How AI Agents Are Reshaping the Future of Auto Marketplaces

This webinar assembled investors, strategists and former marketplace executives to debate one of the most consequential questions facing the automotive classified sector: as AI personal assistants grow more capable of searching, comparing and negotiating on behalf of consumers, do they pose an existential threat to marketplaces — or represent their biggest opportunity? Key themes include: the timeline to mainstream AI agents; who will win the platform battle; how the buyer journey breaks into phases that AI will take over at different speeds; and the concrete strategic moves marketplace leaders should make today.

Panellists: Jonathan Turpin (Principal, AIM Group, moderator) · Martin Schwarzmann (Partner, OC&C Strategy Consultants) · Steve Greenfield (General Partner, Automotive Ventures) · Anoop Tiwari (Founder, A9i; formerly Cars.com) · Henrik Lenerius (CPO, Kaisa)

What Are AI Personal Assistants — and When Do They Arrive?

Jonathan Turpin, AIM Group

When we say AI personal assistants, we mean the next generation of chatbots focused specifically on consumers — ones that don’t just retrieve information but go out and make things happen, both on the internet and in the real world. We already see early versions, but what distinguishes these from today’s tools is the ability to plan ahead, use tools, retain context across sessions, and take action on behalf of users. The question is when they arrive for the average consumer — and whether marketplaces are ready.

Martin Schwarzmann, OC&C Strategy Consultants

These assistants are fundamentally still grounded in the same LLM technology we see today — they build on existing thinking models and will use protocols like MCP to connect to external services. Early versions exist already: the Operator-style AI browser, ChatGPT’s agentic mode, Google’s Gemini agents. But today’s best implementations cost around $200 a month, which puts them well out of reach for the average consumer. They’re useful but not yet consistently useful — people who’ve tried them can see the promise, but it’s not quite there.

In my view, we’re about two years from meaningful AI personal agents reaching the average consumer at a freemium tier — effectively free, with some usage limits. The main bottlenecks are compute efficiency, reliability, and security: these agents are currently prone to prompt injection attacks, and there are serious open questions about trust and safety when an agent is acting on your behalf across the internet.

Jonathan Turpin, AIM Group

Can websites block AI agents from accessing them — and would they even want to?

Martin Schwarzmann, OC&C Strategy Consultants

Technically, yes: services like Cloudflare can already identify certain AI agents. But the more important question is whether you’d want to. These agents represent genuine buyers with real purchasing intent. Blocking them is blocking your own customers. The business case for blocking is weak, except in very limited cases — news publishers trying to protect ad revenue, for instance. For automotive marketplaces, blocking an agent is blocking a lead.

Winner Takes Most: The Race for the Trusted AI Platform

Steve Greenfield, Automotive Ventures

My instinct is that this is a winner-takes-most market. Whichever AI engine earns your trust and becomes integrated into your daily life is the one you’ll rely on for decision support — including car buying. These agents will understand us intimately: they’ll have our calendar, our lifestyle, our purchase history, our negotiating patterns. Once you trust one of them, switching is costly. And that trust will take years to build — which means the platforms that are already embedded in daily life have a structural head start.

I think about Google paying Apple $20 billion a year to be the exclusive search engine on Safari. Those deals reflect how much control of the consumer interface is worth. We’re going to see equivalent battles for AI assistant distribution — through operating systems, through browsers, through smart devices. The big players have billions or trillions to throw at this, and I think a small number of them will end up with dominant positions.

Anoop Tiwari, A9i

I’d push back slightly. The outcome depends on which layer of the stack you’re looking at, and what the agent is actually doing. There’s a difference between an agent running at the browser level, at the device level, inside a standalone ChatGPT-style interface, or embedded within a specific website or app. And there’s a difference between an agent retrieving information and one completing a transaction. At the information retrieval layer, you might see concentration. At the transaction layer, I’d expect more fragmentation — the incentives and trust requirements are very different depending on who is deploying the agent.

Martin Schwarzmann, OC&C Strategy Consultants

The concept of an “app store for AI” — where you install a marketplace skill inside your personal assistant — was tried with ChatGPT plugins and didn’t take off. What’s replacing it is something more dynamic: context engineering. The AI assembles the right API connections on the fly, based on what the user is trying to do. This means the dominant consumer AI interface will set the ground rules — this is how much space you have for discovery, this is what you’re allowed to show — and marketplaces operate within those constraints. The incumbent marketplace doesn’t define the experience anymore. The AI platform does.

