Kavak Completely Rebuilds Company with AI Agents… Automating 96% of Customer Interactions and 95% of Transactions

Kavak stated that it currently invests more in tokens (AI processing cost units) than in knowledge workers, which has enabled it to create 'superhuman' agents. This means that agents outperform the best human employees the company has hired in all critical aspects. Agents are tireless, continuously learn, and demonstrate excellent performance in customer interactions.

These results suggest that Kavak's strategy of integrating AI technology not just as a tool but as the core operating system of the company has been highly effective.

Kavak opted for a unique strategy of building one agent per customer. Every day, 100,000 to 200,000 agents are instantiated with their own virtual machines for specific customers. These agents understand the customer's entire past history and provide personalized interactions, leading to increased customer satisfaction and efficient transactions.

This approach demonstrates that AI agents are designed to provide optimal service tailored to each individual customer's needs and circumstances.

Witnessing the emergence of Transformers and ChatGPT, Kavak clearly recognized that a new kind of company and a new way of building companies had become possible. These technologies marked a crucial turning point, revolutionizing existing business models and heralding the rise of companies driven by artificial intelligence.

This realization significantly influenced Kavak's bold decision to completely redesign the company around AI agents.

Kavak originally started as a used car marketplace but faced a lack of infrastructure in Latin America. This forced the company to build everything in-house: fintech for financing, logistics for vehicle purchase, repair, and sale, and vehicle history services (similar to Carfax). Because basic infrastructure did not exist, Kavak adopted a strategy of vertical integration to provide proper service to its customers.

This vertical integration became the foundation for Kavak to offer a complete customer experience in the complex used car market and to expand the scope and depth that AI agents could handle.

While many companies simply adopt AI tools into existing systems, Kavak decided to redesign the entire company around agents. This was based on the realization that merely providing ChatGPT or cloud services to teams while maintaining existing structures would not lead to true efficiency or improved customer experience. Kavak chose the bold path of actually rebuilding most of its APIs to redesign the company around agents and future capabilities.

This decision demonstrates Kavak's strong commitment to deeply integrating AI into its core business model to drive fundamental change.

The only way for agents to function properly is through training, which requires exposing them to customers, collecting data, and training them through evaluation. This feedback loop is a key element that enables continuous improvement and learning for the agents.

By enhancing agent performance based on data from actual customer interactions, this directly contributes to customer satisfaction and business results.

Kavak's AI agents handle 96% of customer interactions with minimal human intervention. 95% of all transactions are also fully processed by agents. Most of the customer journey, excluding physical car pickup or key handoff, is conducted through agents.

This high level of automation demonstrates that Kavak has successfully transformed AI agents from mere auxiliary tools into core players in business operations.

Kavak invests the same amount of time, engineering resources, tokens, and cost in building its evaluation (evals) system as it does in building agents. This is because evaluation is considered not merely a secondary task, but a core element for the continuous development of agents.

Active investment in the evaluation system is essential for accurately measuring agent performance, identifying areas for improvement, and ultimately achieving better customer experience and business results.

The agents developed by Kavak are not merely customer support or service agents; they are 'sales agents' that handle the complex process of selling vehicles. Selling a car in Latin America is a very intricate process, where buyers choose from over 20,000 SKUs, decide on loans, insurance, and warranties, and sell and trade in their own cars. From 2020 to 2021, Kavak required 15 different highly skilled experts across 15 fields to handle this process. Now, AI agents manage all of it.

These sales agents understand complex customer requirements, provide customized solutions, and streamline all stages of transactions, maximizing sales efficiency.

There are several reasons why Kavak's AI agents are better salespeople than humans. AI agents are specialists in specific areas, have infinite patience, and understand a customer's entire history. They can also formulate long-term plans and never get tired. Importantly, even if an agent makes a mistake, it learns immediately the next day, and this learning is shared not only with that agent but also with the other 200,000 agents, leading to immediate improvement. This is the feedback loop Kavak utilizes, and it is resulting in increased customer satisfaction and performance.

These strengths of AI agents play a decisive role in optimizing customer interactions and transforming the sales process into a more efficient and personalized experience.

By deeply understanding customers and streamlining financial product processes, Kavak has significantly improved conversion and retention metrics. Beyond simple transactions, it understands each customer's personal situation and, when ready, presents deeply personalized interest rates, risk levels, and maximum loan amounts to help close deals.

This deep personalization of financial products enhances customer satisfaction and contributes to building long-term customer relationships.

