Can agentic AI take the pain out of buying a car?

This webinar brought together analysts, investors and practitioners to examine how AI agents are reshaping the car buying journey — for consumers, marketplaces and dealerships. Key themes include: the persistent pain points buyers face, where agentic AI delivers measurable value today, how marketplaces like AutoTrader are positioning for an agent-first world, and what the next 18 months realistically look like for the industry.

Panellists: Kate Cavanaugh (CEO, AIM Group, host) · Jonathan Turpin (Principal, AIM Group, moderator) · Dr Anna-Sophie Smith (McKinsey & Company) · Steve Greenfield (Automotive Ventures) · David Sykes (AutoTrader UK) · Thomas Montagnier (Kaisa)

The Pain Points Buyers Still Face

Dr Anna-Sophie Smith, McKinsey & Company

Starting with the car buying process overall, people still have quite a few pain points. Some are very practical — long delivery wait times top the list, and this varies quite a bit across European markets. But many of the pain points people mention are linked to what I’d call information effects: things in the car buying journey that are too complex or difficult to understand. These are exactly the kinds of things where AI assistance can help — whether that’s providing a transparent benchmark across different cars, or explaining complicated contract terms to a customer. So across the board there are quite a few things that AI tools can address.

How Buyers Are Using AI Today

Dr Anna-Sophie Smith, McKinsey & Company

How do people actually look for cars today? Looking at our UK car buyer survey, we see two things clearly. First, there is no single source where people gather information. Second, there is a very substantial generational divide. Gen X and Boomer buyers focus heavily on traditional sources — comparison sites, dealer websites, in-person visits. Younger generations, particularly millennials, are much more multi-channel and gather information from many different places.

And now there is an emerging share of buyers who have integrated AI tools into their car buying process — and this is not only related to age. When we look at different buyer groups, we see the share of people who used an AI tool in their last purchase varies dramatically. For new cars, around 15% used AI, versus 5% for used. There’s an even larger spread between premium and volume brand buyers. The more complex, high-value, or unfamiliar a car buying decision is, the more likely buyers are to bring AI into their search process.

Three things to take away: customers still have pain points; the journey is fragmented across many sources; and AI has already landed in car purchases. It’s not about if, but how we make the best of it.

Where Pain Is High Enough for Buyers to Delegate to an Agent

Steve Greenfield, Automotive Ventures

The maximum point of pain isn’t so much in the online shopping process — it’s in the face-to-face interaction with other humans. The average consumer buys a car every four or five years, whereas the dealer is effectively a professional negotiator. The information asymmetry is significant: the consumer often doesn’t know what their trade-in is worth, whether they qualify for the finance rate they saw advertised, or whether the protection products they’re being pushed are good value. They often leave the dealership with a dramatically higher monthly payment than they expected, and without knowing if they’ve been taken advantage of.

With a professional, omniscient negotiator that consumers trust — ChatGPT or similar — in their pocket, more and more of the shopping and negotiating process will be outsourced to AI. And increasingly, dealers will be negotiating not with humans, but with AI agents.

We’re already hearing anecdotally that consumers are using AI in real time in the showroom. In some edge cases, buyers are photographing their paperwork just before signing and uploading it to ChatGPT to get a second opinion on whether the deal is fair. Dealers should prepare for this. It’s coming quickly.

Steve Greenfield, Automotive Ventures

What won’t consumers hand to an agent? Wet signatures on deal documents still persist, at least in the US — the virtualisation of the car deal hasn’t fully arrived yet. And the test drive. Despite augmented and virtual reality solutions over the years, nobody is virtualising the experience of physically driving a car.

How AutoTrader Sees AI Adding Value

David Sykes, AutoTrader UK

AI is fundamentally about efficiency — getting buyers to the answer they’re looking for faster, and building confidence along the way. There are a few different stages to consider.

Right at the start of the journey, in the research phase, an agent experience can really support people in understanding what might be most suitable for them. We’ve seen an explosion of new car brands in the UK — the number has grown from around 45 to around 75 — and with that comes an explosion of models. Buyers’ familiarity with the market isn’t where it used to be. Something we’ve explored is AI-generated search categories that allow people to search by needs — “family car”, for example — and have vehicles matched to their requirements.

Moving into the active search phase, an agent layer can help people navigate filters, which can be intimidating. We have an AI-powered free-text search currently in test — you can type in your requirements however you express them, and it matches to the relevant search fields and returns the most relevant results. We see people choosing to use it over traditional filters, so there’s clearly a preference for some buyers.

Another route is via LLMs directly. Using the Model Context Protocol, we’ve integrated with ChatGPT so that buyers can express what they want in a conversational way, and the agent calls AutoTrader directly for real-time availability and pricing, returning results in the native experience.