How AI Agents Will Access Automotive Data

Jonathan Turpin, AIM Group

If I search for a car in ChatGPT today, I mostly get marketplace results — because that’s the most accessible structured source. But we have multiple competing standards for how agents access data: traditional APIs, MCP, agent-to-agent commerce protocols, and more. Will agents need marketplaces to access well-curated automotive data — or will they eventually go directly to the source, bypassing intermediaries?

Anoop Tiwari, A9i

Probably all of the above, at different levels of depth. The more important observation is that the number of listings a site has isn’t what gives a marketplace its advantage — there are already half a dozen sites in any major market with more listings than the dominant marketplace. What gives marketplaces their position is how they use the data: the curation, the consistency, the consumer trust built over years. Raw listing access is not the moat. Insight, context, and trust are.

That said, I think the discovery phase of car buying is the most vulnerable to disintermediation. An agent can already browse across multiple sites and assemble a consideration set. But what it can’t easily do is complete a transaction — submit a lead that a dealer is set up to receive, navigate financing and trade-in calculations, or schedule a test drive. The infrastructure for AI-to-dealer action simply isn’t ready yet. That gives marketplaces a window.

Martin Schwarzmann, OC&C Strategy Consultants

There’s a structural opportunity here for marketplace incumbents. They have existing control interfaces — search, filtering, listing views — but these are not AI-optimised. Whatever the AI-native version of that interface turns out to be — and it will likely emerge through MCP or agent-to-agent protocols — the marketplace that defines it first will have a real advantage. The risk is building it too late, after the generalist players have already set the standard.

The Buyer Journey in Four Stages — Where AI Fits Each One

Anoop Tiwari, A9i

It helps to think about the car buying journey in four stages. First is discovery: figuring out whether to buy a car at all, what type, what brand, what your neighbours drive. Second is decision: you’ve narrowed down to a specific car, a specific dealer, you know the parameters. Third is the deal itself: the actual transaction. Fourth is durability: the ownership experience, service, relationship maintenance.

AI agents will enter each of these stages at different speeds and with different levels of disruption. Discovery is the most accessible and the most threatened — it’s where horizontal chatbots will be most effective, because it doesn’t require deep integration with dealer systems. The ownership phase is actually where I’m most excited about AI in the near term: automatically scheduling service appointments, handling recall notifications, managing insurance renewals. That’s where agents can deliver clear value and build the kind of customer intimacy that leads someone back to you when they’re ready to buy again.

Steve Greenfield, Automotive Ventures

The deal stage is where the biggest long-term disruption will come — and where AI could be most uniquely powerful. Today the car deal is the most complex consumer negotiation most people will face in their lifetime. You’re simultaneously juggling trade-in equity, down payment, monthly budget, financing rate eligibility, dealer incentives, protection product add-ons, and whether the exact car you want is even in stock. That complexity is exactly why nearly all transactions still happen at a dealership — no marketplace has managed to digitise it.

But that complexity is also exactly what AI is good at. An AI companion sitting with you during a dealership negotiation — listening, advising in real time, flagging when a deal parameter is unfavourable — would be transformative. And ultimately this will progress to fully outsourced negotiation. Dealers should understand that their professional negotiating advantage, which they’ve held for decades, is about to be equalized. Every consumer could soon have access to a professional-grade negotiating partner in their pocket.

The Discovery Risk: Why Marketplaces Can’t Cede the Top of Funnel

Jonathan Turpin, AIM Group

The risk I want to probe is sequential. If AI personal assistants win the discovery phase — people start their car search in ChatGPT rather than on a marketplace — does that mean they then lose the search phase too? Because discovery naturally flows into search. Lose the entry point, and you may lose the customer entirely.

Steve Greenfield, Automotive Ventures

That’s exactly right, and it’s the existential question. Think about what happened with physical dealership visits. Twenty years ago a consumer would visit three or four dealerships before buying. The internet and marketplaces reduced that to one or two. The same compression could now happen to online destinations: if consumers conduct their entire research journey inside an AI environment, they may arrive at a marketplace only when they need to act — and maybe not at all. That’s a direct threat to the media business model that most automotive marketplaces are built on.

