The problem
On Marktplaats, the largest marketplace in the Netherlands, car dealers receive large numbers of messages from potential buyers every day. Every message is a lead, but not every lead is equally serious. Sellers had to keep track of all conversations by hand: respond quickly, work out whether someone really wants to buy or is just browsing, make appointments and follow up. That took a lot of time, replies were left waiting and good leads were lost to the dealer who was just a little faster. Marktplaats wanted to raise lead quality in the automotive segment and help sellers handle every conversation in a fraction of the time.
The solution
Together with the Marktplaats product team we built a Conversational AI Engine: a standalone, modular service that recognises and handles incoming buyer conversations straight away. The engine classifies the buyer's intent (browsing, test drive, price question, trade-in), answers questions on its own where it can and makes buying intent explicit, so the seller sees immediately which leads deserve attention. Complex situations are handed over to the seller, who keeps control at a strategic level through a companion app. The engine connects to the existing integrations with dealers' lead management systems. Quality and safety are built in: structured output instead of free text, a second model that checks every answer before it reaches the buyer, and an evaluation framework that tests every change for regression. Our team was fully embedded in the Marktplaats team, with technical leadership from our partners.
The result
After a successful beta with the first dealers, the engine is now being rolled out. The target we built towards is that dealers handle 100% of their conversations in 20% of the time. Buyers get immediate, consistent answers, sellers see at a glance which leads have real buying intent, and the time freed up goes into closing the deal rather than keeping up with the inbox.
