ADSTACK

What AppLovin actually is

Everything here is sourced from AppLovin's own support documentation (support.applovin.com, read Aug 2026). Strip the branding and the platform is one loop with three doors.

The user-acquisition engine

Campaigns are goal-based; the optimization model is fixed at creation and cannot be changed later. Two billing modes: target-goal bidding with CPI billing (cost per result held near target, charged per install) and auto-bidding with CPM billing (spends full budget at lowest cost per result, charged per impression).

Everything else you can do

• Targeting is deliberately thin: countries (~200), per-country budgets and goals, US states/metros, creative-set-level language (50 languages). No demographics, interests, OS versions, or device models. The model does the targeting.
• Creative: 9:16 video ≤60s ≤1GB (re-transcoded), single-file MRAID playables ≤5MB with a full funnel-analytics API, banners/MRECs/endcards. Creative sets (10 files per type) attach to many campaigns; edits propagate globally.
• Measurement is delegated: an MMP is mandatory (Adjust, AppsFlyer, Kochava, Tenjin, Singular, Branch) — ≥7-day click window, ≥24h view-through, lifetime event postbacks, and strict revenue hygiene: no pLTV, no ad revenue, no trials in purchase postbacks.
• Four APIs: Reporting (day-N ROAS/retention cohorts, 0d→1y), Asset Reporting, Playable (HTML) Metrics, and full Campaign Management (1,000 req/min; 100 errors in 5 min = 24h block).
• Price rails by geo tier: CPI-billed caps from $10 CPI / $1,000 CPP (US) down to $0.50 / $150 (tier 4); a multi-geo campaign inherits its lowest tier's cap.
• Retargeting exists (campaign_ad_type: rt) but is otherwise undocumented; retention appears only as reporting columns, never as a goal.

E-commerce demand, game-inventory supply

The newer product: DTC brands and lead-gen sites buy the same game inventory. The ad is a fixed three-act sequence — full-screen video, interactive page, dynamic catalog — in interstitial (skippable after 5s) or rewarded (≤60s) placements.

• Audiences: Universal (everyone), Prospecting (new customers), Discovery (never visited your site). The latter two require Shopify integration or purchase-history uploads (CSV/SFTP ≤200MB).
• Goals: ROAS or CPP at Day 0 or Day 7 ("Day 0 suits quick-purchase items, Day 7 higher-value products"), CPL/CPE for lead gen. Auto-bidding only — no manual bids.
• Signal stack: AppLovin Pixel (GTM & GA4-compatible; page_view → view_item → add_to_cart → begin_checkout → purchase), server-side Conversion API (100-event batches, deduped against the pixel), one-click Shopify app, hashed email/phone identity. Click IDs must survive redirects or attribution breaks.
• Dynamic catalogs: Shopify-synced, CSV feeds, or pixel-inferred; the docs' one lift claim — brands with 5+ SKUs "typically see a 10–25% lift in conversion metrics." A no-feed Offering Catalog covers services/lead gen (six offers minimum recommended).
• Attribution: clicks-only or clicks + 1-day views; purchase windows D0 (24h), D7 (192h), plus D14/D28. Twelve pre-integrated third-party measurers (Northbeam, Triple Whale, Measured, Fospha…).
• Creative doctrine, with receipts: among top-5%-spend videos — median 37s, 67% of spend on 31–60s cuts, 95% captioned, ~40% carry a promotion; interactive pages drive 50%+ of a creative set's clicks.

The supply side

• Unified auction: every bidder (Google, Meta, Unity, ironSource, Liftoff, Mintegral, Moloco, Pangle, Amazon APS…) competes per impression; traditional CPM lines can sit alongside. Formats: banner, MREC, interstitial, rewarded, native, app open.
• Optimization tooling: A/B test any waterfall change (50% default split; docs say promote after 24h stable, >10k impressions/day, >2% incrementality); up to 8 segmented waterfalls per ad unit keyed on IDFA state, device type, or SDK-set numeric segments (0–32,000) with AND/OR/NOT rules.
• Programmatic control: Ad Unit Management API (2,000 req/hr) for placements, per-network CPMs, country bid floors, frequency caps, banner refresh locked to 0/10/15/20/30/45/60/300s.
• Reporting: dashboard (fill rate, eCPM, ARPDAU, impressions/DAU) plus revenue API, user-level revenue API, cohort API, and impression-level S2S.
• QA built in: in-app Mediation Debugger (per-network integration status, live-waterfall isolation), Test Mode, Creative Debugger, and dashboard Ad Review to flag bad creatives across 20+ networks.
• House rules with numbers: banner refresh ≥10s, one ad unit per format per app, ~5 ads per user per session, rewarded video ~1–2/day, "focus on global revenue, not eCPM," and a valid app-ads.txt — "many advertisers do not bid" without it.

What the docs say — and refuse to say

The support site documents ROAS mechanics, never outcomes. These are the only ROAS numbers that exist anywhere in the source:

You choose the target ("1 = 100%", percentage return on the targeted day), and the docs' entire goal-setting doctrine is one sentence: set goals in line with your profitability targets, because unrealistic goals "may prevent the campaign from spending or scaling effectively." Over-aiming doesn't buy better ROAS — it buys silence. Reporting then tracks realized ROAS at 0d/1d/2d/3d/7d/14d/28d/30d/90d/1y, so cohorts accrete long past the optimization day.

Action items: Roblox builds its own AppLovin

Roblox already owns what AppLovin had to assemble: first-party immersive inventory, a closed-loop currency (Robux) that makes every purchase a clean conversion signal, engagement telemetry, and an existing ads beachhead. What's missing is the machine around it. Six phases, each mapped to the AppLovin capability it replicates.