The decision in one line
Do not add a new demand partner until you can explain where the current revenue is gained, lost and constrained.
1. Define the decision before opening a dashboard
A useful baseline is not a giant export. It is a compact view built around a decision: improve fill, raise the value of an impression, protect viewability, reduce latency, or change the mix of inventory. Pick one primary question and a small set of guardrails before comparing numbers.
2. Map the inventory you actually sell
List every meaningful ad experience by template, device and position. Separate display, video, sticky and interstitial formats. A single site-wide average hides the page types that create value and the ones that create friction.
- Template and content type
- Device and viewport
- Ad format, size and position
- Refresh or lazy-load behaviour
- Direct, programmatic and house demand
3. Use a metric chain, not one headline KPI
Revenue is an outcome. Read it through the chain that creates it: eligible page views, ad requests, matched requests, impressions, viewability, price and latency. Page RPM or session RPM is the commercial summary; the earlier metrics explain why it moved.
- Page and session RPM
- Request, match and render rates
- eCPM by meaningful segment
- Viewability and time in view
- Ad latency and Core Web Vitals
4. Segment until the action becomes obvious
Break the baseline down by device, geography, traffic source, template and format. Stop when a segment points to a different action. A low mobile RPM caused by slow rendering needs a different response from a low RPM caused by weak buyer demand in one market.
5. Change one commercial variable at a time
Set a control, a defined test window and a success threshold. Whether you change a floor, size mapping, demand source or placement, keep everything else stable enough to interpret the result. Record both revenue and experience guardrails.
6. Treat user experience as a revenue input
Ads that shift content, delay interaction or crowd the page can weaken the audience that future revenue depends on. Reserve space for slots, lazy-load below-the-fold inventory and avoid loading every placement with equal priority. Measure viewability together with speed and engagement.
7. Earn the right to add more demand
Once the baseline is stable, a new demand partner becomes a testable hypothesis instead of a hopeful integration. Compare incremental revenue after fees against latency, auction overlap, operational cost and any change in user behaviour. If the uplift cannot be isolated, the integration has not yet proved its value.
What to do next
Build one weekly scorecard from this baseline. Use it to choose a single experiment, keep the control visible and write down what changed. That operating rhythm creates better yield decisions — and better future insights — than adding tools without a diagnosis.