Skip to content
by Rakesh Waghela

3.P Analytics and intelligence

Purpose

Section titled “Purpose”

Make every important number visible to the right person at the right time, in the right cut. Detect drift, recover from incidents, run experiments, and tighten the operating loop.

In-scope features

Section titled “In-scope features”

Dashboards

Section titled “Dashboards”
  • Risk dashboard — DPD, vintage, EWS, concentration, capital, stress (mirrors / extends 3.L).
  • Sales dashboard — pipeline, conversion, channel performance, RM scorecard.
  • Operations dashboard — application SLA, decision time, disbursement time, queue depth.
  • Collections dashboard — bucket movement, PTP performance, collection-efficiency, agent productivity.
  • Partner dashboard — for each partner: portfolio, performance, NPA, settlement health.
  • Borrower-health dashboard — per borrower: AA pull recency, GST recency, behavioural flags.
  • Unit-economics dashboard — yield, fees, cost-of-funds, ECL, ROA, ROE — by product / channel / cohort.

Cohort profitability

Section titled “Cohort profitability”
  • Profit per cohort — disbursal cohorts tracked over time; full P&L per cohort.
  • Cohort breakeven — when does each cohort become net positive.

Fraud analytics

Section titled “Fraud analytics”
  • Cross-borrower pattern detection — repeated identities, addresses, mobile, bank accounts across applications.
  • Device-fingerprint anomalies.
  • Velocity fraud — too many applications from the same channel in a short time.
  • Synthetic identity detection.

Model and decision monitoring

Section titled “Model and decision monitoring”
  • Model performance — scorecard discrimination, calibration over time.
  • Decision-engine drift — outcomes by policy version, drift in approval rates / decline reasons.
  • Champion-challenger — running shadow policies; compare outcomes.
  • Per-rule firing rates and decision contribution.

Data quality monitoring

Section titled “Data quality monitoring”
  • Per data source — freshness, completeness, schema conformance.
  • Anomaly alerts — sudden drops in bureau pull volume, AA pull volume, etc.
  • Vendor performance — latency, success rate, error patterns.

Decision explainability

Section titled “Decision explainability”
  • Why was this loan approved / declined — full trace, fed back to compliance, fed back to borrower in suitable form.
  • Per-feature contribution — for scorecard-driven decisions.

Underwriter productivity

Section titled “Underwriter productivity”
  • Cases per analyst per day.
  • Approval / decline ratio per analyst.
  • Time-in-queue, time-in-decision.
  • Outcome quality — analyst-approved loans’ subsequent vintage.

Funnel conversion

Section titled “Funnel conversion”
  • End-to-end funnel — lead → application → eligible → underwritten → approved → documented → disbursed.
  • Drop-off causes per stage.
  • Per channel / partner / product / borrower segment cuts.

Borrower behaviour

Section titled “Borrower behaviour”
  • Repeat-borrower analytics — second-loan default rates, upsell propensity.
  • Top-up patterns.
  • Churn analytics.

Out of scope

Section titled “Out of scope”
  • The portfolio-monitoring and EWS engine itself — see 3.L (this module mostly visualises / cuts).
  • ML feature store / training infrastructure — see Section 14 where stack decisions live.

Key entities

Section titled “Key entities”
  • Metric — definition: name, owner, computation, refresh cadence.
  • Dashboard — collection of metrics with cuts.
  • Cohort (shared with 3.L).
  • Experiment — A/B or champion-challenger.
  • Anomaly — detected by data-quality / fraud / model monitor.

Key workflows

Section titled “Key workflows”
  1. Daily metric refresh — warehouse ETL → metric layer → dashboards.
  2. Anomaly alert — detector → notification → triage queue.
  3. Champion-challenger run — config experiment → shadow execute → compare → decide.
  4. Cohort profitability snapshot — monthly.

Integrations

Section titled “Integrations”
  • Internal data warehouse as source of truth.
  • BI tool — Metabase / Superset / Power BI / Looker / Tableau.
  • Model registry + monitoring (when ML in production).
  • Alerting — Slack, email, on-call.

APIs

Section titled “APIs”
  • GET /analytics/dashboards/{id} — render.
  • GET /analytics/metrics/{id} — query.
  • POST /analytics/experiments — define.
  • GET /analytics/anomalies — list.
  • POST /analytics/anomalies/{id}/dispose.

Events emitted

Section titled “Events emitted”
  • dashboard.refreshed
  • anomaly.detected
  • experiment.completed
  • metric.threshold.breached

Edge cases

Section titled “Edge cases”
  • Metric backfill after data correction.
  • Cut drill-down producing empty buckets — minimum population.
  • Stale dashboard with no notification — freshness watermark visible.
  • Sensitive PII in dashboards — masking by role.

Compliance touchpoints

Section titled “Compliance touchpoints”
  • DPDP — analytics must respect purpose limitation.
  • Sensitive data — masked for non-need-to-know users.
  • Borrower decision explainability — supports FPC obligations.

MVP vs production

Section titled “MVP vs production”
FeatureMVPProduction
Operational dashboards (Metabase / Superset)✓✓
Sales / pipeline✓✓
Risk dashboard (basic)✓Rich, drill-down
Collections dashboard✓✓
Unit-economics dashboard(Phase 2)✓
Cohort profitability(Phase 3)✓
Fraud analyticsBasicMulti-signal
Model monitoring(Phase 4)✓
Champion-challenger(Phase 4)✓
Anomaly detection(Phase 3)✓
Self-serve analytics(Phase 3)✓

Related: 3.L Portfolio monitoring, 5. Architecture, 14. Tech stack, 13.19 Analytics backlog.