Background
Kavak is the largest used car marketplace in Latin America. The company buys, reconditions, sells, and finances cars across Mexico, Brazil, Argentina, Chile, and the UAE. It sells 10,000+ cars a month, runs 100k+ transactions a year, and interacts with close to 100,000 customers every day. Its financing arm covers roughly 60% of the cars it sells in Mexico.
The company was founded to fix a specific problem: buying or selling a used car in Mexico's informal market is risky. More than half of Kavak's customers are buying a car for the first time. For most of them, it's the largest financial decision of their lives.
The Challenge
Kavak strives to treat each customer as if it were their own mom. But between 2020 and 2022, Kavak grew roughly 100x in 25 months. This concierge-level service model took a hit during this time period.
It wasn't humanly possible to give this level of concierge experience to hundreds of thousands of potential customers each week.
The commercial team started rationing attention. Customers were segmented by operational capacity and conversion probability, and some leads were simply let go.
The customer experience team had a visibility problem on top of the capacity problem. With thousands of daily interactions, they could manually review only a handful of those conversations.
Head of Customer Experience Ana Martinez describes those years plainly: "We were basically working blind."
One more constraint made this harder than a typical support scaling problem: the channel had to be voice. In Mexico and across LATAM, customers making a major purchase just want to be heard. Chat alone doesn't build the trust the transaction requires.
Why Vapi
Kavak had already built text agents for chat channels. Voice was the missing piece. Maza found it himself, on a weekend, without a sales process.
I found the Vapi API and I started hacking around. A couple of hours later, I had something that really blew my mind. I sent the demo to my team and my idea back then was: 'Guys, I think this is it. I think we found it.'
What made the platform stick was how it fit the way AI products actually get built. You don't get an agent right the first time.
You need to work around personality, assertiveness, friendliness, and you need lots of trial and error, says Maza. In order to build products by trial and error, you need lots of control. You need to be tweaking the voice, the latency, the performance, the personality, the tools.
Full control to configure in a fine-grained way, plus the ability to implement these changes quickly were key to Kavak's success. Vapi exposed every parameter that mattered, so each iteration shipped in minutes instead of sprints, and production agents were constantly being optimized.
The full rollout placed agents across the full journey: first contact and sales calls, inspection scheduling, financing and underwriting reference calls, proactive daily after-sales updates, and feedback loops from the reconditioning centers.
But Vapi agents don't just handle requests at scale. The real benefit is that they improve the overall customer experience by providing patience, care, and attention at a previously impossible scale.
We started seeing people connect with the agent and speak for 10, 20, 30 minutes and just say ‘Thank you’, this is the better service and experience I’ve ever had
Outcomes
The results showed up in the metrics Kavak cares about most:
- +200% revenue growth
- NPS up ~20 percentage points
- Conversion rates up 30%
- Profitability reached in Mexico, while serving 2x the customers
- 100% of conversations analyzed daily, up from 10 to 30 manual reviews
- 24/7 availability, eliminating the eight-hour service window
Customer are happier, they are converting more, our NPS and customer satisfaction scores just skyrocketed
The numbers come from a continuous loop: business teams build new evaluations in Vapi in minutes, tune voice, tone, emotion, and speed, measure against customer sentiment from live calls, and push what works into production. That loop lets Kavak build agents that connect with its specific customers, first-time buyers who want to hear a trustworthy voice before the largest purchase of their lives.
You don't have 100 transactions. You have 100 people talking to you and needing their cars in daily life. This technology allows us to give differentiated treatment to those 100 people, something we couldn't achieve even with a team of 1,000. - Ana Martinez, Head of Customer Experience
The result is a more personalized, higher-conversion buying journey with upsell paths that didn't exist before - for example, an agent who recognized a customer asking to sell her car actually needed liquidity and shifted the call to a credit proposal.
The question is no longer 'how can we deliver good service?' It's 'what else can we build?' Everyone at Kavak became a builder, and they're building on Vapi.
Two years in, Vapi is Kavak's default platform for building and deploying voice agents.