Jonathan Turpin, AIM Group

How will you treat traffic from bots — block it or welcome it?

David Sykes, AutoTrader UK

We would absolutely welcome it. A buyer has tasked this agent to conduct research on their behalf — we treat them as a buyer whether they’re human or an agent acting for a human. Our recent investment in MCP is an example of that. It’s essentially a search API designed specifically for this, and our first integration has been with ChatGPT, with others to follow.

On the question of agents taking contact details and going direct to the dealer, bypassing AutoTrader: the goal of the buying journey isn’t just going from not having a car to buying a car. The goal is having confidence you’re buying the right car, at the right price, from the right retailer. Building that confidence requires context and specialist knowledge — which is where specialists like AutoTrader come in. We see the natural journey starting in a broader discovery experience and transitioning to a specialist layer where confidence is built before the buyer contacts the retailer.

Who Will Own the Car Buyer Relationship?

Dr Anna-Sophie Smith, McKinsey & Company

There won’t be a single party owning the entire process. Different players have different value at different points in the journey. At the very top of the funnel, there’s going to be the most competition between marketplaces and AI assistants. The personal AI assistant has an advantage there — it already knows the consumer, has their history, understands their needs. That’s something even a platform-integrated AI may not have.

As you move toward the transaction phase, OEMs are getting stronger through agency models and more direct-to-consumer approaches. But I think the right framing is specialisation at different points in the journey, with AI acting as a supportive layer across all players.

Dr Anna-Sophie Smith, McKinsey & Company

On new versus used complexity: I’d push back on the idea that used is more complex. When you have a used car, you know exactly what it is — you don’t need to configure it, you don’t need to take further detailed decisions. It’s almost easier to tell your AI assistant: here are five cars within my budget, help me decide. Whereas the option space in new cars is getting more and more complex — a proliferating number of brands, endless configurations.

David Sykes, AutoTrader UK

Both present different challenges. Used: every item of stock is unique. New: comparing across manufacturers is difficult because BMW and Audi describe options and packs completely differently. A single comparison point that presents both in a consistent way would make new easier. This is where a genuinely agentic experience could enable mass personalisation — comparing on whatever factors matter to the individual buyer, new or used, in one consistent experience.

The Data Advantage for Specialists

David Sykes, AutoTrader UK

Proprietary data is where a specialist like AutoTrader differentiates from a generalist LLM. Starting with the depth and consistency of vehicle taxonomy data — the ability to compare one car meaningfully against another. Then building on that: mileage indicators that compare against the cohort, price positioning against comparable vehicles in the market today, rare spec or features highlighted, boot size versus segment competitors. Every one of those nuggets gets the buyer one step closer to a confident decision.

I think everyone benefits from that level of insight, even buyers who aren’t emotionally invested in cars. They need their car to serve a purpose, and they want confidence it will do so effectively and without problems.

Who Do Buyers Trust?

Dr Anna-Sophie Smith, McKinsey & Company

When we asked buyers this directly in our latest survey, they clearly stated they trust their generalist personal assistants — GPT or similar — more than AI integrated into a marketplace or OEM website. And I think that’s linked to something more fundamental: younger buyers, particularly millennials and Gen Z, have their personal AI very integrated into their lives. They ask it about everything. It has a lot of background on them. The trust is already there.

For older generations who haven’t yet built that relationship with a personal AI, the proprietary depth of data on a platform like AutoTrader may instil more trust. It really depends on the consumer archetype.

David Sykes, AutoTrader UK

There’s also a marketing job for marketplaces to do — communicating why their AI is richer and more specialist than a generalist tool. Buyers may assume it’s just a generic chatbot. Making the point that it’s a user-friendly interface to access deep, specialist data is an easier frame to land.

What AI Voice Agents Are Delivering Today — Kaisa

Thomas Montagnier, Kaisa

Kaisa is a trusted partner to some of the most influential marketplaces and OEMs in the world. More than 80% of Europe’s car dealerships use our solutions daily, and we power more than five million conversations per month across 52 countries.

One of the areas where we specialise is tracking and improving phone conversations. After analysing over a million phone calls with our AI call quality solution, we found that almost 42% of phone leads received no response at all. Earlier analysis only counted obviously missed calls — but with most dealers now using IVR systems, a large share of buyers who call never actually reach a dealer. That’s a major pain point, and marketplaces can tackle it today with AI. Not tomorrow — today.

We’ve gone live with Mobileye in Germany and other partners in recent months, and we’re already seeing impressive results. On paper, the flow seems simple: a call is about to be missed, an AI agent jumps in. But what we found working with partners is that the real game-changer is having an agent that is context-aware and brings genuine value to the conversation. That’s directly tied to the quality of data you connect the agent to.

