General
Flight Control’s campaign optimization process optimizes the distribution of impressions across different variants, such as creatives, messages, and more.
At the core of this process is the optimization card, which you can add to the campaign map — the visual logic that defines how impressions are routed across different decision paths. Outputs of the optimization card can represent creatives, product sets, messages, or any campaign decision path or variant you want to test.
The optimization algorithm evaluates these multiple outputs using statistical analysis, measuring performance against KPIs such as conversions, impressions, clicks, or custom conversions. Based on the results, the algorithm dynamically drops underperforming outputs and redistributes impressions toward higher-performing outputs until the optimization goal is reached.
Optimization is guided by two key parameters:
- The Optimization Goal defines how many outputs should remain active after underperforming variants are eliminated.
- The Serving Coefficient controls how impressions are distributed among the remaining outputs. It acts as a multiplier on performance differences — the higher the serving coefficient, the greater the gap in impressions between top and lower-performing outputs. For example, if one creative performs 10% better than another and the Serving Coefficient is 3, the lower performer would receive 30% fewer impressions.
How Optimization Works
The optimization process consists of three main phases:
1. Learning Phase
In the Learning Phase, the system collects data on all outputs and tracks performance against configured thresholds — such as minimum impressions, confidence level, relative performance difference, and results count. No outputs are removed during this phase. The goal is to ensure each output has accumulated enough data to support a statistically meaningful comparison.
2. Elimination Phase
Once at least one output meets the defined thresholds, the system enters the Elimination Phase. Underperforming outputs are identified and removed based on their relative performance until the number of remaining outputs matches the Optimization Goal, which is the target number of outputs to retain.
Note: If the number of outputs is already less than or equal to the optimization goal, the Elimination Phase is skipped, and the system moves directly to the Steady Optimized Phase.
3. Steady Optimized Phase
After the optimization goal is met, the system enters the steady optimized phase. Here, the Serving Coefficient is applied to dynamically redistribute impressions across the remaining outputs based on their relative performance.
- If only one output remains, it receives 100% of impressions.
-
If multiple outputs remain, each one is weighted relative to the top performer, with the Serving Coefficient amplifying those performance differences.
- Example: If Creative A performs 10% better than Creative B, and the Serving Coefficient is 3, then Creative B receives 30% fewer impressions than A (10% × 3).
- These weights are automatically recalculated hourly to reflect the latest performance data.
To learn more about setting up the optimization card in your campaign, see the Optimization Card article.