Illustrative data: both periods have 100,000 requests. Revenue falls from $160 to $150 while CPM rises.
Average CPM only describes the impressions that still cleared. It does not show the auctions lost when a new floor filtered lower bids. A floor change can make the reported CPM look better while fill, competition and total monetized requests fall enough to reduce revenue.
Segment the result by placement, geography, device and demand path. Compare revenue per session or thousand requests, fill rate, bid rate, bid distribution and request volume against a stable holdout. Watch for buyers shifting spend to other supply and for request or page latency changes that alter the comparison.
Roll back or narrow the rule where incremental CPM does not cover the lost fill. Expand only after the holdout shows a durable revenue lift, and keep monitoring bid density so the rule does not silently reduce auction competition over time.
Compare revenue, not only clearing CPM
Calculate revenue as monetized impressions divided by 1,000, multiplied by CPM. Keep the request population comparable before attributing the change to a floor.
Set up a useful comparison
- Define the affected placement and traffic segments before the change.
- Use a stable, comparable control where practical.
- Record requests, filled impressions, revenue, bid density, and latency.
- Allow enough observation to assess normal variability and changes in traffic mix.
Keep, narrow, or reverse the pricing rule
Keep the change where normalized revenue improves without unacceptable delivery or experience costs. Narrow or reverse it where the evidence shows loss.
One before-and-after day does not isolate the floor's effect. Report uncertainty when traffic mix, seasonal demand, or other configuration changes overlap.
Test your reasoning: Judge the whole result
Original simulated exercise. This is not a real client case.
A pricing experiment improves CPM.
Evidence available
- Control: 90,000 impressions at $2 CPM.
- Treatment: 55,000 impressions at $3 CPM.
- Both groups contain 100,000 comparable requests.
Which group earns more in this example?
Read the answer and next check
Control earns $180; treatment earns $165. Higher CPM does not cover the loss of monetized impressions.
Next check: Assess variability and traffic comparability before deciding whether to keep the rule.