Kavak conducted an experiment by deploying an 'AI CEO' in a Mexican city called Cuernavaca. This AI agent was designed to learn, make decisions, and evaluate its own decisions as a CEO. In this 6-week experiment, the AI CEO did not meet its first-month goal of doubling revenue but achieved a 1.5x, or 50%, increase in revenue.

This experiment proved that AI can achieve significant results not only in improving operational efficiency but also in strategic decision-making. This suggests the possibility of an expanded role for AI in future corporate management.

Kavak developed 'El Mike,' an auxiliary AI to help mechanics. Like the collaboration between the human chef and the rat in the movie 'Ratatouille,' El Mike guides mechanics on how to inspect cars, provides useful tips, and outlines procedures. With the help of this AI, Kavak's vehicle inspection quality once again reached its highest level.

This case demonstrates that AI can be utilized not to replace human jobs, but to enhance human capabilities and improve work quality.

To transition into an AI-driven company, Kavak launched an in-house program called 'Jedi Academy.' This academy includes all Kavak employees, from the CEO to AI engineers and even mechanics. The goal is for all employees to understand AI agents and be capable of developing and deploying them directly.

Through 6 weeks of intensive training, employees acquire the skills to deploy cutting-edge AI agents into production. This demonstrates Kavak's strategy of internalizing company-wide AI capabilities rather than leaving AI to just a few specialists.

This company-wide AI training supports employees in adapting to the new technological paradigm and leveraging AI in their respective fields to drive innovation. As a result, Kavak is building an AI-driven innovation culture across the entire organization.

Kavak emphasizes that human-AI collaboration is crucial for the repetitive learning of agents when implementing AI. Simply having the AI move on to the next step when it hits a wall hinders its growth because data is not accumulated. Kavak designed its system to assign a dedicated agent to each customer, and when an agent encounters a specific problem, it calls an API to 'ask for help' from a human expert.

This approach contributes to accumulating data on complex situations that agents cannot resolve without human intervention, thereby completing the agent's learning loop and continuously improving performance.

Kavak argues that AI implementation must be top-down. A bottom-up approach makes it difficult to drive company-wide change because organizations build their own ideas without a clear strategy. Therefore, leaders must have a clear plan for AI implementation and issue directives across the company. They emphasize that hackathons or free idea generation alone are insufficient to achieve the benefits of AI transformation.

This top-down approach helps the organization build AI in a unified direction, which is essential for driving systemic and cultural change across the company.

Kavak decided to build one agent per customer, and in doing so, discovered that each agent required its own micro-virtual machine. This was a technically risky and challenging decision but proved highly effective in providing highly customized services to customers. This approach allows agents to deeply penetrate individual customer requirements and contexts to provide optimal interactions.

This strategy enables AI agents to function not just as simple automation tools, but as key drivers of customer experience innovation.

The new AI system built by Kavak is implemented in the form of virtual machines equipped with agents that can access memory, evaluation (evals) functions, and a command-line interface (CLI). This virtual machine provides an independent execution environment for each customer-specific agent, enabling highly personalized and stable service delivery.

This architecture supports agents in processing vast amounts of data, making complex decisions, and interacting efficiently with the system.

Kavak explains that truly leveraging new technology requires destroying and redesigning existing systems, drawing an analogy to the introduction of electricity. After Edison commercialized electricity, many factories merely replaced coal engines with electric ones, gaining only about 6% efficiency. However, when entire factories, like Ford's production line, were redesigned around electricity, they achieved a threefold increase in productivity, leading the economic development of 20th-century America.

Kavak states that this analogy applies equally to the computer age and today's AI era. If companies only superficially adopt AI and are unwilling to redesign the entire company, they will achieve only 6-10% improvement, and true tenfold innovation will be impossible. This, they point out, is akin to the innovator's dilemma on an industrial scale.

Kavak emphasizes that this is the most exciting time in human history and the best time for founders. It is the first time anyone can access the world's most powerful tools and intelligence almost for free, or for as little as $20 a month. This democratization of tools is unprecedented in human history, and there are many problems to solve and new realities to build around AI.

Kavak advises new founders to go beyond simply trying AI; to deeply imagine what an AI-centric future would look like and build for it. If they assume AI will continue to develop linearly and build accordingly, they will discover wonderful ideas that bring much value to the world.

This advice conveys the message that in a new technological paradigm, entrepreneurs must have a bold vision and leverage AI as a core driving force for innovation.