RFM Segmenter

SKILL.md

---
name: RFM Segmenter
description: Segment your customer base using Recency, Frequency, and Monetary value scoring to identify Champions, Loyal Customers, At-Risk buyers, and Churned segments — then generate targeted retention and reactivation campaigns for each group.
---

# RFM Segmenter

Most ecommerce stores treat all customers the same: one email blast, one discount, one message. RFM segmentation breaks your customer base into behaviorally distinct groups so you can send the right message to the right person at the right time. This skill takes raw order data, scores each customer on Recency (how recently they bought), Frequency (how often), and Monetary value (how much they've spent), assigns them to named segments, and recommends specific marketing actions for each group.

## Quick Reference

| Decision | Strong | Acceptable | Weak |
|----------|--------|------------|------|
| Data window | Rolling 12–24 months | Last 12 months fixed | All-time (biases toward old customers) |
| Scoring method | Quintile-based (1–5 per dimension) | Quartile-based (1–4) | Binary high/low |
| Segment naming | Named behavioral segments (Champions, etc.) | Numbered clusters | R/F/M score only |
| Monetary metric | Net revenue (post-refund) | Gross revenue | Order count only |
| Recency baseline | Category purchase cycle-adjusted | Fixed 30/60/90/180 days | No adjustment |
| Action specificity | Segment-specific campaigns and offers | Generic retention vs. win-back | Mass email to all |

## Solves

1. **Indiscriminate discounting** — Giving a 20% coupon to Champions who would have bought anyway destroys margin. RFM lets you reserve discounts for At-Risk and Lapsed segments where they're actually needed.
2. **Churn blindness** — Without recency scoring, you don't know a customer is slipping away until they're gone. RFM flags customers before they churn.
3. **Misallocated retention spend** — Marketing budgets spent on customers who already churn (Hibernating segment) have near-zero ROI. Redirect that spend to Promising and Needs Attention segments.
4. **Over-emailing engaged customers** — Sending 5 emails a week to Champions can flip them from loyal to annoyed. RFM helps you calibrate frequency by segment.
5. **Reactivation guesswork** — Win-back campaigns work best with personalized angles. RFM tells you which churned customers were once high-value and worth the spend.
6. **LTV forecasting** — Champions and Loyal Customers are your LTV anchors. Knowing their size helps you set realistic growth targets.
7. **Loyalty program design** — RFM reveals which behaviors (frequency vs. spend) actually predict long-term customer value in your category.

## Workflow

### Step 1 — Export and Clean Order Data
Pull a customer-level order export with: Customer ID (or email), Order Date, and Order Value (net of refunds). Set your analysis window (recommended: rolling 24 months for stable categories, 12 months for fashion/seasonal). Remove test orders, internal orders, and wholesale accounts.

### Step 2 — Calculate R, F, M Raw Values
For each customer:
- **Recency (R)**: Days since their most recent order (relative to today or your snapshot date)
- **Frequency (F)**: Total number of orders in the analysis window
- **Monetary (M)**: Total net revenue from orders in the analysis window

### Step 3 — Assign R, F, M Scores (1–5)
Divide your customer base into quintiles for each dimension. Score 5 = best, 1 = worst.

- **R score**: Most recent buyers get 5; oldest last-purchase dates get 1
- **F score**: Highest order count gets 5; single purchase gets 1 (or lowest frequency)
- **M score**: Highest spenders get 5; lowest spenders get 1

*If you have fewer than 500 customers, use quartiles (1–4) to avoid artificially small quintile buckets.*

### Step 4 — Combine Scores and Assign Segments

Use the following segment mapping (adjustable by business):

| Segment | R Score | F Score | M Score | Description |
|---------|---------|---------|---------|-------------|
| Champions | 4–5 | 4–5 | 4–5 | Bought recently, buy often, spend the most |
| Loyal Customers | 3–5 | 3–5 | 3–5 | Regular buyers with above-average spend |
| Potential Loyalists | 4–5 | 2–3 | 2–3 | Recent buyers, could become loyal with nurturing |
| New Customers | 4–5 | 1 | any | Just made their first or second purchase |
| Promising | 3–4 | 1–2 | 1–2 | Recent buyers, light frequency and spend |
| Needs Attention | 2–3 | 2–3 | 2–3 | Above-average overall but recency declining |
| About to Sleep | 2–3 | 1–2 | 1–2 | Recency dropping, below-average engagement |
| At Risk | 1–2 | 3–5 | 3–5 | Were Champions or Loyal but haven't returned |
| Can't Lose Them | 1 | 4–5 | 4–5 | Haven't bought in a long time, were high-value |
| Hibernating | 1–2 | 1–2 | 1–2 | Low across all dimensions, minimal engagement |
| Lost | 1 | 1–2 | 1–2 | Haven't purchased in longest time, low value |

### Step 5 — Validate Segment Sizes
Healthy segment distributions for a mature ecommerce store:
- Champions: 5–15%
- Loyal Customers: 10–20%
- At Risk + Can't Lose Them: 10–20% (your most urgent attention)
- New + Potential Loyalists: 15–25% (your growth pipeline)
- Hibernating + Lost: 20–40% (normal; focus is on preventing others from getting here)

If Champions exceed 30%, your scoring window is too narrow. If Lost exceeds 60%, customer acquisition is outpacing retention.

