A loyalty programme is a flow of customers through behavioural states. New members arrive, some graduate to active loyalists, some plateau, some leave. Every operator knows this in the abstract. Almost none track it in the data.
Cohort analytics is the discipline of segmenting members into groups that share a defining characteristic, then watching how those groups behave over time. Done well, it explains 80% of programme performance questions that quarterly summaries cannot answer. Done badly, it produces pretty charts that don't change any decisions.
The seven cohort lenses every loyalty programme should run
1. Acquisition cohort (the foundation)
Group members by the month they enrolled. Track each month's cohort separately: how many were still active at month 3, 6, 12, 18? This single view reveals onboarding strength. If the Jan-2026 cohort is at 38% active at month 6 and the Apr-2026 cohort is at 29% active at month 6, something between January and April broke. Probably the onboarding journey, the channel mix, or the first-purchase incentive.
2. Recency cohort (R in RFM)
Group members by recency of last purchase: 0-30 days, 31-60 days, 61-90 days, 91-180 days, 180+ days. The 31-60 day band is the most strategically important — members in this state are still warm but tipping toward dormant. Win-back campaigns deployed here cost 1/4 of the same campaign deployed at 180+ days, with 3× the response rate.
3. Frequency cohort (F in RFM)
Group members by visit count over a 12-month window: 1 visit, 2-3, 4-7, 8-15, 16+. The 1-visit cohort is your largest growth opportunity and your biggest leak — converting one-timers to second-timers is the highest-leverage move in retail loyalty.
4. Monetary cohort (M in RFM)
Group by 12-month spend: ₹0-2k, 2-5k, 5-15k, 15-50k, 50k+. Layered on frequency, this creates the classic RFM grid. Champions (high F, high M) and Big Spenders (low F, high M) require different strategies — Champions need recognition, Big Spenders need retention investment.
5. Category cohort
Group by the category of first purchase (or dominant category). Fashion-first members behave differently from electronics-first members; cross-category migration is a strong loyalty signal. Members who buy in 3+ categories have 2-4× higher CLV than single-category members.
6. Store / channel cohort
Group by acquisition store or acquisition channel (app, walk-in, marketplace, friend referral). Channel-of-acquisition is the strongest predictor of behaviour you have. Referred members repeat 30-50% more than walk-in members on average; channel-mix decisions follow directly.
7. Behavioural / lifecycle cohort
Group by the behavioural state machine: New (first 30 days), Potential Loyalist (2-3 visits, < 90 days), Loyal (4+ visits, < 120 days), Champion (top quintile of CLV), At-Risk (declining frequency), Hibernating (60+ days no purchase), One-Timer (1 visit, > 90 days no second purchase). Migrations between these states are the most diagnostic signal in the entire programme.
A worked example: the "New → Loyal" funnel
Consider the most important conversion in any loyalty programme — moving a new member to "Loyal" within 6 months. This is what justifies the acquisition cost. If 100 members enrol in January and 22 are Loyal by July, the funnel is performing healthily. If only 11 are, the programme is leaking.
| Stage | Cohort | Target conversion (90d) | Diagnostic if missed |
|---|---|---|---|
| Enrolment | 100 new members | — | — |
| First purchase | Used loyalty in first 30 days | 60-75% | Onboarding incentive or enrolment-channel weak |
| Second purchase | 2+ purchases by day 90 | 32-45% | No second-purchase trigger; missing nudge |
| Third purchase | 3+ purchases by day 90 | 18-28% | Habit not forming; no category cross-sell |
| Loyal status | Behavioural "Loyal" by day 180 | 18-28% | Repeat rhythm broken; check channel cadence |
The diagnostic column is where cohort analysis earns its keep. A drop at each stage points to a specific lever. A drop at "Second purchase" means the onboarding journey is missing a triggered nudge. A drop at "Third purchase" means there's no cross-category cross-sell mechanism. The intervention is precise; without cohort funnel analysis, the intervention is "send more emails".
Cohort migration tables: the most underused report in loyalty
A cohort migration table shows how members moved between behavioural states over a period. Rows = source state at start. Columns = destination state at end. Cells = count or percentage.
Example monthly migration (Jan → Feb 2026):
| From / To | New | Potential Loyalist | Loyal | At-Risk | Hibernating |
|---|---|---|---|---|---|
| New | — | 38% | 4% | 12% | 46% |
| Potential Loyalist | — | 54% | 22% | 15% | 9% |
| Loyal | — | 6% | 78% | 12% | 4% |
| At-Risk | — | 2% | 8% | 52% | 38% |
| Hibernating | — | 1% | 2% | 4% | 93% |
The diagonal (Loyal → Loyal: 78%) is the retention story. The lower-triangle (At-Risk → Hibernating: 38%) is the leakage story. The upper-triangle (At-Risk → Loyal: 8%) is the win-back story. A healthy programme has high diagonal, low lower-triangle, and growing upper-triangle. An unhealthy programme has the opposite pattern.
Watch the "New → Hibernating" transition closely. It's the largest leak in most programmes. If 46% of new members go silent within 30 days, your onboarding journey isn't firing or isn't valuable enough. Fix this before fixing anything else.
Combining cohort lenses (where it gets powerful)
A single cohort lens explains a slice. Two combined lenses explain the picture. Three combined lenses are where the strategic moves emerge.
Example combined query: "Show me the 1-visit cohort, acquired through app installs in Bangalore, in the cosmetics category, who haven't purchased in 31-60 days." That's an extremely specific high-leverage segment. With Fundle's natural-language query interface, this is a 12-second question. Without it, it's a half-day analyst exercise that never gets done.
Acting on cohorts (the part most teams skip)
Cohort analysis is worthless unless it changes campaigns. The right operational rhythm:
- Daily: Watch the New cohort's first-30-day funnel. Alert on any 2σ deviation.
- Weekly: Cohort migration table for the week. Highlight the worst-performing transition; assign owner.
- Monthly: 90-day repeat rate by acquisition cohort. Drill into the worst cohort's acquisition channel and onboarding journey.
- Quarterly: Full cohort scorecard for the board — CLV uplift by acquisition cohort over time. This is the proof loyalty drives compounding business value.
What this looks like in practice
A regional fashion chain Fundle worked with had a healthy 41% repeat rate at programme level — which masked a desperate situation in the New cohort. 71% of new members became Hibernating within 60 days because their post-enrolment journey had a single "thank you" message and no second-purchase trigger. Adding a 5-step onboarding journey lifted New → Potential Loyalist conversion from 22% to 38% in 90 days. Overall repeat rate moved from 41% to 47% — entirely from the New cohort fix.
The headline metric (repeat rate) couldn't locate the problem. Cohort analytics could. That gap is what separates a working loyalty operation from a reporting exercise.
See cohort migrations on your data
Fundle's analytics layer ships with cohort migration, multi-lens combination, and natural-language query interface. 30-minute demo on a sample dataset.
FAQs
How many cohort lenses should we run?
All seven listed here, every quarter. Most operators run two or three; the other four are where the differentiated insight is. The marginal cost of adding lenses is low; the marginal value is high.
How frequently should cohorts refresh?
Acquisition cohorts: monthly. Behavioural cohorts: daily. RFM cohorts: weekly. Most operators refresh monthly across the board, which makes the data stale for tactical decisions.
Can we do this in a BI tool?
Cohort tables can be built in Looker / Tableau if you have a strong data team. The challenge is keeping definitions consistent across reports, refreshing in real time, and exposing them to non-analyst users. Fundle ships these constructs native; the analyst team is freed for higher-value work.