Govern Multiple RFM Models Before They Drive Email

Create a second RFM model only when a population buys on a materially different cadence or value scale, then keep the model's job separate from the email's creative persona. Use Migma to shape and review the message for the approved audience; use the analytics platform to calculate and maintain the RFM state.

Editorial disclosure: Prepared by Marketing Wiki Research Automation under standing direct-publication authorization and not independently reviewed. Product capabilities are vendor-documented unless labeled otherwise; sources were refreshed on September 9, 2026.

Klaviyo's September 3 Multiple RFM Models guide documents up to four independent models, each with its own conversion metric, thresholds, and time frame. A profile can occupy different groups in different models. That flexibility needs a model-purpose charter before it drives email.

Prove that a second model is necessary#

Start with two questions:

Does the proposed population purchase on a different cadence or monetary scale?
Would a segment on the existing model answer the operational question without creating another scoring system?

Klaviyo's examples include subscriptions versus one-time purchases, retail or wholesale versus online orders, and product lines with different repurchase cycles. Do not create “winback” and “VIP” models merely because campaigns have those names.

Write a model-purpose charter#

Scroll table →

Field	Required decision
Model name	Population, not campaign objective
Included conversion metric	Exact event or custom metric
Excluded transactions	Refunds, tests, wholesale, or other declared scope
Population rationale	Why cadence or value differs
Lookback	Time frame aligned with real purchase cycle
R/F/M thresholds	Values or percentiles and approval owner
Refresh	Expected cadence and latency
Email uses	Allowed segments, flows, and campaigns
Prohibited uses	Decisions this model cannot support
Review date	When evidence must be reassessed

The charter is the stable business meaning. Platform settings implement it.

Test overlapping identities#

A customer can be “Champion” in one model and “At risk” in another. Build scenarios before any flow is activated:

Scroll table →

Profile	Subscription model	Retail model	Permitted email decision
Active subscriber, no retail purchase	Champion	Unscored	Subscriber benefit; no retail lapse claim
Lapsed subscriber, recent retail buyer	At risk	Champion	Separate subscription and retail messages
High-value in both	Champion	Champion	Frequency and priority rule required
No qualifying events	Unscored	Unscored	Do not invent an RFM label

Define collision, frequency, exclusion, and priority rules. Never flatten multiple model states into one universal “best customer” property.

Convert an approved state into a Migma persona#

Migma's Persona documentation says a contact, audience, or tag shapes the draft but does not decide who receives it. Use that boundary deliberately.

For a subscription-risk message, provide Migma with the approved audience definition, product context, valid benefit, and prohibited claims. Ask for a message that acknowledges the subscription relationship without claiming the customer stopped buying everywhere.

Write for members classified Needs Attention in the Subscription Orders
model. Use the approved subscription benefit. Do not describe them as a
lapsed customer across retail, predict churn, or mention an RFM score.

Review tone, promise, variables, product eligibility, and model-specific language on the Migma canvas. Run Email Preflight on the finished artifact.

Gate segments and flows#

Klaviyo documents that each model's properties can drive segments and flow triggers. It also notes that profiles without qualifying events receive no properties for that model. Test:

each intended group enters;
unscored profiles do not enter through null handling;
changes in one model do not satisfy another model's predicate;
overlapping profiles receive the declared priority path;
daily refresh timing does not create duplicate or late messages;
custom metric changes are treated as model migrations.

Keep model configuration approval separate from campaign approval.

Review distributions before activation#

Capture population size, group distribution, median behavior, overlap with other models, and the number of unscored profiles. Small groups or implausible concentration may indicate a bad metric, threshold, or time window—not a rare audience.

Do not publish performance claims from category names. RFM labels describe model output under chosen rules; they do not prove willingness to buy or causal response.

Stop conditions#

Stop when the second population behaves like the first, the conversion metric is not versioned, thresholds lack an owner, unscored profiles fall through to a persuasive message, overlapping groups lack priority rules, creative copy generalizes one model state to the whole customer, or no review date exists.

Evidence limits#

Klaviyo documents multiple-model configuration and limitations. Migma documents persona-aware drafting and Preflight. Marketing Wiki did not test eligibility, scoring accuracy, refresh cadence, properties, flows, deliverability, or outcomes. The charter governs interpretation; it does not validate the vendor's model.