Niti AI / Industries / NBFC lending

NBFC acquisition & collections · tracked to T+90 DPD

Your constraint map shows which stage is blocked. Nobody can prove why.

Niti is a revenue decision system for NBFC lending: it turns constraint maps into ranked acquisition and collections decisions — with causal root-cause analysis, confidence labels, and outcomes tracked to T+90 DPD. For founders, CBOs, and CROs running ₹50Cr–1,000Cr+ AUM field books, where a wrong root cause fails quietly for a full experiment cycle.

The post-mortem, five months late

A mid-market NBFC saw PAR30 rise in its North Zone. The diagnosis: PD officer quality. Six weeks of officer training followed. PAR30 didn't move. At T+90, the post-mortem found the real cause: a NACH failure from a local service disruption — a collections operations issue, not a credit quality one. The correct intervention happened five months after the original diagnosis.

Separately: three zone-head transitions hit the same geography in eighteen months. Each incoming zone head ran a variant of the same three experiments the last one already ran. The first zone head's findings were in a deck nobody could find. Each repeat cycle cost six months and ₹2-3Cr in suboptimal outcomes.

And a third team ran three simultaneous growth experiments — RM incentive redesign, new zone coverage, credit policy relaxation — none targeted at the actual bottleneck, which turned out to be underwriting TAT. Disbursements stayed flat for six months while three teams blamed each other's interventions.

What it costs

A wrong root cause is a full experiment cycle, wasted, with no record kept.

4-5 months
Average cycle time — intervention, T+90 DPD outcome, re-diagnosis — when the original root cause was wrong. One miss costs one full cycle.
₹3.2Cr/mo
Throughput blocked at the underwriting stage on one field-led NBFC's book — mistaken by leadership for an origination-volume problem.
14-20 mo
Average zone-head tenure. Every transition without a documented experiment registry resets institutional knowledge to zero.

LOS reports show counts. They don't show which stage is accumulating work-in-progress, or why. The gap isn't a reporting gap — it's an architectural one, and no dashboard closes it.

What changes

Constraint, cause, experiment — one closed loop.

"Which stage is actually the bottleneck?"

The constraint map

Identifies the active pipeline bottleneck within 14 days of data ingestion, with a ₹/week throughput number attached. Not a dashboard of stage counts — the one stage actually throttling performing loan output right now.

"Why is it blocked?"

5 WHYs causal analysis — three tiers, one label

Tier 1 is a rule engine: CONFIRMED for known patterns. Tier 2 is a structural causal model: HIGH or MEDIUM CONFIDENCE with an attribution percentage. Tier 3 is ranked hypotheses: HYPOTHESIS — UNVERIFIED when the evidence genuinely isn't there. The system states which one it's giving you and stops rather than guessing — this is Hetu's three-tier engine, running underneath the pipeline data instead of underneath ad spend.

"How do we stop re-running the same failed test?"

The experiment outcome tracker

Every intervention pre-commits a success metric before it runs. Outcome tracked at T+30/60/90 DPD, against both a local metric (the targeted stage) and the system metric (performing disbursements). The loop closes, and re-triggers constraint detection automatically.

"What happens when a zone head leaves?"

An experiment registry that doesn't walk out the door

Every experiment, its root cause, and its T+90 outcome are logged permanently. The next zone head inherits the record instead of repeating the last one's three tests from scratch.

"Is this an NPA problem or an RM problem?"

NPA traced to cohort and origination stage

The 5 WHYs chain traces NPA concentration back through RM cohort and sourcing stage — not just to a portfolio-level number, but to the specific officers and the specific window where sourcing quality dropped.

"Is our incentive structure creating the NPA?"

Incentive-quality cross-check

Disbursement volume and NPA rate read together, per RM cohort — surfacing exactly when your highest-disbursement RMs are also your highest-NPA-rate RMs, which an incentive plan built for volume alone will never catch.

The gate that matters here

Hard gate · confidence before conclusion

A confident wrong narrative is worse than "we don't know yet."

The 5 WHYs engine will not manufacture certainty it doesn't have. When evidence is insufficient, it returns HYPOTHESIS — UNVERIFIED and names exactly what data would confirm it — instead of handing you a plausible story that sends you into another 90-day cycle chasing the wrong fix, the way the PD-officer-training misdiagnosis did.

Proof, not promise

Confirmed in Tier 1 for three of five triggered constraints.

On a design-partner deployment, the active constraint was identified within 14 days of pipeline data ingestion. The 5 WHYs chain resolved to a CONFIRMED root cause in Tier 1 — the rule engine, no statistical guesswork — for three of the five constraints it was triggered on.

  • ₹3.2Cr/month throughput blocked at the underwriting stage, quantified and attributed.
  • 2.8× — the performing-disbursement gap between top-quartile and median RM performance.
  • 73% of NPA from 18% of volume — traced to two specific RM cohorts, not a portfolio-wide average.
  • 3.4× — the NPA-rate gap between the highest-disbursement RM cohort and the lowest.
TriggerTier resolvedLabel
Underwriting TAT bottleneckTier 1Confirmed
Sourcing gap → 2 named ROs, 60 days priorTier 1Confirmed
NPA concentration → cohort attributionTier 2High confidence

Illustrative, from design-partner data. The RCA that traced a sourcing quality gap to two specific relationship officers, sixty days before it appeared in the NPA book, is the one worth asking to see live.

FAQ

NBFC lending — common questions

What decisions does Niti support for NBFCs?

Acquisition and collections interventions tied to a constraint map — where to act, why the stage is blocked, and whether the intervention worked by T+90 DPD.

How does the three-tier causal analysis work?

Tier 1 resolves confirmed patterns from your graph; Tier 2 attributes with confidence intervals; Tier 3 returns ranked hypotheses with falsification tests — never a fake root cause.

What is the 90-day pilot?

₹2L activation + ₹3.5L/mo, with an outcome guarantee tracked to T+90 DPD. The demo starts from one unexplained problem in your book.

How is this verified under the hood?

Confidence labels are computed by Hetu — the same decision-verification engine that runs Niti's D2C budget product.

Explore

Where to go next

Start here

Name one problem in your book nobody has explained to your satisfaction.

That's the demo. We run the three-tier causal analysis on it live: which tier resolves it, what confidence label it gets, and — if it comes back HYPOTHESIS — what data would move it to CONFIRMED. ₹2L activation, ₹3.5L a month, 90-day outcome guarantee tracked to T+90 DPD.

Bring us the unexplained problem