AI agents are actively used in a wide variety of fields: analytics, sales, customer support, HR, inventory management, etc. In the field of advertising monitoring, they allow for the collection and analysis of data on advertising placements on the internet, assessing their effectiveness and identifying competitive strategies. One example of such AI agents is Banner Stat, developed by M-Social.

Banner Stat – is a scraping tool for advertising placements, consisting of three interconnected AI agents that process advertising placements step-by-step, turning raw data from web pages into structured advertising reports necessary for:
- monitoring advertising campaigns;
- competitor analysis;
- discovering new advertising strategies;
- analyzing the effectiveness of advertising placements;
- optimizing media planning

AI Collector

The first AI agent that is part of Banner Stat is the AI Collector. Its goal is the automated collection of advertising banners and their associated metadata from web pages using computer vision (CV).

Mechanics:

• Uses Selenium for rendering web pages and bypassing blocks.

• Uses ViT (Vision Transformer) and CNN (Convolutional Neural Network) for detecting advertising banners on web pages.

• Extracts the DOM structure of pages and matches it with found banners.

• Determines banner coordinates and extracts meta-information for further analysis.

Result:

Collection of metadata about the advertising placement: url, image, display time, placement dimensions, placement platform.

AI Typifier

The second AI agent is the AI Typifier, the goal of which is to determine the type of advertising banner (video banner, display, RTB-block, branded pages, preroll).

Mechanics:

• Uses a CV model (ViT + CNN) to analyze the visual content of the banner.

• Uses feature clustering (size, animation, loading of additional resources) to classify the ad format.

Result:

Categorization of the advertising format.

AI Attributor

And the third AI agent, part of the Banner Stat system, is the AI Attributor, which serves to extract advertising attributes, such as brand, product category, and advertiser, from the collected data.

Mechanics:

• Uses a BERT classifier to determine the ad category.

• Applies an NER model to extract brands and advertisers.

• Analyzes metadata, images, and banner text using NLP and heuristic rules.

• Cross-references data with known brand and advertising platform catalogs.

Result:

Advertising placement attributes: advertiser, product category, brand.

So what is the final result accumulated in Banner Stat?

• Advertising placement metadata.
• Advertising placement format.
• Advertising placement attributes.
• Estimated OTS (opportunity to see) - calculated using an algorithmic methodology based on data from AI agents.

Conclusion

Banner Stat — is not just a tool, but an example of how artificial intelligence is changing approaches to data analysis and advertising campaign management, making them more efficient and effective.