By Ariel Coro, founder of Dealer Growth Hackers and publisher of Dealer AI Guy. Sources are linked throughout; vendor-reported numbers are flagged as vendor-reported. Photo by Alex Knight on Unsplash.

TL;DR

  • Your customers got AI agents first. CarEdge's agents mystery-shopped 100 stores and only 33 answered with a full out-the-door price, while 84 percent of buyers told Accenture they're open to letting AI handle the negotiating.
  • The proven dealer-side jobs are phone and BDC coverage, lead follow-up, and service scheduling. CDK's data says 47 percent of sales appointments and 61 percent of service appointments still book by phone.
  • Reporting, predictive targeting, and desking agents are real but live or die on your data quality and integration. Spyne calls this the execution gap, and it's where most projects stall.
  • AI-assisted lending, hiring, advertising, and automated pricing answers carry legal exposure. Keep transcripts, review them, and control what the agent is allowed to say.
  • Buy like an operator: audit your foundation first, fund one job completely instead of six pilots, and assign a human owner who reads transcripts weekly.

Every vendor in this industry now sells an "AI agent." Some of them are selling a chatbot with a new sticker on it. A few are selling something that will genuinely answer your phones at 2 a.m., book the service appointment, and log the whole thing in your CRM. The problem is that from the demo, you can't tell which one you're looking at.

This guide is the piece I wish someone had handed me before the term got hijacked. It covers what an AI agent actually is, the jobs agents are doing in real stores right now, the numbers behind those claims, the integration trap that eats most of these projects, and the legal exposure almost nobody brings up in the sales meeting. It's long on purpose. Bookmark it, forward it to your GM, and argue with it in the next manager meeting.

One more thing before we start: while you evaluate AI agents for your store, your customers are already using their own. That fact should change how you read everything below.

What an AI agent actually is (and what it isn't)

A chatbot waits for a question and produces an answer. That's it. If the shopper asks something weird, it produces a weird answer, and the conversation dies in the widget.

An agent is different in one specific way: it can take actions to pursue a goal. It can look up live inventory, check the service schedule, send the follow-up text, push a note into the CRM, and escalate to a human when it hits its limits. The test is simple. If the software can only talk, it's a chatbot. If it can talk and do, it's an agent.

That distinction matters because the value isn't in the conversation. Nobody's gross improved because a widget chatted politely. The value shows up when the software completes work your people are dropping: the missed call, the 11 p.m. lead, the third follow-up that never happened, the appointment that never got booked.

Keep that test handy. Half the "AI agent" products pitched to dealers fail it.

Your customers got agents first

Here's the uncomfortable order of operations: buyer-side AI got useful before most dealer-side AI did.

In June, at the first AutoIndustry.ai Summit, CarEdge's Zach Shefska sent AI agents to mystery-shop 100 dealerships live, while he was on stage. The agents emailed and texted stores asking for out-the-door pricing, then logged every reply. We covered the results in detail in our analysis of the 100-dealer mystery shop: 92 of 100 stores replied within 24 hours, 62 gave some price information, and only 33 produced the full out-the-door breakdown the agent asked for.

Read those numbers again. The response rate is fine. The answer quality is the problem. A human shopper might tolerate a payment quote when they asked for an out-the-door number. An AI agent logs it as a non-answer, asks again, and compares your store against the nine others that answered cleanly. CarEdge is already turning that data into its Dealer Transparency Index, which scores stores on fee transparency and add-on behavior based on verified quotes its agents collected.

And shoppers want this. An Accenture study covered in Automotive News found that 84 percent of car buyers are open to AI handling complex tasks like negotiating the deal, and 42 percent said they'd let an AI agent make the final purchase decision for them. We flagged that stat in the daily brief when it dropped, along with CarEdge's public claim of over $20 million in aggregate buyer savings from agentic negotiation, averaging about $2,405 per multi-round negotiation. Those savings came out of somebody's gross.

So when you evaluate an AI agent for the store, you're not adopting a gadget early. You're catching up to the other side of the desk. I wrote a longer strategic piece on where this ends up, AI Will Negotiate Your Customer's Next Car Deal, if you want the full argument.

The jobs AI agents are actually doing in dealerships

Forget the category names vendors use. In real stores, dealer-side agents are doing about seven jobs today. Some are proven. Some are promising. Some are mostly slideware. Here's the honest read on each.

The AI BDC and phone agent

This is the strongest case right now, because the phone is still where dealerships bleed.

