How does Price Genius work?

Last edited: 2024/05/06

1. First, we randomly divide the traffic into three groups: Exploration, Optimization and Benchmark:

   1. Publisher can control the share of Benchmark in the UI.
   2. The traffic managed by Price Genius is further divided into two fractions (this split is fixed):
      1. Exploration (10%) - part used for gathering data on how the inventory reacts to price changes, which is then the basis for modeling (we are running multiple parallel A/B tests).
      2. Optimization (90%) - the part where optimal prices are set.

2. The essence of Price Genius is price prediction/modeling:

   1. Each ad unit gets individually estimated price forecasting models (per 2024.05, we have six various models - based on Machine Learning, Time series analysis, or mathematical modeling).
   2. Each model for each ad unit is recalculated and adjusted daily.
   3. Price Genius automatically selects which of the models best fits a particular ad unit and takes its prediction as an optimal price.

3. Optimal prices are set daily (at night); to do that, we are using the GAMs API connection.

4. We compare the performance of the entire Price Genius part of the traffic with Benchmark and present it in a dashboard.

   <aside> <img src="/icons/info-alternate_blue.svg" alt="/icons/info-alternate_blue.svg" width="40px" /> More:

   Price Genius performance dashboard

   How to modify a Benchmark pricing rule?

   </aside>

5. If Price Genius does not outperform your pricing strategy, it automatically switches to ‘standby mode’ and applies your pricing rules to 90% of the traffic allocated to Price Genius.

   <aside> <img src="/icons/info-alternate_blue.svg" alt="/icons/info-alternate_blue.svg" width="40px" /> More: Automatically disabling Price Genius in case of downlift

   </aside>