Calculating Your Customer Lifetime Value - Dawn of the Data Age Lecture Series
- 1. Dawn of the Data Age Lecture Series Interpreting Data Like a Pro
- 2. Hi. I’m Luciano Pesci… Co-Founder & CEO, EMPERITAS ● Team of economists and data scientists delivering bi-weekly Customer Lifetime Value intelligence so our clients can beat their competitors for the most profitable customers. Founder & Director, Utah Community Research Group, Univ. of Utah ● Teach microeconomics, data science, applied research, & American economic history. 2
- 3. Today’s Lecture Outline ● Teach you how to use your data to find average CLV. ● Explain best practices for including non-monetary CLV. ● Move you beyond average CLV to individualized CLV. 3
- 5. Customer (Lifetime?) Value ● There’s a big difference between Customer Value (CV) & Customer Lifetime Value (CLV). ● Lifetime Value requires a clear birth date & expected death date (churn) of the customer. ○ This can be learned from your historical data. ○ Requires you predict into the future. 5
- 6. Revenue or Profit? ● Customer Value and Customer Lifetime Value can be based on either revenue or profit. ● Profit can be a better metric. ○ It’s a “net metric” because it includes costs related to realizing the revenue (like cost of acquisition). ○ If there’s volatility in profit margins over time, revenue can be a better metric. 6
- 7. Quick & Dirty Customer Value ● The fastest way to CV is to take annual revenue (or gross margin) and divide it by the number of customers for each year. ● This is a starting estimate, but it’s very limited. ○ Doesn’t take lifetime into consideration. ○ Doesn’t address the Pareto Principle. 7
- 8. Survey Based Estimation ● If historical data isn’t available, or IT stonewalls you, use customer surveys. ● The key components to include are: ○ Purchase patterns (items, value, frequency). ○ Loyalty (likelihood to retain). ○ Recommendation (net promoter score). 8
- 10. Power of Word of Mouth ● The single most important non-monetary contribution to CLV is personal recommendation. ● Requires you track recommendations that lead to acquiring new customers, and attribute a portion of that value back to the original customer who recommended. 10
- 11. Social Media Engagement ● Digital word of mouth through social media engagement is a good proxy for recommendations. ● This data adds a rich layer of insight to your calculations, but it requires you assign values to outcomes (and track them). 11
- 12. Willingness to Pay Even More ● App or platform usage behavior can be a great insight into customers who derive higher value from your product or service. ● Economists call this consumer surplus because it’s value they derive above and beyond the price they pay. ○ You have to figure out how to capture this. 12
- 13. 13INDIVIDUAL CLV
- 14. Making Magic Happen ● The real magic with CLV happens when you understand value for individual customers. ● Allows for richer segmentation, persona development, and predicted values that can be tested/validated over time. 14
- 15. Core Components of Individual CLV ● Recency - When did they last purchase? ● Frequency - How often do they purchase? ● Monetary Value - What’s the value of purchases? ● Non-Monetary Value - Do they recommend and/or have a high consumer surplus? ○ Since this makes them less likely to churn. 15Source: cran.r-project.org/web/packages/didrooRFM/index.html
- 16. 90-Day Sprints ● Once you have usable data in hand, it shouldn’t take more than 90-days to get to your first model. ○ This can be complicated by data prep/cleaning/transforming. ○ Assumes you have working knowledge of your industry. ● This should NOT be considered your final model, it’s a starting benchmark on which to improve. 16
- 17. Buy Till You Die ● Incorporates Recency and Frequency to build a predictive model for churn. ○ Uses daily purchase patterns to predict if they will buy in a future “target period.” ● This information gives you a more accurate value for the target time period by using expected value (probability) of purchase. 17Source: cran.r-project.org/web/packages/BTYD/vignettes/BTYD-walkthrough.pdf
- 18. Predict, Validate, Repeat ● Since customer lifetime value depends on the underlying preferences of the individual, it can be very volatile. ● The only way to solve this is to continually test, validate, and update your model. ○ Frequency of purchase should determine how long you run tests & how often you retrain the model. 18
- 19. Groupon Case Study Groupon used a two stage random forest model for each predefined "purchase cohort." The first stage predicted whether or not an individual will purchase in the target time window. The second stage predicted a dollar value for individuals predicted to purchase in the target time window. These predictions were updated daily and the whole model was retrained quarterly. 19Source: An Engagement-Based Customer Lifetime Value System for E-Commerce
- 20. ASOS Case Study ASOS trained a random forest to predict whether an individual will purchase in the target time window and a random forest to predict a percentile of value. It calibrated results from these models by using a logistic regression for the churn prediction and a decision tree for the value prediction. This process was repeated daily. They also experimented with deep neural nets to learn unsupervised embeddings (based on website behavior) as an added feature set for the random forest to improve the churn predictions. 20Source: Customer Lifetime Value Prediction Using Embeddings
- 21. JOIN US FOR THE NEXT LECTURE Step Up Your Survey Research, Thursday December 14th 2017 emperitas.com/lecture