An NPS score of 28 tells the institution that more borrowers are unhappy than delighted, but it does not tell it why — and "why" is the only question whose answer produces action. A management team that sees an NPS of 28 and concludes "we need to improve customer experience" has produced a resolution without a diagnosis. The Customer Insights Agent AI reads every open-text NPS comment from every detractor, classifies each comment by theme using natural language processing, identifies the three causes that appear most frequently in detractor responses and least frequently in promoter responses, and delivers the finding with verbatim evidence in under 4 minutes. The management team that receives this output can improve something specific — not "customer experience."
Why NPS scores without driver analysis are practically useless
NPS methodology asks two things: a 0–10 rating and an optional open-text comment explaining the rating. Most institutions track the score faithfully and ignore the comments systematically. The comments are the data — they are the borrower's direct statement of what produced the rating. An NPS of 28 is not a finding; it is a question. The question is: what are the 28% of detractors angry about, and how does it differ from what the 44% of promoters are happy about? The difference between those two sets of comments is the causal map of the institution's customer experience — the exact things that need to be fixed to move the score.
The challenge is volume. An NBFC with 48,000 active accounts running quarterly NPS surveys receives 8,000 to 12,000 comment responses per quarter. No human team reads them all. The Customer Insights Agent AI does — categorising every comment, scoring the sentiment, extracting the themes that drive detractor scores and the themes that drive promoter scores, and producing a driver analysis that distinguishes between the two.
The NPS driver analysis: Q3 2025 survey results
2 · Collections FPC audit (Driver #2 = regulatory risk)
3 · Charges disclosure at sanction (not fine print)
What makes these three drivers analytically significant — and not just common complaints
The Customer Insights AI identifies drivers not just by frequency in detractor comments, but by the delta between detractor frequency and promoter frequency. Many complaints appear in both detractor and promoter responses — for instance, "high interest rate" appears in 28% of detractor comments but also in 9% of promoter comments (who mention it to say they expected worse). The rate complaint is real but it does not drive detractor scores — borrowers who are otherwise satisfied do not downgrade their NPS to 2 or 3 because the rate is high. They accept the rate. Driver #1 (processing delay with no communication) appears in 68% of detractor comments and 4% of promoter comments — a 64-point delta. That gap is the driver signal.
Driver #2 is not just a satisfaction problem. The NLP analysis classified 184 of the detractor comments containing "collections" as reporting conduct that matches RBI FPC violation criteria — specifically, contact outside permitted hours, contact with family members without consent, and threatening language. These 184 comments represent a compliance event that the collections team and the compliance officer need to review immediately, independent of the NPS score. The Customer Insights AI flags these for regulatory action, not just customer experience action.
Driver #2 is not a customer satisfaction issue — it is a compliance issue. The NPS analysis found it; the compliance team needs to act on it.
An NPS comment that reads "they called my employer and it was humiliating" is three things simultaneously: a detractor score, a grievance, and an FPC violation report. The Customer Insights Agent AI classifies it as all three and routes it accordingly — to the NPS driver analysis, to the grievance system, and to the compliance officer's attention. No single-purpose NPS tool does this. The analysis that treats this comment only as a satisfaction data point has missed the regulatory signal that sits inside the same sentence. Customer Insights AI's value is not that it reads comments faster — it is that it reads them with multiple interpretive lenses simultaneously and routes each finding to the function that should act on it.