Strategy

Customer Segmentation for Loyalty Programs

Silver/Gold/Platinum was a useful framework in 2005. Today, it masks 11+ behavioural states members move through every quarter. A field guide to modern customer segmentation.

2026-02-1112 min read

A tier is a label. A cohort is a state. The retail industry has spent two decades labelling members — Silver, Gold, Platinum — and treating the labels as if they explain behaviour. They don't. They are a marketing construct, not an analytical one.

Behavioural cohorts — derived from data, updated continuously, predictive in nature — explain behaviour. Modern loyalty operators run both: tiers as the member-facing narrative (status, recognition), cohorts as the operational reality (campaign targeting, retention investment).

Why static tiers fail

Three reasons. First, tiers are based on cumulative spend — a backward-looking measure that punishes new high-potential members. Second, tier qualification windows (typically annual) miss seasonal patterns and rolling momentum. Third, tier benefits are uniform within a tier, which makes them generic by definition.

The data

Across Fundle programmes, members in the same tier show 8-12× variation in 90-day repeat rate. The tier explains <15% of behaviour variance. Behavioural cohorts explain 60-70%.

The 11+ cohorts that explain everything

A working behavioural cohort framework, refined across hundreds of programmes:

CohortDefinitionStrategy
ChampionsTop 10% CLV, frequent, recentRecognition + exclusive access. Discount-light.
LoyalSteady frequency, mid-high CLVCross-category nudge. Maintain.
Potential Loyalist2-3 purchases, < 90 daysHabit-forming triggers. Second-purchase incentive.
NewFirst 30 daysOnboarding journey. Welcome to category.
Big SpendersHigh M, low FFrequency-building. Cross-occasion engagement.
PromisingRecent first purchase, high engagementConvert to Loyal. Personal touch.
Customers Needing AttentionHigh historical CLV, declining recencyPersonalised win-back. Investment-worthy.
At-RiskDeclining frequency 30-60 daysUrgent win-back. AI-sized offer.
Hibernating60-120 days inactiveDeep discount + reason to return.
One-TimersSingle purchase, 90+ days inactiveSecond-purchase mission. High-leverage.
Lost180+ days inactive, low past CLVDe-prioritise. Resource cost > expected return.

How to build cohorts that hold up

Three principles that separate cohorts from arbitrary buckets.

1. Definitions must be predictive, not descriptive

A good cohort predicts future behaviour. "Members who spent ₹10k last year" is descriptive. "Members whose visit frequency dropped 40% in the last 30 days" is predictive — they're more likely to churn. Predictive definitions justify intervention.

2. Refresh frequency must match decision cadence

If campaigns run weekly, cohorts must refresh daily (so weekly decisions use 1-day-old data, not 8-day-old). Monthly refresh is too slow for tactical work.

3. Migration matters more than membership

The interesting question isn't "who is in At-Risk today" but "who entered At-Risk this week". The latter triggers action; the former is a static snapshot.

From cohorts to campaigns

Every cohort should have a default campaign or journey associated with it. When a member enters the cohort (e.g., transitions to At-Risk), the journey fires automatically. This is the only way segmentation becomes operational at scale — no human in the loop deciding to send a campaign to a list.

See AI cohorts on your member base

30-minute walkthrough of Fundle's 11+ live behavioural cohorts on a sample dataset.

FAQs

Should we abandon Silver/Gold/Platinum tiers?

No — keep them as the member-facing narrative. Status matters. Just don't use them for campaign targeting; use behavioural cohorts for that.

How many cohorts is too many?

Beyond ~15, you're overfitting. The right number is the smallest set that captures distinct behavioural states. 11-13 is the practical sweet spot for most retail categories.