BFSI Case Study - 95% Faster NPA Evaluation with Colrows

Before

Months of work to bid on a portfolio.

Retail NPA portfolio evaluation took months - constraining competitive bidding speed. The process required manual segmentation, regulatory compliance checks across SARFAESI and DRT, and analysis spanning legal, technical, field-ops, and executive disciplines. Analyst subjectivity introduced inconsistency. Key-person risk compounded with every senior departure.

With Colrows

Semantic reasoning, end to end.

A semantic reasoning layer combining ML for bulk scoring with a knowledge graph for regulatory interpretation and edge-case handling. The AI agent synthesises historical data, legal heuristics, and institutional knowledge - producing an auditable reasoning trail behind every bid rationale.

01 · The challenge

Speed is the only moat.

In India's asset reconstruction market, the window to acquire a distressed retail NPA portfolio is brutally narrow. The firm that bids first - and bids accurately - wins. Every day spent evaluating a portfolio is a day a competitor could pre-empt the deal.

The institution had assembled a formidable multi-disciplinary team: technologists, legal specialists, field investigators, and senior executives. But the evaluation process was fundamentally bottlenecked by human bandwidth and sequential workflows - manually segmenting thousands of accounts, assessing borrower behavioural traits, reviewing KYC and loan documentation, and interpreting an evolving regulatory landscape before arriving at an indicative valuation.

“The process was sound. The people were excellent. But the market doesn't wait for excellent processes.”

01

Evaluation latency

End-to-end portfolio assessment took several months, constraining the firm's ability to bid competitively in fast-moving auctions.

02

Regulatory complexity

Correct valuation demanded fluency in RBI SARFAESI, DRT frameworks, and overlapping sub-regulations - expertise that varied across analysts.

03

Analyst subjectivity

Segmentation logic and file interpretation were anchored to individual analysts' experience - creating inconsistent outputs and key-person risk.

04

Document intensity

Each borrower file contained layered KYC, payment history, field reports, and behavioural indicators - requiring skilled human review throughout.

02 · Why not just train a model

The problem isn't data. It's interpretation.

The obvious question is whether a machine learning model trained on historical NPA outcomes would solve the problem. It's a legitimate hypothesis, and one the institution examined carefully. The answer reveals something important about the nature of the problem itself.

ML model on historical data

Pattern recognition from past outcomes

✓

Strong on bulk scoring

Predicts recovery probability for borrower profiles it has seen before - fast and scalable for routine accounts.

✓

Good at pattern flagging

Surfaces accounts that resemble past defaulters or high-recovery profiles with high throughput.

×

Blind to regulatory change

A model trained on 2024 data doesn't know a 2026 RBI circular changed enforcement thresholds. It reasons from an outdated legal reality - confidently.

×

Survivorship bias baked in

Training data reflects portfolios the firm bid on and won - not the full evaluation universe. Edge cases and passed deals are invisible.

×

Weakest on edge cases

The accounts that determine bid profitability - atypical collateral, complex DRT histories, contested SARFAESI enforcements - are precisely where frequency-based models fail.

×

Cannot explain its reasoning

An investment committee bidding hundreds of crores cannot accept “the model says 34 paise on the rupee.” A rationale is required, not a score.

Colrows semantic layer + AI agent

Reasoning anchored to encoded knowledge

✓

Bulk scoring via trained agent

Historical data still powers pattern recognition - but anchored to the semantic graph, not free-floating.

✓

Regulatory graph, always current

SARFAESI and DRT provisions are modeled as nodes and relationships. Update the graph once; every subsequent evaluation inherits the change instantly.

✓

Explicit handling of edge cases

Unusual accounts are evaluated against the full regulatory and institutional knowledge graph - not extrapolated from statistical frequency.

✓

Traceable, auditable reasoning

Every valuation output carries a reasoning trail - which provisions applied, which borrower signals were weighted, which heuristics drove the final figure.

✓

Investment-committee ready

Outputs are structured as defensible rationales, not probability scores - designed to survive scrutiny at the bid approval stage.

✓

Institutional knowledge, not just data

Legal heuristics, executive judgment, and field-team expertise are encoded explicitly - not hoped to emerge from training data.

The fundamental insight: NPA evaluation is not primarily a data problem - the data existed and was accessible. The bottleneck was interpretation: knowing which regulatory provision applied, what a payment pattern signalled about borrower intent, how to weight collateral quality against recovery probability in a DRT proceeding. Interpretation requires reasoning, not pattern matching.

03 · The solution

Institutional knowledge, encoded.

Colrows was deployed as a semantic reasoning layer - not a rules engine, not a dashboard, not a conventional AI model. The regulatory graph, tribal knowledge, and a trained AI agent work as a unified system. Each component does what it does best.

ML component

What it handles

→

Bulk account scoring against historical recovery patterns.

→

Borrower behavioural trait classification at scale.

→

Payment pattern anomaly detection across thousands of accounts.

→

Collateral liquidation timeline estimation from historical DRT durations.

→

Stratified sampling selection for portfolio representativeness.

