Skip to content

t1k:marketing:iaa:mediation-waterfall-strategy

FieldValue
Moduleiaa
Version0.3.4
Effortmedium
Tools—
/t1k:marketing:iaa:mediation-waterfall-strategy

Mediation Waterfall Strategy helps monetization leads design and optimize the ad network stack that maximizes eCPM across all placements. The choice between programmatic in-app bidding (MAX, Open Bidding) and classic waterfall, the ordering of networks, eCPM floor configuration, and network A/B testing all directly determine how much revenue each impression generates. This skill covers the bidding-vs-waterfall decision, floor-price design, network ordering heuristics, and testing methodology.

  • Designing the initial mediation stack for a game that has cleared 10K DAU and is scaling UA
  • Evaluating whether to migrate from waterfall to bidding (e.g., adopting AppLovin MAX or AdMob bidding)
  • Diagnosing a fill-rate drop or eCPM floor problem after a network change

This skill resolves the mediation architecture decision (bidding vs waterfall vs hybrid), eCPM floor setting, and network prioritization for a given game and ad format. It is used by monetization leads and UA analysts. It supports decisions about which mediation SDK to center around, how to configure floors, and how to A/B test network changes without regression risk. It does NOT cover SDK integration steps or placement code — those belong to the sibling monetization module.

First-look floor setting — use P90, not the average

Section titled “First-look floor setting — use P90, not the average”

For a first-look auction (a buyer that sees inventory before the waterfall, e.g. CloudX First Look), set the floor at the P90 CPM — the value at/below which 90% of bids fall — plus a small margin. This makes the first-look buyer win only the top ~10% of demand, so it does not cannibalise the downstream waterfall. P90 ≠ average: a long-tail bid distribution can put P90 at 2–3× the mean, so you must pull the real distribution.

The math depends on the data grain (the key trap):

  • ILRD (one row per impression) → plain percentile(cpm, 90), no weighting (each row is already one impression). Lookback 7–14d.
  • Mediation reporting (aggregated: app × network × date × format × geo) → impression-weighted P90 via the empirical CDF (sort by eCPM, cumulative impressions ÷ total ≥ 0.90, interpolate). Lookback 15d. A naive percentile(ecpm, 90) on aggregated rows treats a 10-impression row like a 100K-impression row and inflates the floor — weighting is mandatory.

Segment by app × format × geo (US vs ROW). Treat P90 as a starting point; tune on fill + incremental revenue.

Full step-by-step + the reusable AI prompt template: first-look P90 floor methodology.

Deepen iteratively via /t1k:sync-back as real usage drives priority. Priority sources for when deepening: brainstorm bidding-vs-waterfall decision tree, eCPM floor methodology from ARPDAU benchmarks.

  • references/first-look-p90-floor.md — first-look (pre-waterfall) bid-floor methodology: impression-weighted P90 vs plain percentile, by data shape.
  • See brainstorm: theonekit-core/plans/reports/sales-modules-brainstorm-260527-1625.md
  • Sibling module: monetization (per-network SDK setup) — iaa covers cross-network FORMAT and PLACEMENT strategy