Each project arrives with its purpose, milestones and first tasks written out. Here is how the first three read.
Written brief for what lifetime value will decide
Lifetime value numbers get built and then ignored when nobody agreed which decision they were meant to inform. Writing down the two or three decisions CLV will feed, such as the most you can pay to win a customer in each channel or how much a retention offer may cost, fixes the level of detail, the horizon and the audience before any modelling starts.
First tasks
- List the decisions where customer value is currently guessed
- Ask each decision owner what number would change their choice
- Write the one-page brief with decisions, owners and level of detail
- Get the brief agreed by the main decision owner
Order and customer data extract for value analysis
Customer value is built from four facts: who bought, when, how much, and what was later refunded. Finding where each lives, whether the ecommerce platform, the billing system, the payment processor or the accounting package, and pulling a clean extract with at least two years of history is the step most first attempts skip.
First tasks
- List every system that records orders, subscriptions or refunds
- Export customer ID, order date, value and refunds for all available history
- Remove test orders, staff purchases and internal accounts
- Compare the extract's yearly revenue with the finance totals
- Write a short note on gaps and migrations in the data
Customer key for lifetime value calculations
If one person appears as three customers because they checked out as a guest, changed email and later opened an account, their value is split three ways and CLV looks lower and shorter than it is. Choosing a single key for value work, with a documented rule for linking guest orders and duplicate accounts, matters more than any model choice.
First tasks
- Count orders with no customer ID or with a guest checkout flag
- Choose the identifier and matching rule for value work
- Apply the rule and measure how many duplicate customers merge
- Record the rule and its effect in the CLV method note