MBA for EXECUTIVES REUNION 2015 MBA for EXECUTIVES REUNION 2015 WELCOME BACK ALUMNI! CLV, From Inside and Out Daniel McCarthy Peter Fader Bruce Hardie Wharton School of the University of Pennsylvania November 14 th, 2015 We acknowledge the support of the Baker Retailing Center, and helpful discussions with Eric Bradlow, Bilge Yilmaz and Chris Ittner. 1
Customer Lifetime Value: Challenges and Opportunities The customer lifetime value ( CLV ) philosophy, in two sentences: I look at the value of a customer and when you make an investment in a customer you can get return on that capital. Not all customers are created equal. --Charlie Ergen (Dish Network CEO) Customer Lifetime Value: Challenges and Opportunities The customer lifetime value ( CLV ) philosophy, in two sentences: I look at the value of a customer and when you make an investment in a customer you can get return on that capital. Not all customers are created equal. --Charlie Ergen (Dish Network CEO) But what is the value of a customer, or CLV? Customer value Company Value? How about customer-based public company valuation? 2
Customer Lifetime Value: Inside and Out The customer lifetime value ( CLV ) philosophy, in two sentences: I look at the value of a customer and when you make an investment in a customer you can get return on that capital. Not all customers are created equal. --Charlie Ergen (Dish Network CEO) High growth cosmetics But what is the value of a customer, or CLV? retailer, using internal Customer value Company Value? customer data How about customer-based public company valuation? $40B pay TV company, using external, publicly disclosed customer data What is CLV? The present value of the future (net) cash flows associated with the customer A forward-looking concept, not to be confused with (historic) customer profitability How to calculate CLV? It depends on the business setting 3
Contractual vs Non-contractual settings In a contractual setting, we observe the time at which customers die Pay TV provider In a non-contractual setting, the time at which a customer becomes inactive is unobserved Cosmetics retailer The challenge of non-contractual settings: How do we differentiate between those customers who have ended their relationship with the firm versus those who are simply in the midst of a long hiatus between transactions? Non-contractual transaction data 4
The RFM Classification Within the direct marketing literature, there is a strong tradition of classifying customers on the basis of RFM: Recency: date of the customer s last transaction Frequency: how many times the customer has bought from us in a specified time period Monetary Value: average value of transactions in a specified time period Our objective is to develop a model to generate forward-looking estimates of CLV as a function of RFM in a noncontractual setting: E(CLV)=f(R, F, M) CLV From the Inside: High-growth cosmetics retail Consider a cohort of 4,127 new customers who were acquired between February 14 th and August 14 th 2014: We track their initial and subsequent purchases over a 12-month period. For testing purposes, split the 12 months into two periods of equal length: group customers on the basis of recency and frequency in months 1 6 ( calibration period ) compute average total spend for each group in months 7-12 ( holdout period ) Ultimately we use all 12 months for CLV analysis 5
Starting Point for Model Building Assume that the amount spent per transaction is independent of the transaction process. our model of buyer behavior can be separated into: a sub-model for the transaction flow (recency, frequency) a sub-model for revenue per transaction Modeling the Transaction Flow ( Buy til You Die ) Transaction Process: While active, a customer purchases with a random per-period purchase probability Purchase probabilities vary across customers Dropout Process: Each customer has an unobserved dropout propensity Dropout propensities vary across customers This is known as the BG/BB model (Fader, Hardie and Shang 2010). Details and Excel implementation at: http://brucehardie.com/notes/010/. For R users: http://cran.r-project.org/web/packages/btyd/ 6
Model Performance (1 of 4): Calibration-Period Histogram Model Performance (2 of 4): Tracking Cumulative Repeat Transactions 7
Model Performance (3 of 4): Tracking Week-By-Week Transactions Model Performance (4 of 4): Conditional Expectations (Holdout Period) 8
Modeling the Spend Process The dollar value of a customer s given transaction varies randomly around his average transaction value Average transaction values vary across customers but do not vary over time for any given individual The distribution of average transaction values across customers is independent of the transaction process. Details and Excel implementation at: http://brucehardie.com/notes/025/. For R users: http://cran.r-project.org/web/packages/btyd/ Distribution of Average Transaction Value 9