When you feed the agent listing information, it can answer questions about the specific vehicle before qualifying the lead. It already knows the car in question and all related information — so it’s not just asking “how can I help?” It elevates the conversation and drives conversion.

From our first market tests in Spain and Germany, up to 64% of phone leads answered by the AI agent were fully qualified — meaning the buyer answered every qualification question around timeline, financing and trade-in, and chose to share their contact details. What would have been a missed lead becomes a valuable, qualified lead for the dealer.

From the buyer’s perspective, the experience is smoother than an IVR loop or no response at all. They can have a natural conversation, their interest in the car is acknowledged, their questions are answered, and all information they share is immediately sent to the dealer.

When a lead is qualified, the marketplace can send the dealer a notification with a summary of the call and clear qualification tags. Dealers tend to respond much faster because they have full context. Over 80% of those notifications are being opened by dealers — which shows the real appeal of enriched lead notifications.

Bot vs Bot: The Negotiation Question

Steve Greenfield, Automotive Ventures

Since before the internet, the most efficient way to run a new car purchase has been to pit dealers against each other in a reverse auction. AI agents are only going to amplify that. For new cars, once you’ve found the right car and can treat it as a commodity, AI will drive out dealer margin and create dramatic price efficiency — and that could come quickly. Dealers should be aware.

There’s a concept from business school called the zone of potential agreement. On any given day in a negotiation, there’s an efficient frontier of deals that could work for both buyer and dealer. A bot-versus-bot negotiation could very efficiently and quickly explore all those deals within that zone — and may dramatically shrink the time it takes to reach an agreement, while still leaving both sides with a reasonable outcome.

David Sykes, AutoTrader UK

In that world, it almost becomes a zone of inevitable agreement — bots will reach an inevitable price, which is effectively price transparency. And I think that would be good for everyone. The pain point is the negotiation itself. If you can have confidence in the price from the start, that removes the pain entirely. Fixed-price selling already exists in the UK market, and we actually lost a car recently because I kept trying to negotiate on a fixed-price vehicle. So the trend is real.

What’s Deployable in 18 Months vs the Five-Year Vision

Steve Greenfield, Automotive Ventures

For consumers: we’re in the very early innings. ChatGPT only launched three and a half years ago. These models will get exponentially better, and consumers will lean on them for more of their daily lives — not just buying things. For automotive, which is probably the most painful negotiation most people will face outside of real estate, consumer adoption will increase as the tools prove themselves.

For dealers: they’ll be inundated with vendors. As more traffic comes via AI agents, dealers will need to figure out how to present their data in a digestible way — whether via MCP or whatever succeeds it — to ensure they get their fair share of agent-referred traffic. And they’ll focus on efficiency: driving more revenue per employee using AI-enabled tools.

The models are improving faster than almost anyone predicted. I use Claude daily and I’m continuously impressed by how much better it’s got in the last three to six months. I’m very bullish on where capability goes.

Structured Workflows vs Conversational AI

David Sykes, AutoTrader UK

The real value of agentic experiences is that they are flexible and personalised. Structured flows miss that — we could have built those for years. What AI uniquely offers is the ability for the consumer to take the journey in the direction they choose to, when they choose to. It’s about optionality and complementary experiences.

Consumers don’t want AI for its own sake — they want solutions to problems. Part of the current hyperbole leads people to build things because they can, not because they should. AI should be complementary. Some people will dip in and out of traditional journeys, using AI to support specific moments. Others will live within an AI-led experience end to end. Giving people the choice to use it where it adds the most value to them is the goal.

What Agents Still Get Wrong — and What’s Further Along Than People Realise

Thomas Montagnier, Kaisa

The most obvious current limitation is taking action on behalf of the dealer — booking a test drive directly into a dealer’s calendar, for example. Not all dealers use synchronised digital calendars, so the agent can’t always complete that action. The challenge is less about AI capability and more about integration with dealer infrastructure.

What’s further along than most people realise is the ability for AI agents to apply a standardised sales playbook consistently. Not all human sales staff perform at the same level — some forget to ask about financing, trade-ins, or purchase timelines. An AI agent asks every qualification question, every time, and leaves no opportunities on the table. That’s something the industry is systematically underestimating.

Will AI Finally Break the 10-Cars-Per-Month-Per-Salesperson Ceiling?

Steve Greenfield, Automotive Ventures

In the US, dealership sales productivity has been stuck at around ten sales per salesperson per month for roughly 75 years. I’m hopeful that AI changes this — not by making each human push more units through, but because an increasing percentage of sales over time will be conducted without a human at all. That ratio naturally drifts upward. It might be the most objective KPI we have for whether AI genuinely improves automotive productivity.

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.