ask, and what your team actually says, will rewrite your training calendar on its own.

Play two: lock one number stack before you buy anything. One written definition of a lead, one sale date, one approved source list the whole store uses. AI pointed at disagreeing numbers automates the disagreement.

Play three: AI drafts, humans send. Put a language model on lead-response and follow-up drafts, with your BDC editing and sending every message. You get the speed without handing your reputation to an autoresponder.

Play four: a weekly AI readout of play one. Every Friday, one page: the top objections, the questions nobody answered, the appointment asks that never happened. It is the cheapest sales training you will ever buy.

Play five: a weekly market snapshot. Your top five models against the market: price position, days on lot, what competitors moved. One page, every Monday, same source every time.

Run all five for ninety days before you take another vendor demo. You will walk into the next one knowing exactly what you need, which is the one thing vendor theater cannot survive.

How to evaluate before you sign: a one-page scorecard

One page, six lines, each scored 1 to 5.

• Data ownership: what we keep when we leave, and in what format.
• Model transparency: they can name what is theirs and what is rented.
• Reference quality: a store like ours, a year in, willing to take the call.
• Exit cost: what leaving actually takes in dollars and weeks.
• Time to value: days to the first measurable output, not months of onboarding.
• Workflow fit: it works inside how the store already sells instead of demanding the store reorganize around the tool.

My rule: anything under 24 total is a pass, and anything that scores a 1 on data ownership is a pass no matter the total, because every other line on the card is rented the moment you cannot leave.

Print it. Bring it. Score in the room while the demo runs. The discipline of the card matters more than the math.

Closing: own the system or rent the illusion

Over twenty years in this business, from an independent shop in 2005 to building AI platforms today, I am credited with helping generate a reported $2.4B in dealer profit. None of it came from a tool nobody could explain. The dealers who win the next five years will not be the ones with the most AI subscriptions. They will be the ones who own their data, their definitions, and their workflows, and rent only what is replaceable.

If you want the longer version of how I build that, the work page holds the record. If you want it in your store, pricing is one click and one honest conversation away.