Transforming Automotive Retail with Agentic AI

The average franchise dealership in the United States spent $543,539 on advertising last year, according to NADA's 2024 Annual Financial Profile. That's roughly $45,000 a month flowing into Google Ads, social campaigns, third-party marketplace listings, OEM co-op programs, and broadcast media. For a group with 50 stores, the annual marketing budget easily exceeds $25 million.

It's working. The traffic is showing up. The 2025 Cox Automotive Car Buyer Journey Study found that 95% of car shoppers now use online resources as their primary research tool, spending an average of 14 hours online before they ever set foot in a dealership. Google's own research confirms that twice as many buyers start their journey online as at the dealership.

And then what happens?

The buyer lands on a Vehicle Detail Page. They scroll through stock photos. Maybe they click a payment calculator that asks for a credit score before it shows a number. Maybe a chat widget slides in from the corner and asks "How can I help you today?" — and they close it because they've learned from experience that it can't actually help. They compare two or three vehicles in their heads because the site doesn't offer a way to compare them any other way. They leave.

That is the reality for roughly 95 out of every 100 dealer website visitors. Industry benchmarks put the average automotive website conversion rate — the percentage of visitors who submit a form, start a chat, or make a phone call — somewhere between 2% and 5%, depending on how you measure it and who you ask. Shift Digital's 2025 Pulse Report, which analyzed billions of shopping sessions across thousands of dealer websites, confirmed that dealer sites are outperforming third-party marketplaces in engagement, but the absolute conversion numbers remain stubbornly low.

The math is brutal. All that money getting buyers to the website. And the website itself fails to do the one thing it's supposed to do: sell.

The $25 Million Leak

I've been obsessing over this gap for the last two years, and I've come to believe the problem is not traffic, not SEO, not even the website design. The problem is that nobody sells on the website.

Think about what happens on the showroom floor. A good salesperson reads the buyer. They ask about family size, driving patterns, budget. They walk over to two vehicles and explain the tradeoffs in plain language. They sit down, calculate a real monthly payment with today's rates, factor in a trade, and book a test drive for Saturday morning. The entire experience is personalized, visual, and guided.

Now think about what happens on the website. The buyer gets a search bar, a grid of inventory tiles, and a handful of CTA buttons that all lead to forms. "Get Your ePrice." "Schedule a Test Drive." "Value Your Trade." Each one asks for personal information before it gives the buyer anything useful in return. The experience is transactional before it's consultative. It's the opposite of how people actually buy.

Brian Pasch, who has spent nearly two decades studying automotive digital marketing and whose CRM and CTA research reports are the industry's most cited benchmarks, has been quantifying exactly how deep this problem runs. His 2026 CRM Research Report surveyed over 330 dealerships and found that CRM reporting capabilities scored an average of just 6.5 out of 10 — what Pasch called "a failing grade." His work on Form Conversion Rate — FCR, the percentage of visitors who click a CTA button and actually complete the form — reveals that the conversion tools on most dealer websites are part of the problem, not the solution. Slow-loading forms, intrusive popups, and calculators that demand personal information before delivering anything useful are actively driving buyers away. At a typical dealer website converting between 1.5% and 3% of visitors into leads, that means for every 100,000 monthly visitors, 97,000 or more leave without taking a single action.

The question every dealer group should be asking is not "how do I get more traffic?" It's "why is the traffic I already have leaving without buying?"

What Chatbots, Voice AI, and Digital Retailing Got Right — And Where They Stopped

Credit where it's due. The industry has tried to solve the website conversion problem, and every attempt has moved the needle somewhere.

Chatbots introduced the idea that the website shouldn't be a silent brochure. Companies like Gubagoo, LivePerson, and others gave dealers a way to engage visitors in real time. The problem is that chatbots answer questions. They don't sell. When a buyer asks "Should I get the Tucson or the Santa Fe for my family?" a chatbot sends a link to a comparison page. A good salesperson would ask about family size, second-row headroom, weekend plans, and towing needs — and then walk them through both options with a recommendation and a payment. Those are two fundamentally different experiences, and the conversion data reflects it.