### Step 6 — Build Segment-Specific Campaign Plans
Map each segment to a specific message, offer, channel, and cadence. See the output template for the full campaign matrix.

### Step 7 — Set Review Cadence
Re-run the RFM analysis every 30–90 days (shorter for high-frequency categories like consumables, longer for furniture or appliances). Track segment migration: customers moving from Needs Attention to At Risk is a warning signal; customers moving from Promising to Loyal is a success signal.

## Examples

### Example 1 — Fashion Apparel Brand (2,400 active customers)

**Input data summary:**
- Analysis window: 12 months
- Average order frequency: 2.3 orders/year
- Average order value: $87
- Average customer LTV (12-month): $201

**RFM Score Distribution:**
```
Champions (R4-5, F4-5, M4-5): 187 customers (7.8%)
Loyal Customers (R3-5, F3-5, M3-5): 312 customers (13.0%)
At Risk (R1-2, F3-5, M3-5): 203 customers (8.5%)
Can't Lose Them (R1, F4-5, M4-5): 89 customers (3.7%)
New Customers (R4-5, F1): 445 customers (18.5%)
Potential Loyalists (R4-5, F2-3, M2-3): 276 customers (11.5%)
Needs Attention (R2-3, F2-3, M2-3): 334 customers (13.9%)
About to Sleep (R2-3, F1-2, M1-2): 201 customers (8.4%)
Hibernating (R1-2, F1-2, M1-2): 298 customers (12.4%)
Lost (R1, F1-2, M1-2): 55 customers (2.3%)
```

**Key Actions:**
- **Champions (187)**: Early access to new collection + personal stylist offer. No discount. Target: increase AOV.
- **At Risk (203)**: Win-back email sequence — "We miss you" → 10% offer → last chance. Target: reactivate 25%.
- **Can't Lose Them (89)**: Phone/SMS outreach. VIP reactivation offer (20% + free shipping). These were top spenders.
- **New Customers (445)**: Onboarding sequence focused on second purchase (most impactful moment in fashion LTV).

**Projected Revenue Impact:**
- At Risk reactivation (25% of 203 × $87 AOV): ~$4,400 recovered revenue
- Champion upgrade campaigns (AOV increase of $30 for 40% of 187): ~$2,244 additional revenue
- New Customer second purchase (15% conversion of 445 × $87): ~$5,808 additional revenue

---

### Example 2 — Consumable Brand (Coffee Subscription + One-Time Buyers)

**Challenge:** High purchase frequency means standard RFM windows miss the nuance of subscription vs. occasional buyers.

**Solution:** Separate the analysis for subscribers (recency = days since last subscription renewal) and non-subscribers (standard RFM), then merge segments with adjusted scoring.

**Adjusted recency buckets for coffee (avg. reorder cycle = 28 days):**
- R5: Purchased within 21 days
- R4: 22–42 days
- R3: 43–70 days
- R2: 71–120 days
- R1: 120+ days

**Finding:** 340 R1 customers with F4-5 (frequent buyers who've lapsed) = 14% of customer base. This "At Risk Power Buyer" sub-segment exists in most consumable brands but gets missed with standard quarterly RFM windows.

**Action:** Target these 340 with a personalized reactivation campaign: "Your usual order is overdue" + 15% one-time reactivation discount. Expected reactivation rate: 35–45%.

---

## Common Mistakes

1. **Using all-time data instead of a rolling window.** A customer who bought 50 times in 2019 but nothing since 2022 looks like a Champion in all-time data but is Lost in any meaningful current sense. Always use a rolling window.

2. **Not adjusting for purchase cycle.** A 90-day recency cutoff is fine for a general retailer but meaningless for a daily supplement brand where lapsing after 45 days is already a warning sign.

3. **Treating Monetary as gross revenue.** High-return customers can have high gross revenue but negative net revenue. Use post-refund net revenue for M scoring.

4. **Building 11 segments when you only have 200 customers.** With small datasets, collapse to 5–6 named segments. Micro-segments with 3 customers can't be acted on.

5. **Forgetting to exclude wholesale and B2B accounts.** A single wholesale account can dominate your F and M quintiles and skew the entire scoring model.

6. **Running RFM once and never updating it.** Customer behavior shifts over time. Segment scores should be refreshed at least quarterly; monthly is better for fast-moving categories.

7. **Applying identical discounting to all non-Champion segments.** At Risk and Can't Lose Them need discounts; New Customers need education and social proof; Promising customers need curated recommendations. One offer doesn't fit all.

8. **Ignoring segment migration trends.** If 20% of last quarter's Loyal Customers are now in Needs Attention, you have a systemic retention problem that won't be solved by a single campaign.

9. **Not tracking RFM campaign performance by segment.** You can't improve what you don't measure. Always tag campaigns with the RFM segment they targeted and report conversion by segment.

10. **Using RFM as the only segmentation layer.** RFM tells you what customers do, not why. Combine with product category data or survey responses to build fuller personas.

## Resources

- [Output Template](references/output-template.md) — RFM segment report and campaign matrix
- [Segment Action Playbook](references/segment-action-playbook.md) — Campaign strategies per segment
- [Scoring Methodology Guide](references/scoring-methodology.md) — Quintile calculation and edge cases
- [Quality Checklist](assets/checklist.md) — Pre-deployment accuracy checklist