GIGASCHOOL Shows How an AI Agent Saved 47,600 Rubles Per Month: Automating Advertising Without API

Maxim Todorov, the founder of a company testing service promotion through the specialized platform b17.ru, faced a challenge: the channel provided cheap and high-quality leads, but lacked an API, making advertising campaign automation virtually impossible. A marketer's manual work in data collection, lead attribution, and analysis took up to 11 hours per week. After implementing an AI agent, this time was reduced to 1 hour, potentially saving up to 47,600 rubles per month. However, the main outcome was not the savings, but the rapid validation or invalidation of a business hypothesis.

In advertising campaigns on platforms without an API, marketers manually collect data, match leads to ads, and analyze results. This involves hours of routine work that not only consumes time but also leads to outdated data, slows decision-making, and increases the cost per lead. Every such hour is a missed opportunity for quick hypothesis testing and timely strategy adjustment. Today, this burden can be completely removed.

The Pain of Manual Work in Advertising Campaigns Without API

Maxim Todorov's company actively used the b17.ru platform to promote its services. Although leads from this source were cheaper and higher quality than from other channels, working with the platform presented significant difficulties. The main limitation was the lack of an API, which prevented the automation of data collection and ad management. The platform actively fought against bots, so even basic operations had to be performed manually.

Before automation, the marketer spent about 11 hours a week on routine tasks: collecting statistics on impressions, clicks, expenses, and applications, matching leads to specific ads, calculating metrics, and analyzing campaigns. This process was linear and labor-intensive: the more ads, the longer each stage. As a result, a full strategy review occurred no more than once a month, by which time some data had already become outdated. This slowed down hypothesis testing and prevented quick reactions to changes.

Why Traditional Tools Were Insufficient

Conventional automation tools, such as standard scripts or integrations, were useless due to the lack of an API and active resistance from the platform. A fundamentally different approach was needed – a system capable of simulating human actions in a browser, collecting data, and then processing it, taking all limitations into account. The company was looking for not just a way to export data, but an intelligent assistant that could take on the entire cycle of advertising campaign management, from data collection to generating recommendations.

This led to the idea of an AI agent that could operate in conditions where traditional automation was impossible, freeing the marketer from routine tasks and allowing them to focus on strategic decisions.

Designing the AI Agent: Architecture and Functionality

The AI agent was designed as a comprehensive system comprising several interconnected components. Its task was not only to automate data collection but also to conduct deep analysis, identify anomalies, and generate recommendations for the marketer. The system consisted of:

• Browser extension. Responsible for extracting data from b17.ru by simulating user actions. This bypassed the lack of an API.
• n8n platform. Acted as an orchestrator, managing the entire process, from data collection to result delivery.
• Google Sheets. Used for intermediate data storage and processing.
• Language model. Responsible for analyzing prepared metrics, identifying patterns and anomalies, formulating hypotheses, and drafting recommendations.
• Telegram bot. For launching the process and sending reports.
• DataLens. For visualizing the technical and economic performance of the system.

The AI agent was conceived as a system where the language model does not do everything, but focuses on interpretation and hypothesis generation, while data collection, attribution, calculations, and validation are performed by deterministic code. This ensured predictability and reproducibility of results in stages where errors are critical.

Implementation and System Operation

The process began with a manual launch: the user opened b17.ru and activated a browser extension, which extracted statistics and sent them to n8n. The process then continued automatically. The agent collected over 20 parameters for each ad, cleaned and formatted the data, and checked for omissions and duplicates. Ads were then linked to leads from forms for attribution, allowing evaluation of not only CTR but also cost per lead and conversion rate.

The language model received already prepared metrics. Its task was to interpret the data: it differentiated between ads with low CTR and those with high CTR but low conversion, identified anomalies, and suggested reasons for poor performance. The agent then proposed which ads should be kept, modified, or disabled. These recommendations were not automatically applied but were reviewed by the marketer, who made the final decisions.

Results and Unexpected Findings

After launch, the system sent a brief summary and an HTML report with metrics and recommendations via Telegram. For detailed analysis, a table with filters was created, and DataLens was used to monitor system performance, including execution time, launch cost, and the correctness of model responses. The average runtime for one pass was about 124 seconds, and the estimated cost of processing one ad was 52 kopecks.

The main result was not direct savings. The AI agent quickly removed the bottleneck in the upper part of the funnel, confirming the effectiveness of the lead generation channel. However, further analysis revealed that the company's product was not sufficiently adapted to the audience attracted by the advertising. Thus, automation uncovered a deeper problem in the product and lead processing, leading to the decision to discontinue the project for which the automation was created. This prevented further expenses on scaling an ineffective direction.

How to Apply This Approach in Your Business

The GIGASCHOOL case demonstrates that even in the absence of an API and active resistance to automation, AI agents can be effectively used. If your business has processes where routine consumes time, and traditional automation methods are unsuitable, consider the following steps:

• Use browser extensions for data collection. The lack of an API is not a death sentence. If data is visible in the browser, it can be extracted and transmitted for further processing.
• Divide tasks between code and the AI agent. For deterministic operations (collection, calculations, validation), use regular code. Apply the language model where interpretation, analysis, and hypothesis generation are needed.
• Start with narrow but critical points. Automating routine tasks that slow down business hypothesis testing can provide not only savings but also valuable insights into the viability of your products or channels.

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