How to Read RFM Scores 111, 114, and 441 Without Reversing Their Meaning

A Tale of Three Customers in One Thai Online Shop

Imagine a fictional Thai fashion accessories shop called Sai Fah that has been running for two years and has around 3,000 customers in CasperBrain. The shop's owner, Khun Rattana, exports the customer list one morning and notices three profiles sitting near the top of her attention queue. The first shows a score of 111. The second shows 114. The third shows 441. She pauses. Are these all bad because the digits are low? Is 441 somehow strong because the first two digits are high? Should she send win-back discount vouchers to all three?

The answer to every one of those questions depends entirely on which direction the scoring scale runs — and getting that direction wrong is the single most consequential misread a retention analyst can make. This guide explains CasperBrain's convention from the ground up, walks through all three profiles with explicit fictional data, and ends with a checklist you can apply before every campaign decision.

Why Score Direction Must Be Settled Before Any Other Step

RFM conventions differ across tools. Before comparing reports, check the number of score levels, the direction of each dimension and the meaning of the displayed value. Microsoft’s Dynamics 365 Commerce documentation describes its own configurable RFM setup; it is not documentation for CasperBrain. The Shopify guides explain RFM concepts in their own context. Use these references for the general method, then confirm the convention of the product you actually use rather than copying its numbers to another platform.

CasperBrain uses four score levels: 1 is strongest and 4 is weakest on each dimension. For recency, score 1 indicates the strongest relative recency position; score 4 indicates the weakest. That relative position does not establish a universal number of days or a future outcome. State the direction before interpreting the code, and keep it with any report shared with the team.

The risk is concrete. Someone who learned a higher-is-better scale could misread 441 as strong recency and frequency. In CasperBrain, the first two positions instead indicate the weaker recency and frequency quartiles; the last indicates the strongest monetary quartile. The score alone does not prove churn or tell you why buying slowed. In the fictional 310-day, one-order example below, the recorded history gives additional context for a careful follow-up test.

Sources: Microsoft Docs: Set up Recency, Frequency, and Monetary (RFM) analysis in Dynamics 365 Commerce · Shopify: RFM Analysis — Definition and Segmentation Guide (2026)

Anatomy of a CasperBrain Score String

Read a CasperBrain composite as three characters in the order Recency, Frequency, Monetary. R concerns how recently purchases were recorded; F concerns purchase frequency; M concerns recorded spending. In this guide’s fictional examples, we explicitly define frequency as distinct orders and monetary as total purchase value in a declared 12-month window. Those teaching definitions do not establish every product report’s counting, exclusions, ties or time-window behavior. Check the current report definitions and underlying records when applying the method.

The string 114 means R1, F1 and M4: three positions about three different dimensions. Do not interpret it as the integer one hundred and fourteen. Codes 141 and 411 both sum to six, but 141 has stronger relative recency and weaker relative frequency, while 411 has weaker relative recency and stronger relative frequency; both have M1. Those relative positions call for different questions about the purchase history, not an automatic business verdict from the sum.

Keep the composite as a three-character code when interpreting it or passing it to another tool. Alphabetic or numeric sorting can arrange the code values, but neither operation establishes an overall ranking of customer strength or value. Inspect the R, F and M positions separately, read them using the stated score direction, and include the underlying purchase facts in the decision.

Sources: Shopify: RFM Analysis — Definition and Segmentation Guide (2026)

The Four-Quartile Scale: What Each Position Means

For interpretation, use the documented CasperBrain direction: score 1 is strongest and score 4 weakest; score 2 and score 3 are the second and third quartile positions. The digit does not itself tell you a fixed day limit, order count or spending threshold. It also does not establish which records were eligible for scoring or how tied values were handled. Treat the assigned quartile as the starting point for reading the underlying purchase data, rather than inventing those implementation details from the label.

Because boundaries are relative, a customer's score can change even if their raw behavior stays exactly the same. If a large group of previously inactive customers all make a purchase in the same week, the recency distribution shifts, and some customers who were in score 1 may slide to score 2 simply because others have caught up. This is a fundamental property of quartile-based ranking, not a bug. It means you should always inspect the underlying values — actual days since last purchase, actual order count, actual spend — before treating a digit change as evidence that a customer's behavior has changed.

