Customer Lifetime Value TFMA 2014 David Lockwood Direct

Customer Lifetime Value TFM&A 2014 David Lockwood: Direct Wines Terry Hogan: Golden Orb Customer Lifetime Value – Direct Wines

How Direct Wines use customer lifetime value Discover profitable/unprofitable activities, sources, mechanics Previous curve-based approach – Problems with LTV changing from week to week – Purchase of a few additional cases had substantial impact Needed to extend model to other markets Customer Lifetime Value – Direct Wines

Customer Lifetime Value: What is it and why does it matter? New customers are acquired at an average cost per recruit (CPR) Different campaigns generate customers at a different cost per recruit – Drivers: response rate, media, message costs etc. – Relatively simple to measure –Easy to fall into the trap of focusing too much on CPR Different cohorts of recruits will have different values to the business over their lifetimes (LTV) –Drivers: proportion who never buy again, number of orders, average order value (AOV), attrition rate What matters is the difference between CPR and LTV – Added value per recruit Customer Lifetime Value – Direct Wines

Meaning: Cost per recruit What level of ‘cost’ to include? Direct marketing costs (MCPR) Discounts? Extra free? Multibuys? Free shipping? Direct Wines use ‘Full cost per recruit’ (FCPR) – net contribution loss divided by recruits Customer Lifetime Value – Direct Wines

Meaning: Lifetime Value ‘Value’ – Net contribution ‘Lifetime’ – how long? Simple approach – Look at first x years e. g. 3 years – Simple calculation – easily understood Advanced approach – Consider the full lifetime of the customer – Could be 20 years or more – Essential to discount future cash flows Customer Lifetime Value – Direct Wines

Calculation of Lifetime value • What level? Individual • Data intensive • Time-consuming • Use sample data to build models Response code • Not too dataintensive Campaign • Quick • Distinguishes lists, tests • May need to split by time • Good reference point • Lose detail of tests, sources More detail Less detail • Technique – Curves – sensitive to ‘lumpy’ purchases – Time-series – difficult with infrequent purchases – Regression – most powerful/appropriate in this case Customer Lifetime Value – Direct Wines

Regression: a bit of theory Equation: Y = k + ax (+ a 2 x 2 +a 3 x 3. . . ) Aims to minimise the square of the residuals R 2: measure of goodness of fit – % variance explained by line Standard error of estimate: measure of the absolute size of residuals Customer Lifetime Value – Direct Wines

Regression model: predict what? Need to decide a time-frame – let us take 3 years for now What are we trying to predict? – Total revenue? – Revenue per recruit? Regressing total revenue – typically high R 2 – Dominated by cohort size – Doesn’t distinguish well between different campaigns Better to predict revenue per recruit Customer Lifetime Value – Direct Wines

Regression model: Level of detail? Total or incremental revenue per recruit? – Either is OK, but total revenue includes one of the numbers we want to use as a predictor (revenue to date) – Better to make it harder by predicting incremental revenue • Lower R 2 but same standard error By line of business? Category? Product? – Direct Wines have both standard and continuity businesses – We found it more accurate to predict total revenue – Errors tend to be greater if predicting standard and continuity separately Customer Lifetime Value – Direct Wines

Building the model Assemble lots of likely measures about historic campaigns – Need to be on the same basis as the predicted variable • • • Total revenue to date Continuity revenue Gross/Net orders Deseasonalised, last 6 months’ revenue Dummy variables (1/0) to represent categorical variables Point your preferred statistics tool at these numbers – stepwise regression – Beware! Increasing number of variables always increases R 2 without improving model – Many will be highly correlated • individually quite predictive, but don’t add much • A good statistics package will identify the significant factors • Include the measures that significantly improve the R 2 Customer Lifetime Value – Direct Wines

Spread the sales over time using a curve Not essential, but can aid interpretation/understanding - Use remaining portion of curve Curves can be different for different revenue streams Best to do it before calculating contribution so that we can treat past times differently Use the standard error of the model to estimate a confidence interval around the numbers Customer Lifetime Value – Direct Wines

Calculate net contribution from projected revenue Depends on what data is available – Actual margin for past sales – Most costs can be calculated as standard percentage of revenue for simplicity • Within margin of error of predictions – Additionally/alternatively £ per original recruit for marketing costs • E. g. for the first 6 months/ 1 year Customer Lifetime Value – Direct Wines

When can we start to build a reliable model? From about 6 months we can account for half of natural variation We need to build a series of models for different time periods Early on, recruitment details feature. Later, recent sales are more relevant Customer Lifetime Value – Direct Wines

Over time, the results become better These are real people with real lives – Move away – Change in circumstances – It is only a model Bigger campaigns are more predictable – Random variation tends to cancel out After about 6 months we can generally make pretty good predictions – Different in different markets Customer Lifetime Value – Direct Wines

Thank You www. golden-orb. ltd. uk @Golden. Orb. Ltd Customer Lifetime Value – Direct Wines