Marketing Dashboard

Overview

The model projects monthly cash flows by estimating revenue from two sources: (A) new customers acquired during the projection period, and (B) existing subscribers who were already paying before the projection starts. Revenue is calculated separately for web (direct payments via Solidgate) and mobile (app store purchases).

Revenue Multipliers

Multiplier	Value	Meaning
Web Revenue Multiplier	0.82 (82%)	Gross web revenue × 0.82 = net revenue after Solidgate payment processing fees, taxes, chargebacks (~18% deduction)
Store Commission Multiplier	0.692 (69.2%)	Applied to mobile revenue: app store takes a 30.8% commission
Mobile Uplift	Per-app (e.g. 20%)	Ratio of mobile revenue to web revenue, stored in app_mobile_uplift table. E.g. 0.20 means mobile generates 20% of the revenue that web does.

Revenue Formulas

LTV Curve Construction

The LTV (Lifetime Value) curve represents the average cumulative revenue per customer over time. It is built using 6 milestone days: D10, D40, D70, D100, D130, D160.

1. Predicted LTV — from the product_ltv_estimates table, weighted by each country's share of spend (country mix) and each product's share of purchases (product mix). This prevents summing LTVs across products.
2. Observed LTV — computed from real solidgate_orders cohort data using tiered cohort windows: D10 uses cohorts 11+ days old (freshest), D40 uses 41+ days, D70 uses 71+, D100 uses 101+, D130 uses 131+, D160 uses 161+. This ensures each milestone uses the freshest possible data while having sufficient maturity.
3. Blending — observed LTV (80%) is blended with predicted LTV (20%) at each milestone. Observed data is the primary signal; predicted serves as a safety net. If observed data only covers some countries, it is normalized by coverage weight to avoid understating.
4. Monotonicity enforcement — since different cohort ages are used for each milestone, inversions can occur (e.g. D100 < D70 if older cohorts had lower LTV). When detected, the predicted growth pattern is used to extend the curve forward.
5. Missing milestone interpolation — if some milestones have no data, they are filled using typical ratios relative to D100 based on actual billing patterns (weekly trial→monthly, monthly, yearly): D10=25%, D40=55%, D70=75%, D100=100%, D130=125%, D160=140%.
6. Extrapolation beyond D160 — for days 161–180, the model extrapolates using a data-driven decay ratio computed from the last two inter-milestone increments (D100→D130 vs D130→D160), clamped between 0.30 and 0.95.
7. Daily interpolation — between milestones, LTV is linearly interpolated day-by-day. From D0 to D10 it follows a linear ramp from $0.

New Customer Revenue (Source A)

Each projection month, a new cohort of users is acquired. The number of users = monthly_spend / CPA, where CPA is the 90-day average cost per acquisition. Each cohort generates incremental revenue each month based on their age along the ROAS curve (M1, M2, M3, M4, etc.).

Existing Subscriber Revenue (Source B)

Revenue from pre-existing subscribers is based on the 28-day total scaled to monthly of recurring and 1-click payments (from solidgate_orders where payment_type IN ('recurring', '1-click')). This base revenue decays each month using a churn rate derived from the D40→D70 incremental churn from solidgate_subscriptions. The monthly churn rate is capped at 30% to prevent aggressive decay. D10 churn is not used because it includes trial cancellations which overstate churn for active subscribers.

Country Mix

The country distribution is based on the last 30 days of ad spend by country from meta_campaign_daily_by_country. Each country's percentage of total spend determines its weight when blending LTV values. LTV is computed per-country from observed cohort data, then weighted by the country mix to produce a blended LTV curve.

ROAS Anchor Overrides

After running a projection, the base ROAS anchors (M1, M2, M3, M4) are shown as editable fields per app. You can adjust these values and click Re-calculate to re-run the projection with your custom anchors. This replaces the computed anchor values from cohort data while keeping the same curve interpolation and extrapolation logic.

Outflows

Ad Spend: monthly_spend per month (constant, as entered).
Backend Cost: ~8% of total revenue (computed from historical backend costs / total revenue).

Known Limitations

• Ad spend is assumed constant across all projection months. If actual spend ramps up or down, predictions diverge.
• CPA is a 90-day historical average. It does not account for CPA increasing at higher spend levels (diminishing ad returns).
• Rapidly scaling apps (e.g. CloneAI) may show ±20% per-app error due to backward-looking historical data becoming stale.
• D130/D160 observed LTV requires cohorts 131+/161+ days old. For newer apps, these milestones fall back to predicted estimates or typical ratios.
• Mobile revenue is estimated from the web revenue using a fixed mobile_uplift ratio, not from actual app store data.

Backtest Accuracy

Validated via walk-forward backtesting (simulating the model on past cutoff dates and comparing to actuals). Total web revenue prediction error: ±2–6% across multiple cutoff dates. Per-app errors vary: stable apps (Fling, Impresso) typically ±5%, high-growth apps (CloneAI) up to ±20%.