The strongest possible composite is '111': top quartile on all three dimensions simultaneously. The weakest possible composite is '444': bottom quartile on all three. Between those extremes, mixed scores like '114' or '441' tell a story about where a customer is strong and where they are not — and that mixed story is exactly what makes the score actionable.

Fictional Profile A — Score '111': Nong Fai, the Power Buyer

The following profile is entirely fictional and is constructed to teach score interpretation. All names, order IDs, dates, and amounts are invented. Do not treat any figure here as a research result or a product screenshot.

Here is a complete fictional input summary for Nong Fai. The 12-month example has exactly 18 distinct orders, with IDs A01 through A18. Orders A01 through A17 each have a purchase value of ฿2,000; A18 has ฿8,000. Their ages in days are 3, 23, 43, 63, 83, 103, 123, 143, 163, 183, 203, 223, 243, 263, 283, 303, 323 and 343 respectively, measured from one shared calculation date. The total is 17 × ฿2,000 + ฿8,000 = ฿42,000, and the newest order is 3 days old. For this teaching example, the wider fictional population is assumed to assign R1, F1 and M1. These assignments are given inputs, not quartiles calculated from only the three profiles in this article.

A 111 score makes relationship care worth considering, but it does not guarantee future purchases, immunity from churn or the right offer. For this fictional customer, early access, a thank-you note or a relevant recommendation are possible tests. Before offering a discount, ask whether it solves a real customer need and whether the margin can support it. Record the reason for the action and review subsequent observed purchases without assuming the score or the contact caused them.

Fictional Profile B — Score '114': Khun Sombat, the Frequent Low-Spender

This profile is fictional. Khun Sombat has exactly 12 distinct orders in the same 12-month teaching window, with IDs B01 through B12. B01 through B11 each have a value of ฿60, while B12 has ฿140. Their ages in days are 5, 35, 65, 95, 125, 155, 185, 215, 245, 275, 305 and 335 respectively. Total purchase value is 11 × ฿60 + ฿140 = ฿800. The newest order is 5 days old. As a teaching assumption about the wider fictional population, his assigned quartiles are R1, F1 and M4.

Read the assigned positions separately: R1, F1 and M4. In this fictional population, the recorded 5-day recency and 12 distinct orders are assigned the strongest recency and frequency quartiles, while total purchase value of ฿800 in the same 12-month window is assigned the weakest monetary quartile. Monetary compares total value in the declared window, not per-order spend. The small average order value discussed next is a separate calculation from the fully stated example data.

Do not read M4 as strong monetary value. In this particular worked example, 12 purchases totaling ฿800 imply average purchase value of about ฿66.67, so a carefully chosen bundle or relevant additional item is a possible test. That small average comes from the explicit order data, not from M4 alone. Purchase frequency is not website visit frequency, and the score alone does not establish loyalty or future intent. Check product fit, customer preference and margin before choosing an offer; do not label this recent buyer lost simply because monetary rank is weak.

Fictional Profile C — Score '441': Khun Malee, the Lapsed Big Spender

This profile is entirely fictional. Khun Malee made a single purchase at Sai Fah: Order ORD-0007 placed 310 days ago for ฿55,000 — a high-value home decor accessory set purchased as a gift. That is the only order in the 12-month analysis window. Days since last order: 310.

In the wider fictional population stipulated for this lesson, 310 days since the last order is assigned R4, one order is assigned F4, and total purchase value of ฿55,000 is assigned M1. Thus the ordered score is 441. The single order ORD-0007 provides the complete input for this profile. The article does not claim that those raw values produce the same quartiles for every shop or that R4 proves the customer has churned.

A higher-is-better mental model would reverse the meaning of this profile’s assigned positions. Read it instead as weak relative recency and frequency with top-quartile total spending in the fictional population. The 310-day record suggests checking whether the product was a one-off gift, whether the customer wants further contact and whether a relevant new item exists. A careful follow-up may be appropriate, but the score does not establish the reason for the long gap or guarantee recovery. Review the actual context before investing in a broad campaign.

