Ad Mediation

What ad mediation does

Ad mediation is the layer that manages competition among multiple demand sources for an app's ad inventory. Instead of integrating one ad network and accepting its prices, the app integrates a mediation SDK, and the mediation layer decides — per ad request — which of the connected networks gets to serve. The chain: app ad request → mediation → demand sources compete → winning ad → user.

A mediation platform concretely provides:

• One SDK wrapping many. Adapters for each connected network, so the app ships one primary integration instead of maintaining a dozen raw SDKs.
• Selection logic. Historically a waterfall (networks called in order of expected eCPM); now predominantly real-time in-app bidding, where networks submit actual bids per impression.
• Unified reporting. Revenue, impressions, fill rate, and latency per network, per placement, per geography — in one place, on one definition.
• Operational controls. Placement configuration, frequency caps, network on/off switches, A/B tests, and segment-level rules without app releases.

The major platforms are AppLovin MAX, Google AdMob mediation, Unity LevelPlay, and DT FairBid. Note the structural tension: every major mediation provider also operates its own ad network, which bids in the auctions the platform referees. Treat auction transparency and reporting honesty as selection criteria, not assumptions.

Waterfall versus in-app bidding

The waterfall ranks networks by historical average eCPM and calls them sequentially until one fills. Its flaws were chronic: historical averages misprice individual impressions, every sequential call adds latency, and high-ranked networks could cherry-pick premium impressions while returning the rest. Operating a competitive waterfall meant endless manual tier tuning — many teams maintained duplicate instances of the same network at different price floors just to force granular competition.

In-app bidding (the app-world equivalent of header bidding) replaced rank-by-history with bid-per-impression: connected networks submit real-time prices, highest bid wins. The practical gains are flat latency (one parallel auction instead of a sequential walk), per-impression price discovery, and the near-elimination of waterfall operations work.

Reality is hybrid: most setups run bidding networks alongside a residual waterfall of traditional networks, and the mediation layer merges both into one decision. The direction of travel is clear — bidding keeps absorbing share — but "fully bid-based" is still not universal across networks and regions.

Operating mediation well

1. Add networks for incremental revenue, not count. Each network adds SDK weight, initialization time, and a privacy/compliance surface. Measure marginal ARPDAU per added network; most apps plateau within a handful of well-chosen bidders.
2. Watch latency per network. A slow bidder delays every auction it participates in. Mediation dashboards expose per-network response times — enforce timeouts and drop chronic laggards.
3. Use price floors sparingly in bidding setups. Floors made sense as waterfall competition tools; in real-time auctions, aggressive floors mostly trade fill for thin eCPM gains. Test against holdout placements before believing a floor helped.
4. Segment reporting by geography and format. Network strength is wildly uneven across markets and formats (rewarded video versus interstitial versus banner); a network worth keeping in US rewarded slots may be dead weight in LATAM banners.
5. Reconcile mediation-reported revenue against network payment dashboards monthly. Adapter misconfigurations and discrepancy clauses show up as gaps between reported and paid.
6. Mind the SDK bill of materials. Every adapter version pin, every network's data collection behavior, every privacy-manifest entry is maintenance you own. Audit the adapter list each quarter; remove what doesn't pay rent.

Common mistakes

• Adding networks without measuring latency. The revenue gain from bidder #9 rarely survives the auction slowdown it causes — and slow ad delivery suppresses impressions per session across all networks.
• Using historical eCPM blindly. In residual waterfalls, stale averages systematically misroute impressions; in bidding, historical eCPM is a reporting output, not a control input.
• Ignoring SDK size and maintenance cost. Mediation stacks are a top source of app-size growth, ANR rates, and release friction. The marginal network must pay for its engineering drag, not just its media revenue.
• Optimizing eCPM while fill collapses. Revenue is eCPM × filled impressions; mediation tuning that lifts the rate while starving delivery is rearranging deck chairs. ARPDAU is the metric that keeps mediation honest.
• Assuming the referee is neutral. First-party demand bidding inside its own mediation deserves scrutiny: compare its win rates and prices against third-party bidders for the same placements.

FAQ

Do I need mediation if I only use AdMob? A single network is fine for a first launch or a small app — AdMob alone fills most geographies adequately. Mediation earns its complexity when meaningful revenue depends on competition: typically once an app has the scale for multiple networks to bid seriously.

Which mediation platform should I pick? The honest differentiators: bidding-network coverage in your key geographies, the strength of the platform's own demand (it will be a major bidder), reporting quality, and SDK engineering quality. Game-centric stacks gravitate toward MAX or LevelPlay; broader app categories often start with AdMob mediation. Run a split test if the decision is high-stakes — results genuinely differ by app and audience.

Is in-app bidding always better than a waterfall? At equal network participation, bidding wins on latency and operations. Edge cases persist — some networks still pay better through managed waterfall placements than through their bidding adapters. Hybrid setups exist precisely to capture both.

How does mediation interact with user acquisition? Ad revenue feeds LTV, and several mediation platforms (notably AppLovin and Unity) sit on both sides — their networks buy your inventory while their UA products spend your budget. Impression-level revenue data flowing into your MMP closes the ROAS loop; the app monetization path covers the full revenue stack.

Why is my mediation revenue lower than the dashboards suggested? Network-side discrepancies (their count rules), invalid-traffic clawbacks, unpaid house-ad or cross-promo impressions counted as filled, and adapters silently failing on some devices. Reconcile per network monthly; persistent gaps above a few percent justify investigation or removal.