CDK's latest research, which we covered in early August, found that 47 percent of sales appointments and 61 percent of service appointments are still booked by phone. Every missed or fumbled call is pipeline walking out the door, and most stores fumble more than they think. That's not a new problem. What's new is that software can now answer every call, every time, without hold music.

The results from stores doing this are hard to ignore. Paul Sansone Jr.'s auto group runs an AI-powered BDC in its service department that answers every incoming call with no wait time and converts roughly 65 percent of those calls into scheduled appointments, a number he shared with CBT News. Voice follow-up vendors are riding the same wave: Lokam.ai reports 45 percent month-over-month growth, and one of its dealer clients says it added $20,320 in gross profit in its first 21 days just by re-engaging dead showroom leads. Both of those are vendor-and-dealer-reported numbers, so apply the usual salt, but the direction is consistent across every store I talk to: phone coverage is the fastest payback in dealer AI.

If you do one thing after reading this article, pull your missed-call report for last month and multiply it against those CDK percentages. That's the size of the hole an agent would plug.

Compare the companies in this lane in the AI BDC & Phone Agents directory, each with the questions to take into the demo.

The sales agent that works your leads

The second proven job is lead engagement and follow-up: the agent that responds to the 11 p.m. internet lead while it's still warm, answers real questions with real inventory, qualifies the shopper, and books the appointment before a competitor's BDC opens.

The case study making the rounds here is Johnson Honda, written up by CBIZ: autonomous chat agents that captured leads, qualified shoppers, booked test drives, and ran follow-up drove a 27 percent increase in showroom appointment rates and a 26 percent lift in lead-to-sale conversion. We covered it in the brief alongside a detail worth stealing: the store treated the agent as a process change, not a plugin. The agent owned the lead until it either booked or escalated.

The failure mode is just as instructive. If your agent quotes a different number than your website, or dodges a direct pricing question with "when can you come in?", the buyer-side agents from the previous section will log every dodge. Template-and-stall died the day machines started reading the replies.

Vendors doing this job are cataloged in the directory under Sales & Lead Follow-Up.

The website agent

The search data that led me to write this article shows dealers hunting for exactly this phrase: "car dealership website AI agent." What they're really asking is whether the chat widget can finally earn its spot on the page.

The bar to clear: a website agent should see live inventory, real incentives, and your actual service schedule, and it should complete transactions a shopper starts, not just collect a phone number and quit. A widget that answers questions from a stale FAQ is a 2019 product with a 2026 label. And there's a second audience to think about now. As Fullpath's Auto Intelligence Index found, large language models are becoming a significant referral source for car shoppers, which means your site is increasingly read by machines shopping on someone's behalf. We flagged that study in the brief. If your pricing is inconsistent between the VDP, the widget, and the autoresponder, you now have three tools contradicting each other in front of a buyer that never forgets.

We built a free Dealer Website Grader that checks whether your site's technical foundation can even support this kind of shopper. Run it before you buy a website agent, because an agent bolted onto a slow, schema-less site inherits all of its problems.

The Website & Digital Retail directory tracks the platforms and agents competing for this job.

The service drive concierge

Fixed ops is quietly where agents fit best, because service interactions are structured: book, remind, status, upsell, follow up. Machines are good at structured.

The after-hours numbers make the case on their own. Remember that 61 percent of service appointments come in by phone. Now look at who's answering at 7:40 a.m. when the line is six deep at the drive. Sansone's 65 percent call-to-appointment conversion happened in service, not sales. New entrants keep piling into this lane too; in one week we logged both a DealerRefresh launch of an after-hours agent handling WhatsApp, email, and missed calls around the clock, and Reynolds naming BDC accountability and 24-hour triage among its tech-competition finalists. The service scheduling agent has gone from novelty to table stakes faster than any other category on this list.

Service-lane vendors live in the directory under Service & Fixed Ops.

The reporting and DMS agent

Dealers are also searching for "AI agents for dealership management systems" and "AI agent for dealership reporting," and I understand the longing. The dream is asking plain-English questions of your own store: which service advisors are underperforming on alignment penetration, which lead sources actually cash deals, what did we make on the last twenty units of a model.

Be careful here. This category is real, but it depends entirely on data access, and your DMS was not built to share. I've written before about the DMS duopoly and the $129.5 million antitrust settlement that came out of data-access litigation; the incentives that produced that lawsuit haven't gone anywhere. A reporting agent is only as good as the pipes behind it, and in most stores those pipes are the problem. Ask any vendor selling this exactly which fields they pull, how often, and what happens when the DMS changes its export format. Watch how fast the demo confidence evaporates.

