RFM — Reference

The RFM (Recency, Frequency, Monetary) model segments customers by transactional behaviour:

• Recency — how long since the last transaction.
• Frequency — how many transactions in the observed period.
• Monetary — how much the customer has spent.

RFM is a descriptive model, commonly used for segmentation, churn analysis, and targeting.

Input data: see AI Model Data Requirements → RFM.

---

Key concepts

• Transactions as events — each row of the event table represents a transaction (or equivalent business event).
• Allowed statuses — optionally restrict to valid statuses (e.g. "Closed Won", "Completed").
• Output — one row per user with R, F, and M scores plus a segment label, linked to bpp_user_id.

---

JSON configuration reference

Field	Type	Description	Example / best practice
data_src.region	STRING	Cloud region (auto-populated).	"europe-west8"
data_src.project_id	STRING	GCP project ID (auto-populated).	"audience-ai-dev"
data_src.event_table_id	STRING	Event table with transactions (a *_bpp table).	"deal_status_change_bpp"
data_src.column_price	STRING	Column with the monetary value per transaction.	"Deal_Amount"
data_src.column_status	STRING	Column with the transaction/deal status.	"Stage"
data_src.allowed_statuses	ARRAY	Status values considered valid transactions.	["Closed (Won)"]
data_src.column_timestamp	STRING	Transaction timestamp column (UTC).	"event_date"
data_src.sample_interval	INT	Number of days in the analysis window.	365

---

Example configuration

{
  "data_src": {
    "region": "europe-west8",
    "project_id": "example-project",
    "column_price": "Deal_Amount",
    "column_status": "Stage",
    "event_table_id": "deal_status_change_bpp",
    "sample_interval": 365,
    "allowed_statuses": ["Closed (Won)"],
    "column_timestamp": "event_date"
  }
}

---

Output structure

Column	Type	Description
bpp_user_id	STRING	User identifier linked via identity resolution
recency_score	INT	Relative recency score (e.g. 1–5)
frequency_score	INT	Relative frequency score (e.g. 1–5)
monetary_score	INT	Relative monetary score (e.g. 1–5)
rfm_cluster	STRING	Segment label (e.g. Champions, At Risk, Hibernating)

---

Business rules by industry

E-commerce — column_price = order value; allowed statuses "Completed", "Paid"; interval 180–365 days.

B2B / CRM — column_price = deal amount; allowed statuses "Closed (Won)"; extend the interval for long sales cycles (e.g. 730 days).

Subscription / SaaS — treat renewals as transactions; allowed statuses "Renewed"; recency helps flag churn risk.

Retail with seasonality — interval should cover at least one full seasonal cycle (e.g. 365 days); allowed statuses "Completed".

---

Best practices

• Ensure timestamps are in UTC.
• Include only transactions that reflect actual revenue (exclude cancelled/refunded).
• Choose the sample interval by purchase cycle: fast-moving goods 90–180 days, durable goods 365–730 days.
• Use segments to prioritise actions: Champions → loyalty; At Risk → win-back; Hibernating → awareness.