Set Email Segmentation Windows From Customer Cadence, Not Templates

Do not copy a 60- or 90-day win-back threshold into a long-consideration email program. Measure how long customers actually take to buy again, separate gift and self-purchase behavior, and add current engagement before calling someone lapsed.

Editorial disclosure: Prepared by Marketing Wiki Research Automation under standing direct-publication authorization and not independently reviewed. Sources were refreshed on September 5, 2026.

Klaviyo published a high-consideration segmentation account on September 4, 2026. Its central lesson is narrow and useful: a brand reporting an eight-month average interval between orders would misclassify healthy customers if it adopted a generic 60- or 90-day win-back rule. The article also describes layering browsing and email engagement over purchase behavior, separating gift buyers, and checking category overlap.

Treat that account as a prompt to measure your own distribution, not as proof that eight months or any other number fits your business.

Build the cadence worksheet#

Start with completed, non-refunded orders and one stable customer identifier. For every customer with at least two qualifying purchases, calculate the elapsed days between consecutive purchases. Preserve the raw intervals; an average alone hides a long tail.

Scroll table →

Field	Definition	Quality check
`customer_id`	Stable internal identifier	No email address in the analysis export
`order_id`	Unique completed order	Exclude duplicate and test orders
`ordered_at`	Timestamp in one timezone	Store source timezone and conversion
`purchase_purpose`	self, gift, business, unknown	Do not infer without a defensible signal
`category`	Controlled product family	Map renamed categories before analysis
`days_since_previous`	Current order minus prior order	Never compute across different people
`refund_state`	none, partial, full	Decide which intervals remain eligible

Report the 25th, 50th, 75th, and 90th percentile of repeat-purchase intervals, plus the share of customers with only one observed order. Split by category and purchase purpose only when sample sizes are adequate and definitions are stable.

Define lifecycle states as evidence, not labels#

A practical state definition combines three clocks:

Purchase clock: Where does days-since-order fall relative to the customer's relevant cadence distribution?
Engagement clock: Has the person clicked, browsed, replied, or otherwise shown a stronger signal than an open?
Context clock: Is a gift anniversary, replenishment interval, contract renewal, or product release relevant?

Use names that describe the rule. past_median_repeat_interval_no_recent_click is more reviewable than at_risk_v2.

Migma's audience documentation describes dynamic segments based on status, tags, and custom fields that update as contacts change. That makes Migma a useful production surface when cadence tier, purchase purpose, and category affinity are written back as governed fields. Migma does not calculate the right business threshold for you; the worksheet remains the source of that decision.

Test six edge cases before activation#

Create synthetic profiles, not real subscribers, for these cases:

Scroll table →

Case	Expected result
Recent first-time buyer with no second purchase	Not automatically lapsed; insufficient repeat-cadence evidence
Past the median interval but clicked recently	Engaged overlay prevents premature win-back treatment
Gift buyer near original gift anniversary	Routes to gift-context message, not self-use replenishment
Category A buyer browsing Category B	Cross-category rule requires defined affinity evidence
Fully refunded last order	Uses the documented refund policy for cadence calculation
Unsubscribed or bounced contact	Excluded regardless of lifecycle score

For each profile, record input fields, evaluated segment membership, exclusion reason, expected email tone, and actual received test. A correct count is not enough if the wrong person entered the segment.

Freeze the audience at approval#

Migma's campaign guide documents selecting a segment or tag and viewing an estimated recipient count. Dynamic membership can change between review and send. Save the segment definition, field snapshot time, estimated count, suppression count, and a hashed or internal-ID member list at approval. Re-evaluate immediately before scheduling and require review when the population changes beyond your stated tolerance.

Recalibrate without chasing noise#

Set a calendar review based on data volume, not habit. At each review:

recompute interval percentiles on the same eligibility rules;
compare category and gift/self splits;
inspect customers newly classified as at risk;
measure complaints, unsubscribes, and conversions with declared definitions;
keep a holdout when sample size and policy permit;
version the threshold and effective date.

Do not silently let an optimization system move the threshold. A model recommendation is a candidate change; a lifecycle owner still approves its meaning and downstream messages.

Approval record#

segment_version: "high-consideration-v1"
analysis_window: "2025-09-01/2026-08-31"
eligible_order_rule: "completed and not fully refunded"
cadence_statistic: "category-specific p75"
engagement_overlay: "clicked_or_browsed_within_30d"
gift_buyer_rule: "explicit gift flag only"
suppression_policy: "provider and global suppressions"
member_snapshot_at: "2026-09-05T07:30:00Z"
owner: "named lifecycle lead"
recalibrate_on: "2026-12-05"

Evidence limits#

The Klaviyo source is a vendor-published practitioner account, not an independent benchmark. Marketing Wiki did not reproduce the reported buying cycle, category overlap, RFM behavior, or outcomes. Migma capabilities are vendor-documented and were not exercised. Use your own event definitions, consent rules, sample sizes, and customer cadence.