Introducing BAT: Russias First Open-source Simulator for AI Algorithm Testing in Ad Auctions

The artificial intelligence team at Avito has unveiled BAT (Benchmark for Auto-bidding Task) — the first Russian open-source simulator designed to test and optimize bidding algorithms in advertising auctions. The researchers introduced the platform at The ACM Web Conference 2025 in Australia, which is one of the premier global events in the field of machine learning. **The simulator is now available on** [**GitHub**](https://github.com/avito-tech/bat-autobidding-benchmark)**.**

BAT simulates authentic advertising auction environments, allowing developers to integrate their algorithms, execute simulated campaigns, and assess the performance of various models prior to production launch. This facilitates the discovery of solutions that yield more clicks while maintaining the same budget.

According to our estimates, BAT is the first open-source «sandbox» for this task in the past 12 years. Previously, the industry relied on the open dataset iPinYou, established in 2013. While it marked the initial step toward open testing environments in digital advertising, the industry has undergone significant changes since then: new formats have emerged, data volumes have increased, and bidding algorithms have become more complex.

Today, advertising platforms handle millions of requests per second and utilize advanced machine learning models to predict user behavior. These new conditions impose heightened demands on the quality of advertising auction simulations.

BAT is built on real anonymized data, which is 1,000 times larger than the corresponding metric for iPinYou. This allows for the testing of algorithms in conditions closely mirroring the operations of contemporary advertising systems — characterized by high load and intricate scenarios.

According to Avito researchers, BAT will benefit:

**Advertising platforms** — by increasing revenue by 10-20% through more accurate ad placements;

**Advertisers** — by achieving up to 20% more clicks at the same budget;

**Developers** — by enabling experimentation with algorithms without the need to construct complex infrastructure;

**Users** — by seeing more relevant ads.

The BAT platform has the potential to significantly democratize the advertising technology market — smaller platforms and startups will gain access to tools that were previously available only to large players. This will lower entry barriers, simplify experimentation, and accelerate innovation adoption.

How BAT Works

Every time a user opens a search or recommendation feed, the system initiates an instant auction among advertisers. Algorithms calculate bids based on relevance, creative quality, competitor bids, and other factors. The ad with the highest overall rating wins, with the advertiser paying the lowest possible price.

BAT allows for the replication of this process in a testing environment. Developers can connect their algorithms, select scenarios, set parameters, and compare the effectiveness of their solutions. The platform already features five fundamental algorithms from Avito that can be utilized as references.

Avito plans to compare different baseline algorithms on the BAT platform, aiding developers in identifying the most efficient approaches for various tasks. The team also hopes to enhance the platform’s capabilities to accommodate more complex and adaptive automatic bidding models.

To learn more about the work of AvitoTech engineers, visit [our website](https://clc.to/eJezlg). If you want to help people and businesses through technology, consider joining the Avito engineering team. Current job openings can be found [on our career page](https://clc.to/1QTHTg).