Voice AI tackled a different piece of the puzzle. Platforms like Numa, Toma, and others have done genuinely strong work on inbound call handling, service appointment booking, and reducing the epidemic of missed calls at dealerships. Some are now expanding into sales and outbound. One has even tapped into vehicle telematics data, triggering AI calls when a service light goes on. That's real innovation. But voice AI doesn't touch the website experience. The 70% of buyers who research online at 11 PM on a Tuesday aren't picking up the phone. They're browsing the dealer's site. Alone.

Digital retailing tools — platforms that let buyers calculate payments, value their trade, and structure a deal online — got the closest to solving the conversion problem. They proved that buyers will engage with financial tools if the experience is reasonable. But most digital retailing flows still feel like tax software: five screens, personal information upfront, a linear process that doesn't adapt to what the buyer actually wants. The buyer who just wants a quick payment estimate gets the same flow as the buyer ready to sign tonight. That friction shows up in the data. Most digital retailing tools see completion rates well below 30%.

Each of these technologies solved a piece. None of them solved the whole thing. And the reason is architectural: they were all designed as point solutions. The chatbot handles Q&A. The voice tool handles calls. The DR tool handles deal structuring. Nobody built the thing that ties the entire website experience together — the way a great salesperson ties the entire showroom experience together.

The Agentic Shift: From Answering Questions to Completing Journeys

Gartner predicts that 40% of enterprise applications will include task-specific AI agents by the end of 2026, up from less than 5% in 2025. McKinsey estimates that agentic AI could unlock $300 to $400 billion in annual value across the automotive sector alone by 2030. The 2025 Cox Automotive study found that 19% of car buyers already used AI tools during their research — and those buyers reported the highest satisfaction of any segment.

The shift from chatbot to agent is not incremental. It's structural. A chatbot answers questions within a conversation window. An agent acts on behalf of the user across the entire application. In automotive retail, that means an AI that doesn't just chat about vehicles — it actually navigates the dealer's website, compares inventory, configures options, calculates real payments, and books the appointment. All inside one continuous conversation, in the buyer's browser, using the dealer's own website.

That's what we built at Swirl.

Here's what it looks like in practice. A buyer lands on a vehicle page. The AI reads their browsing behavior — what they're scrolling past, what they're lingering on, where they seem stuck. It surfaces a contextual nudge. Not a generic "Can I help?" but something specific: "Curious about real-world fuel economy in stop-and-go traffic?" or "Want to see how the cargo space compares to the Telluride?"

When the buyer engages, the AI doesn't just answer. It acts. It navigates the website on behalf of the buyer. It pulls up two vehicles and generates a side-by-side comparison with real tradeoffs explained in plain language — not bullet points from a brochure, but the kind of detail that comes from analyzing millions of real-world owner reviews, YouTube walkthroughs, Reddit threads, and automotive forum discussions. It opens the configurator. It calculates a monthly payment using current rates and incentives. And then it books the test drive at the nearest location for the time slot that works.

The buyer goes from browsing to booked in one conversation. No forms. No five-screen process. No waiting for a BDC callback the next morning. The lead that arrives in the CRM via standard ADF XML isn't a name and a phone number. It's a complete buyer profile: the vehicle they configured, the trim they selected, their budget, their trade-in, and when they're coming in.

Why External Context Is the Moat

The reason buyers trust this experience is that the AI doesn't just know what the manufacturer published. It knows what real owners say.

We built a proprietary context graph trained on over 100 million real customer signals — YouTube reviews, owner forums, Reddit threads, Amazon Q&A, TikTok, and automotive communities. When a buyer asks whether the back seat fits three car seats, or how the paint holds up in Arizona sun, or whether the hybrid battery loses range in cold weather, the AI answers from real-world experience, not a corporate FAQ. It sounds like a knowledgeable friend who happens to have read every review ever written about that vehicle.

This is a structural advantage that compounds over time. Every deployment adds to the graph. Every buyer conversation enriches it. And it creates a moat that no chatbot vendor can replicate by scraping product pages or ingesting a brochure.

The manufacturer provides the specs, the warranty, the pricing. The real world provides the trust. An agentic system that combines both is qualitatively different from anything that came before it.

The Numbers That Changed Our Company

BYD gave us the chance to prove this in production. Not a sandbox. Not a pilot dashboard. A live deployment on their consumer website across multiple vehicle models.

