Half the Money Goes to LLMs: How We Automated Support and Sales Chats with AI

LeaderChat originally was an online chat with human operators. The team noticed that about 90% of customer inquiries were repetitive, and classic chatbots were ineffective as they required exact phrasing. The advent of modern LLMs solved this by understanding context and intent. The product was rebuilt in about a month without losing any clients. A separate service was developed that works with various LLMs via API, replacing the operator interface. The cost of processing a lead was reduced by about ten times compared to human operators. The company sells a result, not a platform, handling setup, integration, and AI training for clients. Currently, the service works with Telegram, VK, MAX, and a website widget, with plans to support Avito and Instagram. The team of three part-time people generates a stable profit of 100-150 thousand rubles per month. About half of the costs go to LLM usage and infrastructure. The company does not invest in large advertising, relying on referrals and partnerships, such as with CRM U-ON.Travel for travel agencies. The article concludes that the technology alone does not solve anything; it's important to find a process that already works and make it faster, cheaper, or more convenient.