Average E(CLV) by RFM Segment Total E(CLV) by RFM Segment 10
CLV By Acquisition Channel Customers acquired through email are much more valuable than other channels high value per acquisition ( VPA ) Moving from CLV to Company Valuation Our original leading questions: What is the value of a customer, or CLV? Customer value Company Value? How about customer-based public company valuation? 11
Moving from CLV to Company Valuation Our original leading questions: What is the value of a customer, or CLV? Customer value Company Value? How about customer-based public company valuation? Moving from CLV to Company Valuation Our original leading questions: What is the value of a customer, or CLV? Customer value Company Value? How about customer-based public company valuation? 12
Augmenting Financials with Customer Metrics Financial Data + Customer Metrics Stock Price? Discounted Cash Flow Valuation Model Benefits: 1. De facto standard valuation method in Finance 2. Flexible (changes in debt / working capital / operating expenses) 3. General (values whole enterprise) 4. Naturally integrates customer metric projections! Income Statement / Balance Sheet Projections Stock Price / Treasury Rate Information Customer Metric Projections Free Cash Flow Projections Wtd Avg Cost of Capital Projections Operating Asset Value Non Operating Assets Customer Asset Values Shareholder Value Net Debt 13
Customer-Based DCF Modeling Process We estimate shareholder value via the following process: Date Customers Sales Free Cash Flows PV(FCF s) NPV(FCF s) We must model/forecast future customer base Revenue/User Q1 2013 Q2 2013 Q3 2013 $100 $200 $350 Profitability, Balance Sheet Effects Weighted Average Cost of Capital $50 $100 $175 $49 $48 $46 $143 E(Stock Price) 20 Non Operating Assets, Debt, Shares Outstanding Modeling the Customer Base We model customers acquired in cohorts, who churn over time: Total customer base: sum of active customers across cohorts Acquisitions over time (Modeled Statistically) Churn by cohort over time (Modeled Statistically) 14
Illustrative Application: Dish Network Incorporated in 1995 Customer data available beginning Q1 2001, through Q1 2015 Satellite pay-tv provider: The Customer Metric Data 15
Estimated Ending Customers, Adds and Losses Great Recession increased losses, decreased adds. Seasonality: Lower Q1 adds and losses Higher Q3 adds and losses Secular decline Validation vs Wall Street Analyst Projections 3,454 quarterly sales predictions by analysts Predictions made 5/2001 12/2014 Varying time horizons How does our predictive accuracy compare? Proposed Model Operationalization: E(Sales(t)) = E(Avg # Custs(t))*E(ARPU(t)) E(ARPU(t)): ARIMA(0,1,0) + Time Trend Example: Analyst Prediction: $2,120M in Q3 05 Our Prediction: $1,988M in Q3 05 2 quarters ahead Actual Sales: $1,989M in Q3 05 5/5/2005: Q1 2005 Results Reported 7/28/2005: Analyst Forecast Made 8/9/2005: Q2 2005 Results Reported 11/8/2005: Q3 2005 Results Reported 16
Validation vs Wall Street Analyst Projections Validation vs Wall Street Analyst Projections RMSE and MAE off Analyst Average Forecasts *: (Source: Zack s, Company Financials) *Actual equipment sales (4% of sales) netted out of all predictions 17
Projecting Revenues Total Customers Projecting Revenues Multiplied by ARPU E(Monthly Revenue) / E(Average Customers) 18
Projecting Revenues Multiplied by ARPU Forecast continuation of trend Projecting Revenues Equals Revenues: 19
Valuation Results Careful accounting for costs. Valuation Results then adjust for other assets and convert to per share: Estimate from Model 20
The Unit Economics of Recent Robbie Summary financial figures for an about tobe acquired customer: Distribution of CLV Probability of loss: 31% Probability of return < WACC: 36% Correct return benchmark? (Koller, Goedhart, Wessels) The Value Relevance of Customer Tenure Survival probability for Recent Robbie 21
The Value Relevance of Customer Tenure versus Longtime Larry: > 50% higher expected residual lifetime! The Value Relevance of Customer Tenure 22
The Value Relevance of Customer Tenure Summary and Conclusions We don t need internal customer data to fit good CLV models CLV can be estimated from inside and out! Our models work well in both contractual and non-contractual settings A thorough customer-base analysis requires careful consideration of the zero class CLV analyses can be used to more precisely estimate company valuation Proposed customer-based DCF model yields important insights into new customer profitability and the importance of customer tenure 23
Questions? Professor Peter Fader faderp@wharton.upenn.edu www.petefader.com Twitter: @faderp Customer Centricity: Focus on the Right Customers for Strategic Advantage http://bit.ly/fadercc MBA for EXECUTIVES REUNION 2015 MBA for EXECUTIVES REUNION 2015 WELCOME BACK ALUMNI! 24