Analytics

Loyalty Analytics: From Points Liability to Customer Intelligence

Most loyalty dashboards measure activity. Almost none measure outcomes. Here is the KPI stack that proves a loyalty programme drives business — and the analytics architecture behind it.

2026-02-0716 min read

Ask any retail CMO for their loyalty dashboard, and you will get the same four numbers: members enrolled, points issued, points redeemed, redemption rate. Helpful operational telemetry, but useless as a measure of business impact. None of them answer the only question the CFO will ask: "Is this programme generating incremental revenue?"

The shift from activity metrics to outcome metrics is the single most important upgrade a loyalty operator can make. It is also the cheapest — the data is already being collected; only the reporting frame is wrong.

The two layers of loyalty analytics

Modern loyalty analytics splits into two layers. Operational metrics — points issued, redeemed, breakage, liability — are the meter-readings of the programme. They need to be accurate and real-time, but they're not what proves the programme works. Intelligence metrics — repeat rate, frequency, ATV growth, cohort migration, CLV uplift, programme incrementality — are what proves the programme works.

A complete analytics stack runs both. A weak stack runs only the operational layer and calls it analytics.

The 10 metrics that actually matter

MetricWhat it measuresHealthy benchmark (Indian retail)
Repeat rate (90d)% of members who purchased in last 90 days38-55% by category
Visit frequencyAvg visits per active member per quarter2.2-4.8 by category
ATV growth (cohort)Average transaction value uplift by cohort+8-22% within 12 months
Cohort migration% of new members moving to "Loyal" within 6 months18-28%
Predicted CLV12-month forward CLV per memberCategory-specific
Churn risk (30-day)% of members predicted to churn in 30 days12-22%
Earn-burn ratioPoints redeemed / points earned30-45%
Points liabilityOutstanding-points financial liabilityTrack trend, not absolute
Programme incrementalityRevenue from members vs propensity-matched non-members12-28% lift on engaged segments
Member NPS (loyalty-specific)NPS among programme members vs non-members+10-20 points premium

Repeat rate — the single most underrated metric

If your dashboard only reports one outcome metric, make it 90-day repeat rate by cohort. It is the cleanest measure of whether the loyalty programme is doing its job, and it requires no fancy modelling — just disciplined cohort tracking.

Most operators don't track it because they don't define cohorts consistently. The right definitions: acquisition cohort (month enrolled), category cohort (first category bought), channel cohort (acquisition channel), store cohort (first-purchase store). Each tells a different story; together they explain repeat behaviour to a level traditional reports never reach.

Practitioner note

Track 90-day repeat rate week-over-week, not month-over-month. Programme drift shows up early in weekly numbers and is averaged away in monthly ones. The cohort that drops 4% WoW is the early-warning signal that something — channel mix, store ops, campaign cadence — broke.

Cohort migration: the diagnostic that explains everything

A healthy loyalty programme is a flow chart. New members enter "New". Some move to "Potential Loyalists" (2-3 transactions). Some move further to "Loyal" or "Champions". Others slide back to "At-Risk" or "Hibernating". Each transition is a behavioural signal; together they form a migration map.

Reports that show the headcount in each segment today are nearly useless. Reports that show transitions week-over-week — how many New became Potential Loyalists, how many Loyal became At-Risk — are diagnostic gold. When repeat rate drops 4%, cohort migration tells you where it broke (typically: New cohort failing to graduate to Loyal in months 2-3).

Predicted CLV — the underlying compass

Customer Lifetime Value is the right strategic compass for the entire programme. Done well, it lets you invest proportionally in each member from day one. Done badly (the way most platforms do it: historical revenue extrapolated), it misleads.

Modern CLV modelling uses transactions, visit frequency, category breadth, recency, channel engagement, demographic signals and seasonality — combined in a probabilistic model that updates with every new transaction. Fundle's CLV model fires at first transaction and refines continuously; reports show CLV uplift per cohort over time, not point-in-time CLV.

Programme incrementality — the CFO's question

Incrementality is the only number that proves the programme is doing its job. The question is simple: of the revenue generated by members, how much would have happened without the programme? Without an answer, every other number is suspect.

The right answer uses propensity-matched control groups. For every targeted campaign, hold out a small fraction (5-10%) of the eligible cohort, matched on observable behaviour. After the campaign window, compare the treatment cohort's revenue to the control. The delta is incremental.

The reality

Most loyalty platforms cannot run automatic control groups. They report "campaign generated ₹X revenue" but cannot say what would have happened without it. Fundle holds out a control by default on every campaign; this is the difference between reporting and intelligence.

Points liability — the financial discipline most teams skip

Every point issued is a financial liability — a promise to deliver value the member can call in. Programmes that don't track liability live with quarterly surprises that look like accounting errors to the CFO and never get fully fixed.

The right liability dashboard shows: total outstanding points, monetary equivalent, projected burn over next 12 months, breakage estimate (points that will expire unused), and movement (points earned this period, redeemed, expired, adjusted). The math is simple; the discipline of running it monthly with finance is what most teams skip.

The natural-language layer

No CMO will become a SQL analyst. The most underused capability in modern loyalty platforms is a natural-language interface to the data. "Show me at-risk fashion members in Mumbai who spent more than ₹10,000 last quarter, with their phone numbers ready for a WhatsApp push." Three years ago this took a 4-hour analyst job. Today, a well-built LLM-backed interface answers it in 12 seconds with a downloadable audience.

Fundle Brain ships this surface; if your current platform does not, you are leaving capability on the table that your team will quietly route around — and the routing-around is what creates the 4-hour analyst jobs you don't see on the dashboard.

A quarterly board pack template

  1. Programme summary: total members, active members (90d), repeat rate by category cohort
  2. Cohort migration: net flow into Loyal/Champions, net flow into At-Risk/Hibernating
  3. Incrementality: top 5 campaigns by incremental revenue, control-vs-treatment lift
  4. CLV uplift: median predicted CLV today vs 12 months ago, by acquisition cohort
  5. Points liability movement: outstanding points, monetary equivalent, breakage projection
  6. Channel mix: SMS / WhatsApp / RCS / push / email reach and engagement
  7. Top 5 opportunities the AI agents have flagged for next quarter

A two-slide version of this, exported one-click from the platform, is what gets loyalty programmes funded for another year. A 14-slide vanity deck of "points issued" is what gets them cut.

See your loyalty data through Fundle's analytics lens

30-minute walkthrough. Real cohort migrations. Real incrementality reports. Real natural-language queries on real numbers.

FAQs

How is this different from a BI tool like Looker or Tableau?

BI tools require modelling. Loyalty constructs (tiers, points, redemption, RFM segments) need to be built in SQL or LookML before they can be reported on, and re-built every time the schema changes. Fundle ships these constructs native; reports are accurate without modelling effort. Most teams still connect a BI tool downstream for ad-hoc analysis, but the core daily reporting lives in the loyalty platform.

How do we measure programme incrementality if we never set up control groups?

You can backfit, partially: select propensity-matched non-members of similar pre-programme behaviour, and measure post-launch difference. It's imperfect, but better than nothing. Going forward, set up automatic holdouts on every campaign — the cost is small, the analytical clarity is enormous.

What's a healthy breakage rate?

Category-dependent, but typically 15-30%. Too low means you over-paid for liability that gets called in. Too high means rewards aren't compelling and members are giving up. Track the trend more than the absolute number.