Overview
Our optimization model is a machine learning system that continuously analyzes campaign performance and adjusts how media is allocated to improve results over time.
By learning from real campaign data, the optimizer helps you maximize performance without manual intervention, making smarter, faster decisions at scale.
What the Optimizer Does
The optimizer automatically:
Identifies which inventory is driving the strongest results
Shifts delivery toward higher-performing opportunities
Reduces spend on underperforming segments
Continuously adapts as performance changes
This enables campaigns to improve efficiency while maintaining scale.
Why It Matters
Manual optimization can be:
Time-consuming and difficult to scale
Reactive instead of proactive
Limited by incomplete data or human bias
As campaign complexity increases across publishers, geographies, and timing, it becomes harder to identify what’s working, and act on it quickly.
The optimizer solves this by continuously monitoring performance and making data-driven adjustments in real time.
How It Works (High-Level)
1. Measure Performance
The optimizer analyzes recent campaign activity using outcome-based metrics like conversion or response rate.
This ensures decisions are based on real results, not just delivery metrics like impressions.
2. Identify High-Performing Segments
Inventory is grouped into performance segments based on factors such as:
Publisher and supply source
Geography (e.g., ZIP code)
Deal or inventory type
Time of day
The optimizer compares these segments to determine where performance is strongest.
If a campaign is early and has limited data, the model may incorporate broader historical patterns to guide initial decisions.
3. Adjust Allocation
Instead of directly changing bids, the optimizer adjusts how traffic is distributed:
Higher-performing segments receive more opportunity to win impressions
Lower-performing segments receive less
This allows the campaign to naturally shift toward better-performing inventory.
4. Continuously Learn and Adapt
The optimizer continuously evaluates new data and updates its strategy throughout the campaign.
As performance patterns change, the model adapts, ensuring optimization remains aligned with current conditions.
How to Use Campaign Optimization
The optimizer is designed to support performance-driven campaigns, especially when:
You are optimizing toward outcomes like web conversions
Campaigns run for multiple weeks (allowing learning to compound)
You want to reduce manual optimization effort
It works best as an always-on optimization layer, complementing your campaign strategy.
Best Practices
Start Broad
Allow flexibility at launch (e.g., broader publishers, geographies, and timing).
This gives the model more data to learn from and improves optimization outcomes.
Maintain Continuity
The optimizer learns within each line item:
Learning does not transfer across line items
Restarting a line item resets learning
Recommendation:
Use fewer, longer-running line items to maximize performance gains.
Allow Time to Learn
Initial signals may appear within a few days
Stronger optimization typically develops over ~2 weeks
Giving the model time ensures more stable and effective optimization.
Requirements & Minimums
To enable effective optimization:
Conversion Tracking: Attribution pixel must be implemented and functioning correctly
Minimum Spend: $500 per line item
Minimum Duration: 14-day line item flight
The optimizer uses a rolling learning window, and sufficient data volume is required to generate meaningful insights.
Important Considerations
Optimization impact may be limited early in the campaign while the model is learning
Highly restrictive setups (e.g., narrow geos, limited inventory) can reduce effectiveness
Learnings are specific to each line item and do not carry over