Customer Insights Agent AI

Invoked when: weekly NPS data batch is received from the survey platform — analysis runs across all responses collected in the prior 7 days

• –Processes every NPS response received in the batch — the numeric score (0–10) and the verbatim open-text response where provided. Classifies each response as Promoter (9–10), Passive (7–8), or Detractor (0–6), and extracts the primary theme from the verbatim response: the specific aspect of the LendingIQ experience the borrower is commenting on — processing speed, communication quality, interest rate, disbursement experience, repayment process, or app usability.
• –Identifies the top 3 drivers of Detractor scores for the current week — the themes most frequently cited by borrowers scoring 0–6 — and compares them to the prior 4-week baseline. A driver that has increased its share of Detractor responses week-on-week is an emerging issue; a driver that has been in the top 3 for more than 4 consecutive weeks is a systemic problem requiring structural change, not an operational fix.
• –Surfaces illustrative verbatim examples for each top driver — two or three verbatims that are representative of the theme, not selected to be the most extreme but the most typical. Verbatims give the product team the borrower's voice alongside the statistical pattern — the combination of frequency data and authentic language is what makes an insight actionable rather than abstract.

Invoked when: daily support ticket batch is processed — all tickets closed or updated in the prior 24 hours are clustered and themed

• –Processes every support ticket in the daily batch — the issue description, the category assigned by the support team, the resolution applied, and the resolution time. Groups tickets into thematic clusters based on the underlying issue rather than the surface category: tickets categorised as 'account query' may cluster into 'EMI date confusion', 'outstanding balance discrepancy', or 'pre-closure process uncertainty' — three different underlying problems that the surface category obscures.
• –Tracks cluster volume trends — the number of tickets in each cluster per day, the week-on-week change, and the resolution time per cluster. A cluster that is growing in volume is a product or process problem getting worse; a cluster with significantly higher resolution time than others indicates an area where the support team lacks clear resolution guidance or authority.
• –Cross-references high-volume support ticket clusters with the corresponding stage in the borrower journey — if 40% of EMI confusion tickets come from borrowers in their first month of repayment, the issue is at the repayment onboarding stage, not in the product itself. Journey-stage attribution of support clusters is the difference between fixing the product and fixing the explanation of the product.

Invoked when: monthly analysis cycle runs — portfolio repayment rate data from the LOS is correlated with feedback and support ticket themes from the same period

• –Correlates repayment rate trends with feedback themes and support ticket clusters from the corresponding period — looking for the feedback signals that preceded or accompanied repayment rate changes. Where repayment rates have fallen in a specific borrower segment, the agent searches the feedback corpus for themes that were elevated in that segment in the weeks before the decline: dissatisfaction with communication frequency, confusion about the repayment process, or frustration with EMI amount rigidity are common precursors to repayment behaviour changes.
• –Distinguishes between feedback-correlated and feedback-uncorrelated repayment changes. A repayment rate decline with no corresponding feedback signal elevation is likely driven by external factors — macro conditions, seasonal income variation — rather than a LendingIQ product or process failure. A decline where a specific feedback theme has been elevated for the prior 6 weeks is more likely to be operationally addressable.
• –Produces a monthly root cause report with the top 3 operationally addressable causes of churn — the feedback-correlated patterns that the product, marketing, or operations team could plausibly change — ranked by the estimated number of borrowers affected. The report is intended to inform the product roadmap and the operations review; it is not a causal proof but a structured hypothesis that the team can investigate and test.