Fifty thousand calls in, I keep coming back to the same question: what have we actually learned?
It would be easy to answer that with a feature list. I don't think that's the honest answer, so here's the one I actually believe: AI doesn't change the game for AutoRetail's aftersales business. Efficiency, retention, and margin do - the same fight aftersales leaders have always been in, on a P&L that runs at 1–2%. What's changed is how much of that fight can now be fought proactively instead of reactively, and that's the only reason I think AI belongs in this conversation at all.
Our own research puts real numbers behind that, not just conviction: a 10% improvement in retention alone can drive roughly 30% revenue growth, and the vehicles AutoRetail businesses lose fastest - the ones past warranty - are worth up to three times as much per year as new ones. That's the business case for taking retention as seriously as new-car sales, before AI ever enters the conversation.
So I don't start with the technology, and I don't let my team start there either. We start with strategy - where an AutoRetail business is actually losing money or customers - and only then ask what AI can accelerate. That's a discipline, not a slogan, and it's the one thing I check on every rollout we run.
To stress-test whether we're actually holding to it, I sat down with Marc and Maz, who've spent the last year and a half building and running this across AutoRetail, and asked them the questions I'd want answered if I were the one being sold to: real calls, messy systems, and customers who don't care how clever the tech is, only whether it actually worked.
Nicholas: Let's start here. What actually makes our approach different from everyone else selling AI to dealerships?
Marc: We learned this the hard way. We assumed that if a dealership had all its data, setup would be plug and play. It isn't. Most of how a dealership actually runs isn't written down anywhere - it's passed on by word of mouth. One customer gave us full documentation on how they handle bookings and pricing, and it was still only half the picture. Once we started testing, we kept hearing “actually, we do it differently” or “there's a caveat nobody wrote down.”
Every dealership runs a bit differently, even on the same system. So instead of building per dealership, we build per pattern: this is how Toyota dealers work, this is how BMW dealers work, this is how Mercedes dealers work. That's repeatable. The voice part, how it sounds, is the easy part. The hard part is the data: making sure it's accurate, flows through correctly, and a booking ends up looking exactly like it would if the customer care team had done it themselves.
Maz: And dealerships need to treat this like hiring someone, not installing software. It needs onboarding and training. The ones who get real results are the ones who stay involved in setting it up, not the ones who buy the product and expect it to run itself.
Nicholas: That's what every dealership is really asking, even when they phrase it as “does it sound good?” Does it actually work?
Maz: It can, but not because it sounds convincing. Latency has dropped by hundreds of percent industry-wide over the past year, so voice AI already feels close to a real conversation. That's not the differentiator anymore. What matters now is how well it actually reflects the business and connects to it.
Marc: A lot of early competitors got this backwards. They focused on making it sound good - and it sounded great, but it couldn't do anything: no access to the right data, no way to interpret it. It became a front door that annoyed customers and burned out the dealerships that bought it. Anyone can build a chatbot that talks. The hard part is making something that works across thousands of calls, stays predictable, and is easy to debug when something breaks.
Picture a dealership group with 35 locations taking around 200 calls an hour, all day, every day, all year. Once you can handle that reliably, you've built something that creates real value for the customer, not just something that saves the dealership money.
Nicholas: That's what actually matters to me. Not whether it feels human, but whether it's consistently moving the numbers that matter - retention, bookings, cost per call. What does that look like day to day?
Marc: Transfer calls are a good example. When the AI hands a customer to the workshop, it talks to the service advisor first and passes along the context before the customer even gets on the line. That matters as much as the call with the customer, because the more the advisor already knows, the better they can handle it. It works alongside your team, not instead of them - which means your frontline staff need to be part of giving feedback on how it's actually performing.
Nicholas: Where do you actually see this succeed or fail?
Maz: The moment an AI system isn't properly connected - when it doesn't tie together booking, product data, and service history - it just becomes a new source of leads. That means more work for the service team, who now have to call people back, because a customer who spoke to someone, even an AI, expects a call back. You've added work instead of removing it, and the booking still isn't done.
Marc: Every extra step between the customer and a finished booking is a risk. If someone has to read out a license plate and the AI gets a letter wrong, that's a bad moment. The more context the AI already has, the fewer questions it needs to ask - it's confirming details instead of collecting them from scratch. That's also what makes a customer feel recognized instead of thinking “great, another chatbot.”
Maz: Most of the time, the moment that matters is the simplest one. Most people calling in don't want a conversation. They want: “I need a service.” “Is this about your Volvo? You're booked.” Done. Nobody wants a fifteen-minute chat - they want it handled.