The platforms and systems of record are profiled under Data, DMS & Platforms.

The predictive targeting and equity agent

"AI predictive targeting for dealerships" shows up in the search data in four different phrasings, which tells you the marketing has landed even where the product hasn't. The pitch: an agent that watches your DMS and service lane for customers in an equity position or nearing lease-end, and works those lists automatically.

The mining math has always been sound, and this was "AI" in dealerships before the current wave. What's changed is the acting part: instead of printing a list for the BDC to ignore, an agent opens the conversations and books the appointments itself. Lokam's dead-lead reactivation number above is this category wearing a different hat. It works when the data underneath is clean. It embarrasses you when it isn't; nothing torches trust like congratulating a customer on a car they traded in eight months ago.

Predictive targeting and equity players map to Marketing & Advertising in the directory.

The desking and negotiation agent

This is the frontier, and the fog is thickest here. Startups are pitching agents that structure deals and even negotiate with the buyer's agent directly. Some real product exists: badco.ai launched a desking platform this year, and CarEdge is running agent-to-agent negotiations from the consumer side today. But before you let software quote structured deals, read the compliance section below twice. A tool that quotes the wrong all-in price at scale isn't a glitch. It's a liability multiplier, and the FTC's pricing-transparency push means the receipts will exist.

The frontier players, buyer-side included, are under Buyer-Side & Desking.

The integration trap: why most dealer AI projects stall

Spyne's Auto Retail Intelligence Quarterly put a name on the thing I see in stores every week: the execution gap. Its argument, which we've tracked across several briefs, is that the AI divide is no longer adoption versus non-adoption. It's connected versus bolted-on. By 2027, stores that embed AI across CRM, DMS, inventory, F&I, marketing, and service will pull away from stores running a drawer full of disconnected AI gadgets.

The disconnected version has a cost you're already paying. AutoSuccess calls it the frankenstack tax: duplicate data entry, reporting that never reconciles, and tools that each hold a different version of the customer. Add an unsettling stat from CBT News: the average dealer now has more than nine vendors communicating with consumers, and most stores aren't centrally capturing what any of them said. Every AI agent you add without integration is another voice speaking for your store that nobody is supervising.

So the boring questions are the whole ballgame. Does the agent write to your CRM or around it? Does it read live DMS data or a nightly copy? When it books a service appointment, does your scheduler actually see it? Who gets the transcript, and where does it live when a customer disputes what your store told them? A GM I respect, Kevin Pitts at BMW of Reading, got so fed up with vendor bloat that he built his own stack. You don't have to go that far. But his instinct was right: own the wiring, or the wiring owns you.

The legal part nobody demos

Nobody sells you the compliance slide, so here it is.

CBT News ran a warning in late July that every GM should read before the next vendor demo, and we led the brief with it: AI-assisted lending decisions, hiring, and advertising are all areas where a dealership can violate discrimination or advertising law by accepting AI recommendations without human review. Jim Ganther of Mosaic Compliance Services has been making the same point on the industry circuit: AI raises dealership liability, and it requires more oversight, not less, because it operates at scale.

Scale is the key word. A salesperson who misquotes a price creates one problem deal. An agent that misquotes prices creates a documented pattern, timestamped, in writing, across every conversation it ever had. The same federal pricing-transparency scrutiny that CarEdge's transparency scoring feeds on applies to your automated quotes too. If an AI is answering pricing questions for your store, you need to know what it said, be able to prove it, and be able to shut it off the moment it drifts.

Three questions to put in every vendor conversation, verbatim: What does the agent do when it doesn't know the answer? Where are the transcripts stored and for how long? Who reviewed the pricing, financing, and disclosure language it's allowed to use? A vendor who answers those crisply has thought about your risk. A vendor who pivots back to the demo has not.

None of this is a reason to sit out. It's a reason to buy like an operator instead of a fan.

What about budget?

The honest answer on pricing is that the market hasn't settled, and anyone quoting you a universal number is guessing. What I can tell you is how the wind is blowing. Kerrigan Advisors' fourth annual OEM Survey found 59 percent of OEM executives projecting that AI will increase future dealership profits, a result we broke down in the brief. When the factories expect AI to raise store profitability, factory programs and co-op money tend to follow, and so does pressure to show progress.