Side-by-Side Decision Matrix for 111, 114, and 441

The table compares three different ordered score strings and the questions their fictional purchase histories suggest. The codes do not share the same digits: 111 contains three ones, 114 contains two ones and a four, and 441 contains two fours and a one. Read each position independently. A digit in the monetary position cannot replace a recency or frequency fact, and adding the digits loses that distinction.

Moving from 111 to 114 means monetary rank changes from score 1 to score 4 while the other assigned digits stay the same. It does not by itself prove that absolute spending fell; check the underlying values and comparison population. Moving from 441 to 141 changes the recency position from score 4 to score 1, but still does not prove a campaign caused the change. Keep the report date, definitions and raw purchase facts with every score comparison.

The Reversal Trap in Practice: How Other Platforms' Conventions Create Misreads

Convention confusion can arise when a teammate has learned another tool’s ratings. Confirm that tool’s documented scale rather than assuming a larger digit always means stronger behavior. Then translate the result into a description such as strongest or weakest on a particular dimension. A code containing 5 cannot be a valid CasperBrain four-level composite; do not construct counterfactual CasperBrain meanings for an out-of-range code.

Before comparing reports across tools, translate the code into tool-neutral language such as strongest recency band or weakest frequency band. Do not assume that another tool uses four groups, the same eligible records, the same definitions or the same share of customers per group. Even similarly named strongest bands need not be interchangeable. Keep CasperBrain’s four-level convention separate and compare actual purchase facts when a cross-tool decision matters.

One practical test: if your team's instinct is that '444' sounds like a strong score because four is a large number, the reversal trap is already present. In CasperBrain, '444' is the weakest possible composite across all three dimensions. If '111' sounds like a bad score because one is a small number, the same trap is operating. Correct the mental model before opening the data, not after.

Sources: Microsoft Docs: Set up Recency, Frequency, and Monetary (RFM) analysis in Dynamics 365 Commerce · Shopify: RFM Analysis — Definition and Segmentation Guide (2026)

Reading Score Changes Over Time Without Overreacting

An RFM score is a snapshot. It captures where a customer stands relative to the rest of the base at the moment the calculation runs. Between two calculation dates, three different things can produce a digit change: (a) the customer's own behavior changed, (b) other customers' behavior changed and shifted the quartile boundaries, or (c) the analysis window itself shifted and now includes or excludes different orders.

Before acting on a score change, compare the underlying values available to your role and the definitions used on both dates. A rank can change when the comparison population changes even if a given raw value remains the same. A rise from 5 to 45 days since purchase is a different observation from a rank change with unchanged recency. Record which observation you actually have before choosing a follow-up, rather than guessing a cause from the new digit.

RFM classification and cohort retention analysis answer different questions and should not be confused. RFM tells you where a customer stands relative to peers at a point in time. Cohort analysis tracks whether a group of customers acquired in the same period continues to buy over subsequent periods. A customer can look strong in RFM terms and still show declining cohort retention, or vice versa. Use the right lens for the question you are actually asking.

Sources: Shopify: RFM Analysis — Definition and Segmentation Guide (2026)

Permissions, Data Exports, and Score Visibility in CasperBrain

CasperBrain supports authorized customer CSV exports; field visibility follows the user’s permissions. Inspect the fields actually available in the current view or export before briefing a campaign. Do not invent a rule that a particular role can see segment names but cannot see score digits unless the current permission contract establishes it. If a needed field is unavailable, ask the administrator about the appropriate access rather than reconstructing private values from other labels.

Keep only the fields needed for the team’s next action, and share customer-level exports through the shop’s approved process. A descriptive segment label is a summary, not a reliable substitute for all three score positions or the underlying order history. Use the actual authorized records for interpretation and keep the scoring convention beside the handoff.

CasperBrain provides retention analysis and planning. CasperCall is a separate telesales execution product. Check the shop’s actual configured workflow before deciding how to hand off a list or task; the score itself does not prove that a call, message or synchronization has taken place. Keep preparation, execution and observed purchases as separate records so the team can review what actually happened.

