First, a quick primer on how cars actually get sold in Korea
In the U.S., you've got Manheim and ADESA running massive dealer-only auctions, the kind of infrastructure where you can't even walk in without a dealer license. Korea has a similar backbone, but there's a startup called HeyDealer that's spent a decade doing something different: connecting private sellers directly to dealers through its own real-time online auction. Think ACV Auctions, but for Korea, and with a decade of proprietary data. Dealers place bids. Highest bid wins. Simple in concept. Brutal in execution.
The client and the problem hiding in plain sight
Our client is one of the largest dealerships in Korea. Smart team. Clear-eyed about what was slowing them down.
Here's the situation: dealers are not sitting at a desk all day refreshing auction pages. They're on the floor selling cars. And yet, winning at auction requires real research. For any given vehicle, a dealer needs to know:
- What is it selling for at retail, right now?
- If it doesn't sell retail, can it be exported? For how much?
- What's the refurbishment cost going to look like, given this car's accident history, options, tire wear?
- After all of that, what's the maximum I can bid and still make money?
Before our engagement, answering those questions meant hopping between multiple pricing sites, doing mental math, and translating a senior dealer's 15 years of intuition into a number, in under 20 minutes per vehicle.
That's not a knowledge problem. That's a throughput problem.
The baseline numbers (not pretty)
A dedicated dealer working a full day on HeyDealer auctions could participate in 50 bids/day. Of those, they'd win about 4%, roughly 2 cars.
Two cars a day. After a full day of work.
There are exactly two levers here: bid on more cars, or win more of the ones you bid on. We pulled both.
Lever 1: Kill the tab-switching
The biggest bottleneck, when we asked dealers directly, was unambiguous: looking up retail comps.
Finding a comparable vehicle across multiple retail platforms and assembling a price range took 15 to 20 minutes per vehicle. For difficult units, longer.
So we built an agent that does it for them.
The moment a dealer clicks on a vehicle in HeyDealer, the AI agent goes to work: it identifies comparable listings across retail platforms, pulls current pricing data, and surfaces a clean summary before the dealer has even finished reading the vehicle description. No new tabs. No copy-pasting. Just the data, already there.
Twenty minutes becomes thirty seconds.
Lever 2: Turn intuition into a system (without killing the intuition)
Knowing retail comps is only half the equation. The other half is knowing what you will spend to make that car sellable again.
We spent real time with the team, interviewing both senior dealers with years of pricing instincts and newer staff who were still building theirs. We mapped out how they think about accident location and severity, options, tire condition, and the company's internal margin requirements.
Then we codified it. Not into a rigid formula, but into a tunable model that each dealer can adjust to match their own judgment.
The result: click a car, and the agent returns a recommended bid price. It has already factored in retail comps, estimated refurbishment costs, historical auction prices for similar units, and the dealer's required margin. The dealer can accept it, adjust it, or override it entirely. The system learns what matters to each user.
Bid smart, not just fast.
What happened
Bids per day: 50 → 200
Win rate: 4.0% → 15.8%
That's not just efficiency, that's compounding. More bids × higher win rate = a fundamentally different business outcome. The team also noted meaningful savings in the administrative overhead of tracking and managing auction records, which had previously been a manual, error-prone process.
What this is really about
A lot of AI work in enterprise gets positioned as cost reduction. Fire people, save money, repeat.
This is the opposite story. The dealers are still there. Their expertise is still essential. What changed is that the routine information work, the tab-switching, the comp-hunting, the mental arithmetic, is now handled before they even have to think about it.
That's what a well-deployed AI agent actually does: it doesn't replace judgment. It clears the runway so judgment can land more often.
If you're sitting on a workflow that should be faster and you're not sure where to start, we'd be glad to take a look.
partnership@nurie.ai | nurie.ai