Google tried to replicate the marketplace experience with Vehicle Listing Ads, and for a period it was seen as an existential threat. It turned out to be overblown — but the lesson isn’t that there’s no threat, it’s that the threat takes longer to materialise than feared. AI personal assistants are a more fundamental challenge, because they don’t just surface listings — they can do the entire discovery journey without a marketplace touchpoint.

Anoop Tiwari, A9i

Marketplaces should stop thinking of discovery as a fight they can win against horizontal chatbots and start planning for a world where consumers arrive already informed. The AI will have done the discovery. Your job shifts to being the place where the consumer validates, confirms and transacts — with the confidence that your data, your dealer network and your trust signals are better than anything the AI built its consideration set on. That’s a different value proposition, but it’s a defensible one.

AI Negotiation: The Biggest Disruption Still Coming

Steve Greenfield, Automotive Ventures

Since before the internet, the most efficient way to get the best price on a new car has been to pit dealers against each other in a reverse auction. AI agents are going to take that to its logical conclusion. They can simultaneously contact every dealer in the country with inventory matching your requirements and let market forces drive to the efficient price. For new cars — which are commodities — this could happen very quickly and dramatically compress dealer margin.

Henrik’s point about dealer-side AI agents handling buyer agent inquiries is the natural counter-move. When a buyer’s agent is pinging every dealer in the country, dealers need their own AI to screen and prioritise those inquiries — understanding which are serious buyers and which are the equivalent of noise. And eventually those two AIs will be negotiating directly with each other. The marketplace that facilitates that AI-to-AI channel, and provides the trust and verification layer that makes it credible, could capture significant value.

Martin Schwarzmann, OC&C Strategy Consultants

This is the scenario where the marketplace can make a big bet: build an agentic-first experience that covers discovery, decision and negotiation as an integrated flow. It’s expensive and uncertain — I wouldn’t claim it will succeed — but the marketplaces that try it will have a real head start over horizontal AI agents, for at least a few years. The cost of attempting it is falling rapidly as the underlying models improve.

Kaisa: Three Categories of Agent AI at the Marketplace-Dealer Intersection

Henrik Lenerius, Kaisa

At Kaisa we work with Europe’s leading marketplaces, OEMs and dealers to improve what happens between the initial lead and the eventual sale. When we look at where AI agents add value in that funnel, we see three distinct categories evolving over time.

The first category is AI-powered personalised engagement — improving conversion after the first contact has been made. A simple example: a buyer calls about an SUV and mentions in the conversation that she has two dogs who need to fit in the car. Using AI to understand that context from the call, we can automatically send her a text message on the morning of her test drive appointment reminding her to bring her licence, sharing directions, and saying “I hope to see your cute dogs in the store today.” That kind of personal touch, derived entirely from what was said during the call, drives show-up rates significantly. It costs nothing for the dealer to deliver at scale. That’s not futuristic — it’s live.

The second category is AI agents actively managing leads. No missed phone calls — an AI answers, asks qualifying questions, responds to common buyer questions and tries to book a test drive. For form leads, an AI calls back instantaneously when the form arrives, before the buyer has had time to move on to a competitor. And at the marketplace layer, an AI agent can insert itself at the start of a phone call to collect financing and trade-in information before passing the enriched lead to the dealer. Speed to lead has a very high correlation to conversion — these use cases attack that directly.

The third category is the future state: AI-to-AI. When buyer-side personal assistants become capable of contacting dealers directly, dealers will be flooded with low-intent inquiries — an AI agent might contact every dealer in a country with inventory matching the buyer’s requirements. The natural response is a dealer-side AI that screens those inquiries, answers questions about the car, assesses intent, and ultimately handles scheduling and, in time, early-stage negotiation. Kaisa is already working with marketplaces across Europe on bringing these solutions to their dealer networks. We believe this is where marketplaces have a real opportunity to capture additional value by packaging these capabilities as part of their offer.

What Marketplaces Should Do Right Now: Three Priorities

Anoop Tiwari, A9i

First: accept that the battle for dreaming and discovery is probably lost to horizontal chatbots. Consumers will increasingly do that part of the journey inside an AI environment, not on your platform. Your best response is to use those technologies yourself and ensure you’re still the destination when consumers arrive ready to act. Go deep into decision and deal. And do everything possible to maintain relationship continuity in the ownership and durability phase — that’s your best re-entry point to the next purchase cycle.