Pre-Action Checklist: Six Steps Before Acting on Any RFM Score

The following checklist applies to any score you intend to act on, regardless of whether it is one of the three profiles discussed in this guide or a different composite entirely. Work through all six steps before briefing a campaign.

Step 1 — Confirm scale direction. State aloud or in writing: in CasperBrain, 1 is best and 4 is weakest on every dimension. Do this every time, especially when new team members are in the room.

Step 2 — Read all three positions separately as R, F and M. A sum of 6 can come from ten distinct codes: 114, 123, 132, 141, 213, 222, 231, 312, 321 and 411. They all have the same sum but different position values. Summing or averaging the digits therefore hides the information you need; examine the three dimensions and the purchase records instead.

Step 3 — Inspect the underlying values, not just the composite. Pull the actual days-since-last-purchase, distinct order count, and total spend from the export. Verify that the digit assignment makes intuitive sense given those numbers and your knowledge of the quartile boundaries.

Step 4 — Consider how much purchase history is available. One or two orders can leave limited evidence about a customer’s habits, but they do not force a particular frequency quartile: the assigned rank depends on the comparison data and scoring definition. A new buyer may simply not have had much opportunity to return. Use the length and context of the history to decide how confidently to interpret F.

Step 5 — Note the analysis window. A score computed on a 6-month window is not directly comparable to a score computed on a 12-month window for the same customer, because the quartile boundaries and the included orders differ. Check which window was used before comparing scores across time periods or across segments.

Step 6 — Separate observed purchases from causal lift. A treated group’s purchase rate describes what happened; it does not establish how much a campaign caused. A comparison group can help, but a difference between non-random groups may also reflect pre-existing differences. A suitably designed randomized comparison is stronger evidence. Label ordinary before/after reporting as observation and avoid attributing every subsequent purchase to the contact.

Sources: Shopify: RFM Analysis — Definition and Segmentation Guide (2026) · HubSpot: RFM Analysis — A Data-Driven Approach to Customer Segmentation

Key Limitations: What RFM Does Not Measure and Where It Can Mislead

RFM captures past purchase behavior. It does not capture product margin, so a customer with M = 1 may have spent ฿55,000 on low-margin items while another customer with M = 2 may have spent ฿30,000 on high-margin items at greater net profit to the shop. Always check margins when deciding how much to invest in retaining or re-engaging a high-M customer. A monetary score is a revenue signal, not a profitability guarantee.

RFM does not tell you why purchases stopped. A customer with R = 4 might have stopped buying because of a product quality problem, a life event such as relocation or job change, a competitor's promotional pull, or — as the Copper Cow Coffee example in Shopify's guide illustrates — because the product is a deliberate special-occasion purchase and the customer is entirely satisfied. Each scenario calls for a different response, but the score alone cannot distinguish between them. Qualitative signals — survey responses, support interactions, return history — are needed to complement the quantitative score.

In a small or highly tied dataset, ranked groups can provide less distinction than their labels suggest. No universal customer-count cutoff is established by this guide. Inspect the actual records, the distribution of values and the report definitions before treating a rank as a decisive customer difference. A small score change should prompt a question, not an automatic expensive campaign.

Finally, RFM does not predict future behavior with certainty — it surfaces patterns for prioritization. Observed purchase rates within a segment describe what customers in that segment did historically. They are correlated with, but do not guarantee, future behavior. The method is a prioritization tool, not a forecast engine. Use it to decide where to run a test, then measure the result against a control group to determine whether the action produced lift.

Sources: Shopify: RFM Analysis — Definition and Segmentation Guide (2026) · Shopify: Customer Scoring Models — Types and How to Build One

Sources and editorial method

Examples are fictional and contain no customer records. Prepared by the CasperBrain team with AI drafting assistance; facts and language are checked before publication.

• Shopify: RFM Analysis — Definition and Segmentation Guide (2026)
• Shopify: Customer Scoring Models — Types and How to Build One
• HubSpot: RFM Analysis — A Data-Driven Approach to Customer Segmentation
• Microsoft Docs: Set up Recency, Frequency, and Monetary (RFM) analysis in Dynamics 365 Commerce