11.11 Chapter Review
11.11.1 Chapter summary
Forecasts learn from historical patterns, while projections and scenarios are conditional on stated assumptions. Reliable future-oriented work preserves time order in validation, uses compatible denominators and periods, reports uncertainty, and makes CLV components such as margin, retention, churn, acquisition cost, and discounting explicit.
11.11.2 Common mistakes
- mixing partial and complete time periods;
- selecting a model using random time splits;
- extrapolating far beyond the historical range without qualification;
- presenting scenarios as statistical intervals;
- assuming recommendation effects and then calling them forecasts;
- calculating utilization with incompatible denominators;
- treating revenue CLV as profit;
- including new customers in retention rates;
- applying one churn rate to all tenures without checking; and
- comparing CLV across cohorts with unequal observation windows.
Quality standard: Every forecast, projection, and CLV result should expose its time horizon, population, inputs, assumptions, validation, uncertainty, and intended decision use.
11.11.4 Exercises
Exercise 1. An NVRW campaign costs $18,000 and acquires 300 members. Expected annual revenue is $360 per member, contribution margin is 40 percent, expected active lifespan is three years, and discounting is ignored. Calculate CAC and the simple margin-based CLV.
Check Your Work
CAC is $18,000 divided by 300, or $60 per acquired member. Annual contribution margin is $360 times 0.40, or $144. Three-year contribution is $432, so simple margin-based CLV after CAC is $432 minus $60, or $372. The estimate assumes every acquired member remains active for three years.
Exercise 2. Use data/nvrw/derived/monthly_visits.csv to compare a seasonal naive forecast, a moving average, seasonal regression, and one ETS or ARIMA model. Validate each method using rolling twelve-month forecast origins.
Check Your Work
All methods must use the same complete monthly series, forecast horizon, rolling origins, and error metrics. The seasonal naive method is the baseline. Select a method using out-of-sample performance across origins rather than in-sample fit, and inspect whether errors are concentrated in particular facilities, seasons, or unusual periods.
Exercise 3. Reproduce data/nvrw/derived/utilization_projection.csv, then create low and high scenarios by varying both population and utilization. Do not describe the scenarios as confidence intervals.
Check Your Work
The base calculation should equal projected population multiplied by the stated utilization rate for the same geography, service, and period. Low and high scenarios should expose both altered inputs and preserve the same units. Label them conditional scenarios. They are not statistical confidence intervals because their bounds come from selected assumptions.
Exercise 4. Recalculate CLV for each membership type when retention is five percentage points lower and higher than the central estimate. Identify which assumption has the largest practical effect.
Check Your Work
Change retention by 0.05 in each direction while holding margin, CAC, discount rate, and horizon constant. Keep rates within zero and one and use consistent annual periods. Compare the dollar and percentage change in CLV by membership type. Retention will usually have a larger dollar effect for high-margin products, but the conclusion should be based on the recalculated sensitivity table.