AdMob eCPM Dropped but Impressions Stayed Stable

Direct answer

Confirm first that total impressions are genuinely stable across two comparable, completed periods. Then use weighted eCPM for the same impression set as revenue and split it in a fixed order: country, format, placement or ad unit, ad source, and time. If the decline disappears inside those segments, the blended average is hiding a mix shift. If comparable segments decline together, broader demand or auction value becomes a more credible direction.

Do not assume lower eCPM means an SDK defect or that a higher floor will restore revenue. eCPM is an outcome of served impressions, their markets and formats, source contribution, auction conditions, and timing. The diagnostic goal is to find where the lower value entered the total before changing the serving configuration.

Start with the metric relationship

Revenue ≈ Impressions × eCPM ÷ 1,000

This relationship is for decomposition, not a forecast or earnings promise. Use weighted eCPM: total revenue divided by total impressions and multiplied by 1,000. A simple average of segment eCPMs gives each segment equal weight even when their impression volumes differ.

Stable impressions define this diagnostic branch

A small normal fluctuation can be acceptable, but the impression total should be close enough that lower value per thousand impressions explains most of the revenue movement. If impressions also fell materially, diagnose both quantity and value instead of forcing the case into an eCPM-only explanation.

Weighted eCPM is the relevant blended measure

Recalculate from total revenue and impressions for the exact scope and period. Do not average displayed country or ad-unit eCPMs unless you weight each segment by its impressions.

Confirm the comparison before explaining it

Use equal-length, completed windows with the same weekdays, reporting timezone, apps, ad units, and currency. Avoid comparing a partial day with a full day. Mark holidays, large live events, campaign shifts, and reporting delays. If the decline appears only against one unusual reference period, add a second normal period before treating it as a sustained change.

Calculate total revenue, total impressions, and weighted eCPM for both windows. Quantify how much of the revenue gap the eCPM change would explain at the observed impression volume. The formula is approximate because reporting adjustments and metric definitions can differ, but it prevents an impression loss from being mislabeled as a pure pricing problem.

Split 1: country

Compare each major country's impressions, revenue, and weighted eCPM, then calculate its share of total impressions. Two different effects can lower the blend: eCPM can fall inside a country, or a lower-value country can gain impression share while country-level eCPMs remain stable. Keep these effects separate because one points toward GEO demand and the other toward traffic mix.

Rank countries by their contribution to the lost revenue, not only by percentage change. A small country with a dramatic fall may matter less than a modest decline in the largest market. Hold format and placement constant when possible so a changed rewarded-versus-banner mix does not masquerade as a country effect.

See how country mix changes blended eCPM

Split 2: format

Separate rewarded, interstitial, banner, native, and other formats. They represent different inventory, user moments, and auction values, so a shift in format share can move blended eCPM while the experience inside each format is stable. Compare the same format inside the same countries before deciding that demand moved everywhere.

If one format fell broadly, continue by placement and source within that format. Check whether its country share, app version, or traffic source changed. Do not use a higher-performing format's eCPM as the expected benchmark for another format; the purpose is before-and-after diagnosis within comparable inventory.

Split 3: placement and ad unit

A stable format average can hide movement between placements. One rewarded placement may appear earlier in the user journey, reach a different user cohort, or receive a larger share of impressions after a product change. Compare placement-level impressions, revenue, weighted eCPM, show behavior, and app version without assuming that attention or auction value stayed constant.

Look for the smallest placement or ad-unit group that contains most of the decline. If a release changed frequency, trigger timing, or screen exposure, keep the evidence scoped to that placement. A placement shift can alter both user mix and source competition even when total impressions remain stable.

Split 4: ad source

For mediated inventory, compare source-level impressions, revenue, and contribution share. A high-value source may contribute fewer impressions while lower-value sources replace the volume, leaving total impressions stable but reducing weighted eCPM. That is a source-mix explanation, not proof that every source's auction value declined.

If one source changed, inspect its affected countries, formats, placements, mappings, adapter state, eligibility, and time boundary. Keep rate definitions consistent across reports. Do not remove several sources at once; that changes competition and makes the original source contribution impossible to observe cleanly.

Split 5: time and period

Compare hour of day, day of week, and equivalent seasonal periods. Advertiser demand and user mix can move within a day even when daily impressions are unchanged. A daily blend may fall because more impressions were served in lower-value hours. Check whether the decline is continuous, limited to certain hours, or aligned with a holiday, event, month boundary, or campaign change.

Use more than one reference period when seasonality is plausible. A year-over-year comparison can add context, but only if app scale, countries, formats, and placements remain comparable. The objective is not to find a universal seasonal benchmark; it is to see whether the affected segments moved together at a shared boundary.

Turn the split into a diagnosis

Summarize the surviving pattern with scope and timing. ‘Weighted eCPM fell’ is an alert. ‘Rewarded eCPM fell across the three largest countries after the month boundary while source shares and placements stayed stable’ is a testable demand direction. ‘Country eCPMs stayed stable, but lower-value GEOs gained twelve points of impression share’ is a mix diagnosis. These statements lead to different next checks.

Choose one action that matches the affected scope and could disprove the explanation. Preserve a control country, format, placement, or source when possible. Define the expected signal and observation window, then track total revenue alongside eCPM so a higher price metric does not conceal lost volume elsewhere.

Use the full AdMob revenue-drop flow · Review revenue after a price-floor change