The budgeting mistake I keep seeing isn't overspending. It's spreading spend across six disconnected pilots instead of funding one job done completely. A store that fully solves phone coverage, integrated to the scheduler and the CRM with transcripts reviewed weekly, will beat a store running six half-configured trials every time. Fund jobs, not categories.

How to evaluate an AI agent before you sign

A short field guide, earned the hard way:

Start with your foundation, not their product. Most AI disappointments are foundation problems wearing an AI costume. Bad data, broken processes, and a slow website will sink any agent you hire. We publish a free Dealer AI Readiness Checklist that scores where your store actually stands, and a deeper 75-point audit skill in our free, MIT-licensed Dealer AI Skills collection. Run one of them before any vendor meeting. It changes the conversation.

Shop from a map, not from inbound pitches. We keep an independent Dealer AI Company Directory: every vendor we track, organized by job, with our coverage history on each profile and no pay-to-play. Start there instead of with whoever emailed you last.

Demand the agent test. Can it take actions, or only talk? Make the vendor show the action trail: the CRM entry, the booked appointment, the escalation to a human, live, in their own system.

Mystery-shop it. Vendors demo the happy path. You should shop the unhappy one. Ask their live deployment at another store a pricing question, a trade question, and something off-script, at 9 p.m. on a Sunday. You're doing to them exactly what CarEdge's agents did to those 100 dealerships. Better you find the cracks than a buyer's agent with a scorecard.

Check the wiring before the brains. Integration list, in writing: which CRM, which DMS, which scheduler, read or read-write, real-time or batch. The smartest agent with no hands is a kiosk.

Assign a human owner. Every store where I've seen agents work has one person who reads transcripts weekly, tunes what the agent may say, and owns the escalation path. The agent is a hire. Nobody hires a BDC rep and skips the one-on-ones.

Where this goes next

Both sides of the desk are getting agents, and the two curves are going to meet in your internet department. Buyer agents are already scoring stores on answer quality. Dealer agents are already answering phones and booking appointments at rates humans don't sustain. The stores that win the next few years will be the ones whose systems can talk to both: machine-readable answers for the buyer's agent, machine-executed follow-through from their own.

The window where this counts as early adoption is closing. The mystery shop already happened. The transparency scores are already being compiled. Your customers already told Accenture they're comfortable sending a machine to do the deal. The only question left is whether the machine that answers is yours, and whether it makes your store look sharp or sloppy.

Start with the readiness audit, fix the phones, and wire in one agent properly. Then read this again in six months and see how much of the "frontier" section has become table stakes. Based on the last twelve months of writing the daily brief, my money says most of it.

Frequently asked questions

What is an AI agent for a car dealership? Software that can pursue a goal by taking actions, not just answering questions. A dealership AI agent can look up live inventory, answer pricing and service questions, book appointments, update the CRM, and hand off to a human when needed. If it can only chat, it's a chatbot, whatever the label says.

What's the difference between an AI agent and a chatbot? Actions. A chatbot produces answers; an agent completes work: booking the appointment, sending the follow-up, logging the conversation. The practical test in any demo is to ask the vendor to show the action trail in their own systems.

What do AI agents actually do in dealerships today? The proven jobs are phone answering and BDC coverage, internet lead follow-up, website engagement, and service scheduling. Reporting agents and predictive targeting are real but depend heavily on your data quality. Desking and negotiation agents are the frontier and carry the most compliance risk.

Do AI agents for dealerships actually work? The store-level numbers we've covered include a service BDC converting about 65 percent of answered calls into appointments and a Honda store reporting a 27 percent lift in showroom appointment rates from autonomous lead engagement. Results depend on integration; disconnected tools underperform everywhere.

Are AI agents a legal risk for dealerships? They can be. AI-assisted lending, hiring, and advertising can trigger discrimination or advertising violations if nobody reviews what the AI recommends, and automated pricing answers fall under the same transparency scrutiny regulators apply to human quotes. Keep transcripts, review them, and control what the agent is allowed to say about price and financing.

How should a dealership start with AI agents? Audit your foundation first, then fund one job completely instead of several pilots. Phone coverage is usually the fastest payback given how much sales and service volume still books by phone. Assign a human owner who reviews transcripts and tunes the agent weekly.

Free dealer AI tools

Start with the foundation: install Dealer AI Skills, then run Dealer Website Grader to find technical website leaks and DealerAEOAudit to see whether AI assistants recommend your dealership.

See all free tools →

Reading about AI won't train your team

The stores pulling ahead aren't the ones with the most tools — they're the ones whose people know how to use them. That takes a strategy, a readiness baseline, and real training.