DIVISIO / Blog
Clustering

Churn was creeping up. The RFM Wizard found the customers worth saving.

By Nikita

The setup

Thread & Loom ships a monthly box of curated basics. Elena owns retention. Every cancel is her problem, every win-back is her win. The business looked healthy on the surface, but the trend line under it didn’t: monthly churn had drifted from 4.1% to 5.6% over two quarters. Small numbers, ugly compounding.

The problem

Elena knew the retention playbook cold. The framework she wanted was RFM (Recency, Frequency, Monetary), the bread-and-butter way to score who’s slipping and who’s gold. The trouble was building it. Her data lived in a raw orders export: one row per order, an order date, an order total, a customer ID. Turning that into a clean per-customer RFM model usually meant a favor from analytics and a two-week wait she didn’t have.

“I didn’t need a data science project,” she said. “I needed RFM segments by Friday.”

The turning point

She opened Divisio’s RFM Wizard. Instead of a blank clustering screen, it walked her through the three questions that matter and mapped her columns to them:

  • Recency → her last_order_date. The wizard noticed it was a date and derived the days-since-last-order variable for her, so she didn’t have to compute it by hand.
  • Frequency → total_orders.
  • Monetary → lifetime_value.

It also caught something Elena would’ve missed: her file looked transaction-level in places, so it flagged duplicate customer IDs and warned her before the numbers came out wrong. She pointed it at the deduped per-customer view, and it applied sensible defaults (four segments, scaling on) and ran it.

Out came the four segments every retention marketer is hunting for. And one of them had her name on it: Lapsing Loyalists, high lifetime value, solid past frequency, but recency sliding into the danger zone. Customers who used to be the best and were quietly on their way out. Exactly the people a win-back should be spent on, and exactly the people her one-size blast was failing.

How she did it

  1. Dropped the orders export into Divisio.
  2. Opened the RFM Wizard and mapped Recency / Frequency / Monetary to her columns.
  3. Let the wizard derive days-since-last-order from the date and heeded its duplicate-ID warning.
  4. Accepted the defaults (four segments, scaling on) and read the resulting groups.
  5. Exported the segment definitions as business rules and rebuilt them as a Klaviyo segment to trigger a win-back flow.

The payoff

Elena built a three-email win-back flow in Klaviyo aimed only at the Lapsing Loyalists: a “we miss your style” note, a personal restock nudge, and a small returning-member perk. Because it was pointed at high-value customers who were drifting, not the whole list, the economics worked the first month.

The flow recovered roughly 18% of the lapsing group before they cancelled, and pulled overall monthly churn back under 5%. The framework she’d always trusted finally took an afternoon to build instead of a sprint to request.

“RFM isn’t new. Being able to build it myself, correctly, before lunch, that was new. The wizard even derived recency from a date column and stopped me from running it on dirty data.”

Elena, retention manager at Thread & Loom

Feature spotlight: RFM Wizard (Pro)

A guided Recency / Frequency / Monetary template that meets marketers where they think. Map three columns, and Divisio handles the rest. It derives a days-since variable when recency is a date, warns you if your data looks transaction-level instead of per-customer, and applies sensible defaults so you get trustworthy RFM segments without building the model by hand.