Unified Neuroprofile: How VK Combined User Data to Create a 'Smart' Feed and Boost Ad Revenue

VK has unveiled a new technology capable of consolidating fragmented data on user behavior into a single ecosystem. At the core of this innovation is an AI model that generates a so-called "neuroprofile" — an anonymized representation of a person's interests based on the analysis of their actions across various platform services. The solution has already been implemented in "VK Video," "VK Clips," and the "VKontakte" social network, demonstrating significant growth in audience engagement metrics.

Transformer Technology: From Isolated Actions to a Holistic Portrait

Unlike traditional algorithms that often view user actions in each application separately, VK's new model uses a transformer architecture. It evaluates behavior not as a set of disparate events, but as a single anonymized sequence. The system takes into account likes, reposts, comments, content skips, and numerous other signals, correlating them with context.

VK specialists note that the model understands the difference in behavior across different platforms. For example, in "VK Video," users tend to watch content for a long time and like it less often, whereas in "VK Clips," they scroll through the feed quickly but interact with it actively. In the "VKontakte" social network, patterns can be even more complex. The neuroprofile consolidates these assessments into a single representation, allowing algorithms to more accurately interpret user intent and distinguish stable preferences from random clicks.

Implementation Results: Growth in Engagement and Watch Time

The practical application of the new technology has already yielded measurable results. The first update was the implementation of neural ranking, which determines the order of content display in real-time. The system matches the user's neuroprofile with the characteristics of the materials, pushing more relevant content to the top of the feed.

Statistics show that users have become more active in interacting with the updated feed: the frequency of content sharing increased by 15%, opening materials in full-screen mode increased by 10%, and the number of likes increased by 5%. The "VK Clips" metrics stand out, where total content watch time grew by 5.5% after the launch of the new AI model. For users, this means more accurate recommendations that adapt faster to changing interests, even if they switch from one service to another.

Project Economics: Investments and Impact on the Ad Market

Although VK has not officially disclosed the exact investment amount for development, sources in the IT industry estimate the cost of creating such systems to be in the range of 0.8 to 1.2 billion rubles. These funds go towards paying ML specialists and programmers, purchasing graphics accelerators, processing big data, and integrating the solution into several products simultaneously.

The economic effect of implementing the technology is already visible in the company's financial reports. In 2025, VK's revenue amounted to 160 billion rubles, of which 98.3 billion came from online advertising. The entire growth in advertising revenue was driven by the "Video and Clips" segment: video ad revenue grew by 68% to 6.5 billion rubles. Analysts note that the increase in watch time in "Clips" directly converts into additional ad inventory, allowing the company to earn more without the need to attract new users.

Contradictory Data

The question of the exact cost of developing the new AI model remains a subject of debate. VK officials have not provided specific figures, limiting themselves to general words about the scale of the project. At the same time, a source in a major IT company familiar with the development details estimates the investment at 0.8–1.2 billion rubles. Forbes, citing its sources, names a figure of 1 billion rubles. The discrepancy in estimates may be due to the fact that the development cost may include both direct costs for creating the model and indirect expenses for infrastructure and integration, which different experts account for differently.

Future of the Ecosystem and Data Protection

The implementation of neuroprofiles marks a new stage in the development of the VK ecosystem. The company is moving from segmented data analysis to creating a single digital footprint for the user, which allows for a significant increase in targeting accuracy and personalization. At the same time, VK emphasizes that the neuroprofile is an anonymized representation, which should guarantee compliance with personal data protection regulations. However, experts note that combining data from different services requires increased attention to security and algorithm transparency so that users understand exactly how their recommendation feed is formed.