How the diagnosis works
Why blended eCPM should be reviewed last.
Revenue is the final symptom. Work upstream from exposure and serving, then check mix, placement, sources, timing, and only then weighted pricing.
01
Impressions
- What it answers
- Did the revenue drop start with less ad exposure?
- Metric definition
- Count shown ads for a matching period. Pair it with DAU, sessions, or eligible moments when available.
- Common misread
- Stable eCPM does not protect revenue when fewer ads are shown.
- What to check next
- Compare requests and fills for the same country and placement.
Directional sample02
Requests, matches and fills
- What it answers
- Are requests still arriving, and are they turning into filled opportunities?
- Metric definition
- Keep requests, matched requests, fills, show rate, match rate, and fill rate distinct because platforms define them differently.
- Common misread
- A low source-level rate is not automatically an app-level serving failure.
- What to check next
- Split the affected serving metric by country and format.
Directional sample03
Country mix
- What it answers
- Did impression share move toward lower-value GEOs?
- Metric definition
- Compare country-level impressions, revenue contribution, and weighted eCPM across matching periods.
- Common misread
- A lower blended eCPM does not prove global demand fell.
- What to check next
- Keep Tier 1 pricing separate from traffic composition.
Directional sample04
Placement and ad unit
- What it answers
- Is one format, placement, or ad unit driving the movement?
- Metric definition
- Keep placement, ad unit, and format separate whenever the export distinguishes them.
- Common misread
- A global average can hide a single rewarded or interstitial issue.
- What to check next
- Compare the same placement by country and app version.
Directional sample05
Ad source and mediation
- What it answers
- Did one bidder or waterfall source lose contribution?
- Metric definition
- Read bids, fills, wins, impressions, revenue share, and eCPM in the context the platform provides.
- Common misread
- A bidder that wins less often may still be healthy if total serving is stable.
- What to check next
- Check availability, adapters, floors, consent, and source configuration.
Directional sample06
Time and external events
- What it answers
- Was the decline concentrated in an important hour or event window?
- Metric definition
- Compare peak hours, nearby normal hours, event days, and non-event days instead of only daily totals.
- Common misread
- A live event can look like a pricing or mediation failure in a daily chart.
- What to check next
- Inspect impressions per active user before changing settings.
Directional sample07
Weighted eCPM
- What it answers
- Did pricing move after upstream volume and mix signals stayed stable?
- Metric definition
- Use revenue divided by impressions, weighted across matching segments; never treat it as a raw mean of segment eCPMs.
- Common misread
- A higher eCPM can still produce less revenue if fill or exposure fell.
- What to check next
- Compare country, format, source, and time-window pricing before changing floors.
Directional sampleDirectional diagnosis
What this method does not claim
- It is a directional CSV diagnosis, not final attribution.
- It does not replace platform reporting or internal analytics.
- It does not guarantee higher revenue; it narrows the first checks.