Nicholas: This is the one I care about most. “Everyone else is doing it” isn't a real reason to spend money on anything.
Marc: A dealership told me recently they miss about half their calls. The obvious fix is to hire more people, but if your margin is one or two percent, that's not realistic. That's the real case for AI: not that it's exciting, but that the alternative doesn't work.
One of our clearest results was a recall campaign where every affected customer was contacted and booked in automatically. We got a 40% conversion rate. Recall calls are normally a headache for everyone involved, but when the customer already knows what it's about and can just book instead of being pushed to a form, it works better for both sides.
Maz: But it only pays off if the reason behind it is right. Is the dealership doing this to actually improve things for the customer, or just to cut headcount? Look at Klarna - they replaced customer support with AI, satisfaction dropped, and they went back to a model where a human steps in the moment something's off. Satisfaction recovered. If AI removes a real pain point, customers don't mind that it's automated. If it doesn't work and they have to repeat themselves five times, that's just a worse version of the same problem.
Dealerships that only see this as a way to cut costs won't get much out of it. And outbound calling - reaching out before the customer has to call in - is actually one of the better ways to cut down inbound volume in the first place. That's where a lot of the real payoff shows up.
A note from Nicholas: this is exactly why we don't start with “how do we use AI.” We start with “where are AutoRetail businesses losing customers today”: missed calls, forgotten follow-ups, appointments that never get booked, customers who quietly go elsewhere. The recall example above is a good case. Nobody set out to “use AI.” They set out to solve one specific, expensive problem.
Nicholas: We bring this up constantly internally. Why is it so central to whether any of this actually works?
Maz: For most dealerships, the DMS holds most of the data you need to give a customer real context: bookings, service history, appointments. But it's not enough to just connect to it. You have to organize that data so it's actually useful in the moment you're talking to someone. There's also data that lives outside the DMS entirely - vehicle registries, deferred work, health checks - so you're really building a full picture across every system, not plugging into one database.
Marc: Even dealerships sitting on all this data often can't answer something as basic as their own customer retention rate. They have the data, not the tools to read it. Without cleaning it up first, whatever AI you add just gets lost in the noise. Some dealerships still book appointments on paper. So part of our job is telling dealerships what they need to fix before AI can even help. On the outbound side specifically, when something fails, it's almost never the AI or the voice - it's bad data. That's why we clean the service data first and use it to figure out the right time to call someone, based on mileage or service history, before the AI ever picks up the phone.
Nicholas: How do we decide something's a success before we call it done?
Maz: Start with visibility: an easy way to see every AI interaction, call transcripts, how customers actually respond. That's what lets us improve things and confirm it's actually performing, because we're judged on results. AI that doesn't complete bookings or convert calls is just added cost, and someone still has to finish the job by hand.
Marc: In practice, it means everyone involved has signed off - sales, customer support, dealership management, and the staff who'll actually work with it day to day. You roll it out carefully and make sure your team has context before a transferred call lands on their desk, so it doesn't feel like extra work dumped on them out of nowhere.
And it looks different for every dealership. Some need more bookings, so a light setup connected to the service department is enough. Others are already busy and need something that can finish a booking start to finish. There's no single blueprint - it depends on what the dealership actually needs.
Nicholas: Last one. Say a friend is an after-sales director trying to convince his CEO to invest in this. How does he make the case in a minute?
Marc: Start with one specific problem - something that already eats up a lot of manual work. That's where AI actually helps. If you just say “I want AI in my dealership,” where does that even go - sales, parts, service? In after-sales it's simple: “I want more bookings with less manual work, and I want to invest in AI to get there.” That's something a CEO can actually judge.
Nicholas: And the other way around: if you're the dealership evaluating a vendor, what would you watch for?
Maz: First, how it connects to our systems, and whether it adds noise or actually fits our data. Second, how they plan to bring our employees into the rollout, with everyone informed. Third, and only third, how the voice sounds and whether it's compliant. In that order.
Marc: I'd put it a different way: the worst thing you can do is bolt on a system that sits outside everything else you run. That fails on day one no matter how good it sounds. A good conversation builds trust, and you need that too - but without the integration, there's nothing underneath it.
Here's what I take from this, and what I expect us to be held to. The technology itself - the voice, the latency, the model - was never the point, and I don't want it to become one internally either. What matters is whether a customer's call gets answered, their appointment gets booked, and their follow-up actually happens, every time - and whether all of that shows up in retention and the bottom line, not just in a demo.
Fifty thousand calls in, that's the standard I'm holding this company to going forward. Lead with strategy. Accelerate with AI. Customer first, hype last.