The results after 90 days:

Over $10 million in incremental sales in 90 days. 13% of engaged website visitors booked a test drive — the industry baseline is between 2% and 3%. That's a 5x improvement on the same traffic, same ad spend, zero changes upstream.

The engagement rate reached 27%. To contextualize: most chat widgets see engagement rates between 3% and 5%. Buyers weren't clicking and closing. They were having substantive conversations averaging several minutes per session.

Time on site increased 75% for visitors who engaged with the AI. They weren't bouncing. They were exploring, comparing, configuring, and buying.

Total incremental revenue attributed to the AI over 90 days: over $10 million. On the same traffic the website was already receiving.

We didn't change the website design. We didn't rebuild the tech stack. Leads push through standard ADF XML — the same format every CRM in the US already accepts. The deployment was live in weeks, not months.

What This Means for US Dealer Groups

I'm sharing these numbers because I believe US dealer groups deserve to see what's possible. The conversation in this market has been shaped by chatbot vendors, voice AI platforms, and digital retailing tools — and all of them have added real value. But the core website conversion problem remains largely unsolved.

Consider the economics. A dealer group with 50 stores spending $500,000 a month on marketing generates millions of website visits annually. If 97% of those visitors leave without taking any action, the cost of that leakage is staggering. Brian Pasch's research has shown that for a group like Morgan Auto Group, a single percentage point improvement in website conversion rate can translate to $50 million in additional annual revenue. We moved a live deployment from 2-3% to 13% — and generated over $10 million in incremental sales in 90 days. That's not a projection. That's an auditable number from a production deployment.

The technology to make the dealer website as productive as the best salesperson on the showroom floor exists today. Not by replacing the salesperson — that's not the point and never was — but by doing the job nobody is currently doing: selling to the buyer in the hours, days, and weeks before they ever walk in.

The Entry Point Is Simpler Than You Think

One of the biggest misconceptions about deploying AI on a dealer website is that it requires deep integration with the CRM, the DMS, and half a dozen vendor APIs. It doesn't. Not to start.

The AI sits on the website as a lightweight integration. Leads push into the CRM via ADF XML — the universal lead format that every major CRM platform in the US already supports. No gated API access. No DMS integration. No six-month IT project. The AI trains on publicly available product data, manufacturer specs, and our proprietary context graph. The only thing the dealer provides is access to their inventory feed, which they already publish to every marketplace and their own website.

Deeper integrations — personalized financing with real lender rates, trade-in valuations from book services, DMS data for returning customers — those come in Phase 2, after the dealer has seen the conversion numbers and decided to expand. The entry point is clean, fast, and low-risk.

A Challenge to Progressive Dealer Groups

We're looking for a small number of US dealer groups who want to be first.

Specifically: progressive groups with 25 or more rooftops, who already invest in their digital experience, who track their website conversion data, and who understand that the first movers in agentic retail will build a structural advantage that's very hard to replicate. Groups whose leadership doesn't just talk about digital transformation but has the digital team, the budget, and the appetite to prove it.

I'm not interested in a product demo that leads to a six-month evaluation cycle. I'm interested in a conversation between people who understand the problem and want to see what the data looks like when you actually solve it.

The buyers are already on your website. They're researching at 11 PM. They're comparing vehicles in their heads because the site doesn't help them compare any other way. They're leaving because nobody on the site walks them through the purchase the way a great salesperson would.

That's the gap. Agentic AI closes it.

Frequently Asked Questions

Sources

• NADA 2024 Annual Financial Profile of America's Franchised New-Car Dealerships — nada.org
• 2025 Cox Automotive Car Buyer Journey Study — coxautoinc.com
• Google Think Auto: Car Buying Online Research Statistics — thinkwithgoogle.com
• Shift Digital 2025 Pulse Report: Digital Automotive Shopping Trends — shiftdigital.com
• Gartner: 40% of Enterprise Applications to Include AI Agents by End of 2026
• McKinsey: Empowering Advanced Industries with Agentic AI — mckinsey.com
• Brian Pasch, Pasch Group: Automotive Website Conversion Optimization Project — paschgroup.com
• Demand Local: 39 Dealership Website Conversion Rate Benchmark Statistics — demandlocal.com

Originally published on LinkedIn.