GIGASCHOOL: How an AI agent saved 47,600 rubles per month and identified a product problem

Manually collecting statistics from advertising platforms without an API is hours of monotonous work that eats into the budget and slows down decision-making. GIGASCHOOL, an educational platform, faced this problem when promoting services through b17.ru. By implementing an AI agent, the company reduced the marketer's time for analyzing advertising campaigns from 11 hours to 1 per week and identified a critical product problem, which saved 47,600 rubles per month on potentially ineffective campaigns.

In modern marketing, time is money, and manual data processing leads to double losses: first, on the salary of a specialist who copies numbers, and then on lost opportunities due to slow decisions. This is especially acutely felt when an advertising platform does not provide an API, forcing marketers to spend hours on routine instead of strategic analysis. But even such pain can and should be automated to reveal the true business problems.

The reality of the problem: routine that kills efficiency

Maxim Todorov, the founder of a company promoting services through b17.ru, faced a classic dilemma: the channel attracted high-quality and inexpensive leads, but scaling was impossible due to the lack of an API. The b17.ru platform, aimed at psychologists, was not adapted for automated data export and actively resisted bots. This meant that every operation, from collecting statistics to managing ads, required manual intervention.

The marketer spent about 11 hours a week on routine tasks: collecting statistics on impressions, clicks, and expenses, attributing leads to specific ads, calculating metrics, and performing basic analysis. These hours were spent switching between the website, spreadsheets, and internal systems, leaving little time for strategy. A full review of the advertising campaign occurred no more than once a month, by which time some data was already outdated, reducing adaptability and efficiency.

The path to an AI agent: when conventional automation is powerless

Traditional automation tools could not solve the problem with b17.ru due to the lack of an API and anti-bot protection. It was necessary not just to automate individual steps, but to create an intelligent system capable of imitating human actions on the site, collecting data, analyzing it, and suggesting solutions. The company realized that an AI agent was needed that could not only process information but also interpret it, freeing the marketer from the role of an operator.

The goal was not to completely exclude humans from the process, but to elevate them to a higher level, where they focus on strategy and decision-making, rather than manual data collection and consolidation. The AI agent was supposed to take over manual export, spreadsheet consolidation, metric calculation, initial anomaly detection, and drafting recommendations.

How the AI agent was designed: orchestrating data and intelligence

The AI agent was designed as a complex system comprising several interconnected components. This was not just a language model, but an entire pipeline based on n8n, a browser extension, Google Sheets, a Telegram bot, and DataLens. Thanks to this architecture, the agent could not only collect data but also process, analyze, and present it in a convenient format.

System components:

• Browser extension. Responsible for extracting data from b17.ru, imitating user actions and bypassing anti-bot protection.
• n8n. Acted as an orchestrator, managing the entire process: from data collection to results delivery.
• Google Sheets. Used for intermediate storage, cleaning, and standardization of data.
• Language model. Responsible for analyzing prepared metrics, finding patterns, anomalies, and formulating recommendations.
• Telegram bot. Served as an interface for launching the process and receiving reports.
• DataLens. Used for monitoring the technical and economic performance of the system.

The agent's scenario was designed so that the user initiated data collection through a browser extension, and then the entire process continued automatically. The agent collected more than 20 parameters for each ad, cleaned and standardized them. It was important that data collection, cleaning, and metric calculation were performed by regular code, ensuring predictability and reproducibility of results. The language model was involved at the stage of data interpretation and hypothesis generation, not calculations.

Implementation: from pilot to strategic tool

The development and implementation of the system took about 80 hours. Maxim Todorov started by manually assembling the first nodes, then used Perplexity for code generation within the nodes, and finally connected Claude Code. This approach allowed for process control and ensured the correctness of the logic.

The agent was successfully integrated into the workflow. The marketer launched data collection via the browser extension and then received a ready-made report in Telegram and HTML format with metrics and recommendations. The system also included a table with filters for convenient ad sorting. The average run time was about 124 seconds, and the estimated cost of processing one ad was 52 kopecks.

An separate error handling circuit was provided in n8n, which allowed for tracking failures and promptly resolving them. The correctness of the model's response scheme was 88% even when performing injection tests, demonstrating the system's high stability.

Results: savings, efficiency, and most importantly, problem identification

Although the campaign was stopped before CPL reduction and annual ROI could be fully verified, the agent showed significant savings on routine operations. The estimated payback period for the system would have been 3-6 months.

However, the main result was not this. The agent helped quickly collect the necessary data and confirmed that the b17.ru advertising channel was working and generating leads. But then it became clear that the problem lay deeper: the product was not sufficiently adapted to the audience that the ads were bringing in. The AI agent quickly identified this bottleneck, showing that there was nothing to scale yet, and allowed the company to make a strategic decision to discontinue the project, thereby saving significant funds and time on further ineffective advertising campaigns.

How to implement this in your company

The GIGASCHOOL case demonstrates that even without an API and with active bot protection, it is possible to automate data collection and analysis if the task is approached comprehensively. Here's where to start if you have a similar problem:

• Use browser extensions. If there is no API, this is an effective way to extract data by imitating user actions.
• Orchestrate the process. An AI agent is not just a language model, but an entire pipeline of various tools that work in conjunction. Use platforms like n8n to manage data flows.
• Separate tasks for AI and code. Code is better at collection, cleaning, validation, and calculations, ensuring predictability. The language model is effective for data interpretation, pattern finding, and hypothesis generation.
• Start small. Automate the most painful and routine part of the work, where the benefit will be obvious and the risk minimal. This will help get initial results quickly and convince the team of the solution's value.
• Monitor metrics. Use dashboards (e.g., DataLens) to monitor system operation, cost, and response correctness.

If this case sounds like what's happening in your company, our manager can help: he'll analyze your business and niche for free and point out where an AI agent would bring a real result in your case. Message the manager