AI Copilots Are a Dead End. Here's What Actually Works | Kavak CEO

影片總結

Kavak CEO Carlos García Ottati says AI agents now handle 90–95% of customer interactions after a year of flat growth during the transition.

Kavak founder and CEO Carlos García Ottati describes the company’s end-to-end used-car marketplace, which operates across Latin America and the Middle East, processes about 10,000 transactions a month, and employs around 3,500 people. After being defrauded while buying and selling cars in Mexico, he saw an opportunity in a market where 90% of transactions were informal and 40% involved fraud. Kavak built its own e-commerce, reconditioning, financing, and logistics infrastructure, then spent four or five years focused on Mexico City before expanding. García says the company’s early investment in connected data systems made its AI shift possible: employee copilot tools were barely adopted, so Kavak instead deployed agents to handle complex customer workflows alongside humans. The transition left the business flat in 2023, but agents now sit in the middle of 90–95% of customer interactions. He says Kavak later grew four times and became profitable, and advises founders to solve essential problems, remove user friction each week, and reassess decisions with fresh data.

章節

1. 0:00Kavak's scale: 10,000 monthly transactions, 3,500 people, and operations across Latin America and the Middle East
2. 4:58His parents stayed optimistic through every move—even in a Venezuelan village where the only car went in reverse
3. 8:08His parents’ principles through his father’s 2002 prosecution—and the relationship skills Carlos learned from moving
4. 9:50How a fraudulent car sale during his Colombia-to-Mexico move sparked Kavak’s car-pricing idea
5. 13:05Fraud, 5% used-car financing in Mexico, and Carlos’s vision of cars as a path to the middle class
6. 22:46Why Kavak spent four to five years in Mexico City before expanding, using market size and a $10,000 ticket threshold
7. 28:00Kavak’s 2017 robot-pipeline memo: building a shared ontology to handle edge cases at scale
8. 31:13Why Kavak stayed in Mexico City: 60,000 car SKUs and warranty-failure data
9. 35:31Humans with blood vs. agents with electricity: Kavak’s agent-and-human workflow
10. 40:53The year everything got worse before it got better: flat 2023 and agents handling 90–95% of interactions
11. 45:38The year everything got worse before it got better: 10,000 employees, a market collapse, and first-time buyers
12. 50:31The year everything got worse before it got better: build for real needs and remove frictions
13. 55:09The year everything got worse before it got better: 24-hour reassessment, fintech, and user friction
14. 59:38Treat problems like a protein bar: pressure is earned, and gratitude keeps it in perspective
15. 1:00:38Why Carlos fires himself every year: imagining a CEO for Kavak’s next phase
16. 1:05:52Why Carlos fires himself every year: lawn-care subscriptions, $20,000, and his first email