how-to-rfmlisted

# how-to-RFM-using-claude-python Decision-first RFM: the output is a ranked budget recommendation - which segments to protect and which to win back, each with a dollar figure - not a table of scores. Showcase walkthrough (Everrest retail case, real executed charts): https://github.com/phoebefu6/phoebe-data-skills - `docs/how-to-rfm-using-claude-python/` ## The 6 steps ### 1. Input Collect the transaction tables (customers, orders, order_items or equivalent) and the business context - specifically the decision on the table (retention budget, win-back campaign, VIP program). RFM needs only: a customer key, an order timestamp, and an order value. Confirm the grain and pick the analysis reference date ("today"). ### 2. Sample data (only when real data isn't available yet) Write a seeded generator that draws each customer from a lifecycle archetype (champion, loyal, at-risk, can't-lose-them, hibernating, one-and-done...) so the segmentation has structure to recover. Plant patterns on purpose - whale concentration, high-value lapsers, a one-and-done acquisition channel, a promo cohort that never reactivated, a returns-heavy group - and keep a ground-truth label column to validate against. With real data, skip this. ### 3. Objective Frame ONE budget decision ("where should the CRM team spend Q3 retention budget, and who is quietly most valuable?") plus 3-5 sub-questions (value concentration, lapsing high-value customers, low-value channels, segment sizes, best next dollar). Ev