Days until cumulative paid revenue equals paid media spend. Based on paid channels only (media cost basis).
View:
Confirmed (reached within data)Extrapolated beyond M12Projected (immature cohort)
* Paid Only: paid install revenue vs paid media spend. Bars with lower opacity = geometric extrapolation beyond M12 using observed M6→M12 growth rate.
Payback Period Detail Table
Cohort
Payback (yrs)
Payback (days)
Crossed at
M12 ROAS
Status
LTV Curves by Channel
Cost-weighted ROAS progression. Computed as total channel revenue / total channel cost at each window.
Channel
D7
D14
D30
D60
D90
D120
M4
M6
M12
* Organic ROAS is N/A (no cost). See Revenue Per Install table below for organic benchmark.
* TikTok LTV data is from Aug 2024 – Feb 2025 campaigns (prior wave). Recent TikTok (launched Wk 5 2026) is running higher D7 (~34-56%) but has no long-window LTV data yet.
* "Ironsource (legacy)" is Q3 2024 Ironsource data ($49K total spend) — not representative of current Applovin campaigns which launched Wk 10 2026.
Multipliers vs D7 Baseline
Channel
D14/D7
D30/D7
D60/D7
D90/D7
D120/D7
M4/D7
M6/D7
M12/D7
Retention Analysis
Install-weighted retention rates. Includes organic as benchmark.
Channel
D1
D7
D14
D30
D60
D90
D120
M4
M6
Retention Deltas
M4 bump = D90→M4 change (positive = users returning). Paid vs Organic gap at D30.
Channel
D90→M4 (pp)
M4→M6 (pp)
Paid vs Organic D30 (pp)
Organic Benchmark
Revenue per install ($) by window. Organic shown as baseline; gap shows how paid compares.
Revenue Per Install
Channel
D7
D14
D30
D60
D90
D120
M4
M6
M12
Gap vs Organic (%)
Green = within -10% of organic, Yellow = -10% to -30%, Red = below -30%
Channel
D7
D14
D30
D60
D90
D120
M4
M6
M12
Trend
Target Calibration
Historical multipliers and back-calculated D7 floors for 60-day payback at 140% (incl. platform fees).
Channel
Hist D7
D30/D7
D60/D7
D90/D7
D7 for 70% D30
D7 for 80% D30
D7 for 140% D90
Recent D7
Proj D90
Status
Recent D7 actuals: Google ~35%, Facebook ~24%, TikTok ~56% (from last 30 days of campaign data). Applovin: too early (launched Wk 10 2026).
Last 30 Days vs Prior 30 Days
Campaign performance from weekly data. Last 4 weeks vs prior 4 weeks.
By Channel
Channel
Spend
Spend Δ
Installs
eCPI
eCPI Δ
D7 ROAS
D7 Δ
D30 ROAS
By Region
Region
Paid Spend
Paid Installs
Organic Installs
Spend Δ%
Last D7 ROAS
Prior D7 ROAS
D7 Δ
Android vs iOS
OS
Paid Spend
Paid Installs
Organic
D7 ROAS
Prior Spend
Prior Paid
Prior Org
Prior D7
Spend Δ%
Projected LTV (pLTV)
Cascading Ridge regression projections for immature cohorts (Sep 2025 – Mar 2026). Backtested on Jan–Aug 2025 holdout. Q1 2026 sourced from campaign data.
Paid only (Google, Facebook, TikTok). Toggle between projected gross revenue and revenue per install.
Month
Installs
Ad Spend
D7
D14
D30
D60
D90
D120
M4
M6
M12
D90 ROAS
M12 ROAS
Paid Monthly × Channel
Month
Channel
Installs
Cost
D7
D90
M12
D90 ROAS
M12 ROAS
pLTV by Channel
Channel
Installs
Cost
D7
D30
D60
D90
D120
M6
M12
D90 ROAS
M12 ROAS
pLTV by Region
Sep–Dec 2025 LTV data + Q1 2026 campaign data. D7 is actual; longer windows are projected from LTV cohorts only.
Region
Installs
Cost
D7
D90
M6
M12
pLTV by Platform — Paid
Platform
Installs
Cost
D7
D90
M6
M12
D90 ROAS
M12 ROAS
pLTV by Platform — Organic
Platform
Installs
D7
D90
M6
M12
Multiplier Model Backtest
Step-up multiplier accuracy vs 20 fully mature SGM cohorts. MAPE = mean absolute % error. Replaces D0 Ridge model (was 31-110% MAPE).
Experimental: predict LTV from day-zero revenue only. Use for early budget signals — confirm with D7 model after 8 days. D0 aggregate error: D90 ≈ ±9%, M12 ≈ ±7%.
D0 Backtest (Jan–Aug 2025)
Target
Agg Actual
Agg Predicted
Agg Error
MAE
N
D0 Early Signal — Paid Monthly
What the D0 model would have predicted from day-zero data alone. Compare against D7 model above.
Month
Installs
Ad Spend
D0 (actual)
D7 (pred)
D90 (pred)
M12 (pred)
D90 ROAS
M12 ROAS
D0 vs D7 Model Comparison — Paid Monthly
Side-by-side: D0 early signal vs D7 production model. Divergence > 15% is a flag.
Month
D90 (D0 model)
D90 (D7 model)
Δ
M12 (D0 model)
M12 (D7 model)
Δ
Model Health Monitor
Self-correcting system: tracks prediction accuracy over time, replaces predictions with actuals as cohorts mature, and flags when retraining is needed.
Retraining Status
D7 Model — Drift Metrics
Window
Bias
MAPE
Samples
Trend
Status
D0 Model — Drift Metrics
Window
Bias
MAPE
Samples
Trend
Status
Prediction vs Actual — Live Tracker
Shows D7-model predictions alongside actuals as cohorts mature. Green = verified, gray = awaiting maturity.
Cohort
D7
D30
D60
D90
M6
M12
Model Backtest Accuracy
Trained on pre-2025 cohorts (channel×month aggregates), tested on Jan–Aug 2025. Aggregate error = (predicted − actual) / actual.
Window
R²
MAE ($/inst)
Agg Actual
Agg Predicted
Agg Error
N
Revenue Forecast — 2026
Two-layer model: DAU × ARPDAU for total monthly revenue (incl. elder players) + new cohort LTV. Webstore-adjusted. UA scaling 12%/qtr, 2 content drops.
Scenario Summary
Monthly Revenue — Actual & Forecast
Jan 24 – Mar 26 = Adjust actuals (webstore-adjusted). Apr – Dec 26 = base case forecast. Elder % = revenue from users acquired >30 days ago.
For each of the last 6 months (Oct 2025 – Mar 2026), the model was trained only on data available before that month. This tests true out-of-sample forecasting accuracy. Naive baseline = "next month = last month."