AI Agent Automates Advertising on a Platform Without API, Quickly Tests a Hypothesis and Shuts Down the Project in Time

Maxim Todorov built an AI agent system to automate advertising on b17.ru, a specialized platform for psychologists that has no API and actively blocks bots. The system used a browser extension to extract ad data, n8n for workflow orchestration, Google Sheets for intermediate storage, an LLM for analysis and recommendations, a Telegram bot for delivering reports, and DataLens for monitoring. Before automation, a marketer spent about 11 hours per week manually copying data, matching leads to ads, calculating metrics like CTR and CPL, and preparing reports. The agent reduced this to about 1 hour per week, with the LLM interpreting prepared metrics and offering recommendations, while deterministic code handled data collection, attribution, and calculations. The system was tested in production for over a month, processing each ad run in about 124 seconds at a cost of 52 kopecks per ad. The estimated monthly savings were 47,600 rubles, plus a potential 15,000 rubles from reduced CPL, with a payback period of 3 to 6 months. However, the main outcome was that the agent quickly generated enough leads to show that the product itself was not well-adapted to the audience, leading Maxim to cancel the project. The case demonstrates that lack of API does not prevent automation, LLMs should not handle everything, and sometimes automation's purpose is to reveal that scaling is not yet viable.