Episode-level NPS shows a spread inside one brand that brand-level NPS hides: India FS Pulse

Generated 2026-08-22 by analysis/04_survey_nps.py. Every figure below is interpolated from a computed value, so the prose cannot drift from the data.

This module runs on synthetic data. 2,000 respondents were generated from a documented, seeded model (analysis/04_survey_nps.py, seed 20260820) because episode-level loyalty benchmarks for Indian banking are proprietary. The method is the deliverable. The numbers illustrate it and must not be quoted as market fact.

The answer

Measured at the level of the customer episode rather than the brand, digital-first providers do not have a loyalty advantage. They have an episode-shaped one. On this synthetic panel the neobank cohort scores +47 NPS on “Make a payment” and -47 on “Get support”: a 94-point spread inside a single brand. A blended brand-level NPS (+12) hides that entirely.

Three supporting arguments

1. The brand-level number is an average of opposites. Neobank overall NPS is +12, against +5 for private banks and -30 for public banks. Ranking on that single figure would misdirect investment, because the underlying episodes disagree with each other.

2. Automated episodes and human episodes separate cleanly. Digital-first wins the episodes a machine completes end to end (“Make a payment”). It loses the episode that requires a person (“Get support”), where Private bank leads. Loyalty here is not about brand affinity; it is about whether the episode had to escalate.

3. The gap is where the churn risk lives. An episode scoring -47 is a detractor-generating machine, and it sits inside the same relationship that Sub-module A identified as the monetisation route. Cross-sell attempted on the back of a failed support episode converts poorly.

So what

Measure and manage at episode level, not brand level. Episode NPS is actionable; brand NPS is a scoreboard.
Fix the escalation path before monetising the relationship. The credit and distribution pathways in Sub-module A depend on trust that the “Get support” episode is currently spending.
For diligence: ask a target for episode-level NPS. If they only have a blended figure, they do not know which part of their franchise is working.

Method

Ratings drawn from a per-(cohort, episode) normal latent model, clipped to 0-10, with a small age adjustment applied to the digital-first cohort. NPS computed the standard way: %promoters (9-10) minus %detractors (0-6). Re-running the script reproduces these figures exactly.