Second: protect your media business aggressively. Ad revenue depends on consumers spending time on your platform. If attention shifts to AI environments, that business is at severe risk. Think now about alternative revenue models that don’t depend on on-platform engagement.

Martin Schwarzmann, OC&C Strategy Consultants

The big bet — and I acknowledge most marketplaces aren’t ready to take it — is to build an agentic-first experience as deep and as early as possible. End-to-end, from discovery through deal. Those who try it will have an advantage of at least a few quarters, possibly years, over the generalist horizontal agents. The alternative, more incremental move is to pivot your value proposition toward the sell side: become the AI enablement platform for your dealer network, not just the listing aggregator. That’s a business model pivot, which is hard, but it’s where durable value could be built.

Steve Greenfield, Automotive Ventures

Three things. One: make this your number-one strategic priority. The share of your traffic coming from AI chatbots is small today but growing very fast. Extrapolate that line two years out. Assign your best people to it now. Two: don’t let your competitor be first to announce a strategic partnership with OpenAI or another major AI platform. That deal gives them a structural advantage in how their inventory and experience are surfaced inside the dominant AI interface. Think about co-opetition — it’s worked for large companies before. Three: data quality is your moat. Marketplaces have dominated because they had the data. The question now isn’t whether you have more listings than anyone else — it’s whether your data is so high-quality, so well-structured, and so uniquely insightful that AI agents will always need to come to you. Build for that.

Can Specialist Marketplaces Beat the Generalist Chatbots?

Jonathan Turpin, AIM Group

The question from our audience: can a specialist automotive marketplace build a specialist AI experience that is genuinely better than what a generalist chatbot offers?

Martin Schwarzmann, OC&C Strategy Consultants

Yes, in the short term, definitely. Marketplaces have a business model that allows them to invest more per interaction than a freemium AI tool — a few cents per query is within budget for a marketplace where each query could lead to a transactional event worth thousands. They have proprietary data the generalist doesn’t have. And they can get there faster, because they don’t need to solve the general-purpose AI problem — they only need to solve the automotive AI problem. The question is defensibility: how long before the horizontal agents get good enough on automotive specifics to close the gap? That window is probably two to three years and shrinking. The cost of building falls every year, which also means the cost of copying falls every year. Move now.

Steve Greenfield, Automotive Ventures

The analogy is dealer groups asking why they can’t build their own marketplace. The answer is the same: scale, cost, complexity — and comprehensiveness. Consumers and AI agents alike want to look across all inventory in one place. A dealer-built tool only shows a keyhole. The same logic makes specialist marketplaces viable against generalists: automotive search has specific nuances, specific data requirements, specific trust signals that a generalist model trained on the general web won’t replicate easily. That specialist depth is defensible — for now.

AI for Selling, Not Just Buying — The Dealer Operating System

Anoop Tiwari, A9i

Most of the conversation about AI in automotive focuses on the buyer side, but the sell side is arguably moving faster. Dealers are already deploying AI receptionists for inbound phone calls, AI-driven follow-up messaging, and AI-powered CRM analytics. Media buying is being automated. Stock management is being optimised. These aren’t futuristic experiments — they’re live and producing results. This operational AI transformation at dealerships will happen faster than the consumer behaviour shift, because the incentives are immediate and the infrastructure is already in place.

Marketplaces that want to defend their position with dealers should be thinking about this seriously. Dealers are already no longer just buying listing packages — they want tools that help them sell. Marketplaces that can become the AI enablement platform for their dealer network — packaging solutions like what Kaisa provides — will be harder to replace than marketplaces that only list inventory.

Henrik Lenerius, Kaisa

The vast majority of dealerships are not very mature on AI yet. There are some tech-forward operations — particularly in markets like Sweden — that have been very successful at using technology to reduce dependency on any single channel. But most dealers are car companies, not tech companies, and they need vendors to lead them through this transition. That’s where the opportunity lies for marketplaces and for specialists like Kaisa: acting as the trusted enabler that brings AI capability to a dealer base that can’t build it themselves.

This webinar was hosted by AIM Group and sponsored by Kaisa. Panellists represent their own views. The transcript has been edited for clarity and readability.