Semantic graph

What it handles

→

Current regulatory standing of each account under SARFAESI and DRT.

→

Edge-case interpretation where historical precedent is thin or absent.

→

Legal enforceability assessment against latest RBI circulars.

→

Cross-referencing executive and legal heuristics for unusual account types.

→

Generating the auditable reasoning trail for bid rationale documentation.

The synthesis: the AI agent sits above both layers - using the ML component for speed and scale on routine accounts, and the semantic graph for correctness and explainability on complex ones. Neither alone is sufficient. Together, they cover the full evaluation surface.

AI agent layer

NPA evaluation agent Portfolio pricing engine Risk segmentation module

Semantic layer

Colrows knowledge graph Regulatory ontology Tribal knowledge models Behavioural signal library

Knowledge sources

RBI SARFAESI DRT framework Legal SME Tech analyst playbooks Field investigation protocols Executive heuristics

Data inputs

Loan files KYC documents Payment history Borrower behavioural data Historical portfolio outcomes

01

Portfolio ingestion

The agent ingests the raw portfolio and pre-classifies accounts against the semantic parameter graph.

02

Graph-anchored segmentation

Accounts are segmented by regulatory standing, recovery probability, borrower profile, and collateral quality - at scale.

03

Sampling & deep analysis

The agent selects representative samples and performs deep file analysis, surfacing behavioural traits and risk indicators.

04

Indicative valuation

Drawing on historical outcomes and the regulatory model, the agent derives a valuation with a full reasoning trail.

04 · The hidden risk

Regulatory drift and the cost of static models.

In NPA markets, the firms that consistently overpay are almost never the ones with bad analysts. They are the ones whose evaluation frameworks have quietly fallen behind the regulatory environment. Regulatory drift is silent, accumulative, and expensive - and it is the specific failure mode that a purely data-trained model cannot detect.

Risk

RBI circular amends SARFAESI enforcement threshold

A model trained before the amendment keeps applying the old threshold to every account. Entire account categories shift legal enforceability status - the model doesn't know. Valuations drift from reality, silently.

Risk

DRT procedural update changes timeline assumptions

Historical recovery timelines underpin ML probability estimates. When DRT processing times shift, every timeline-dependent valuation becomes systematically miscalibrated - with no way to surface this to the analyst.

Colrows

Semantic graph updated once, agent inherits immediately

When a regulatory change occurs, the Colrows graph is updated at the node and relationship level. Every subsequent evaluation - that day, that hour - reflects the new reality. No retraining cycle. No lag. No silent miscalibration compounding across a portfolio.

05 · Explainability

The bid rationale problem.

In asset reconstruction, a valuation that cannot be explained is a valuation that cannot be defended - to an investment committee, to a regulator, or to a counterparty challenging the bid. Explainability is not a nice-to-have. It is a governance requirement. The Colrows agent produces structured reasoning outputs alongside every valuation.

Sample account evaluation output

Illustrative · Account #NPA-2847

Account profile & regulatory standing

Account typeRetail secured - home loan, LAP

Outstanding₹18.4L principal + ₹6.2L interest

SARFAESIEligible. Notice served, possession not taken. Section 13(4) applicable.

DRT statusNo active proceedings. Clean enforcement path.

CollateralRegistered urban property. Circle rate ₹22.1L. Legal title clear.

Behavioural signalPartial payments post-NPA classification - intent to settle flagged by agent.

Reasoning trail & indicative valuation

Recovery pathSARFAESI possession → auction. Estimated 8-11 months per DRT duration model.

Regulatory riskLow. No sub-judice complications. Enforcement risk: minimal.

Heuristic appliedLegal SME rule: partial payment post-NPA reduces haircut by 8-12% vs. silent account.

ML signalProfile similarity to historical high-recovery cohort: 74th percentile.

Indicative value₹14.8L-₹16.2L (58-63 paise on total outstanding)

Reasoning: clean legal standing, settlement intent, and strong collateral coverage support above-average recovery. Colrows recommends inclusion in bid at upper-band pricing.

What changed in production.

>95% Reduction in evaluation cycle time (months → hours)

100% Regulatory coverage - RBI SARFAESI & DRT modeled

4× Expert disciplines unified: legal, technical, field, executive

100% Auditable reasoning trail behind every bid rationale

“The investment committee didn't ask us to justify the number. For the first time, the number came with its own justification - traceable to regulation, to precedent, to the account itself.”
Post-deployment reflection

06 · Outcomes

Evaluation as competitive advantage.

Hours

End-to-end evaluation turnaround

What previously consumed multiple months now completes within hours - enabling the firm to bid on portfolios previously logistically inaccessible.

Zero

Regulatory blind spots

Every account evaluation is automatically cross-referenced against the full SARFAESI and DRT regulatory surface - not just what one analyst happened to know.

4 → 1

Expert disciplines unified

Legal, technical, field, and executive knowledge - previously siloed in individual experience - now operates as a single coherent reasoning system.

↑ Bids

Portfolio pipeline expansion

With evaluation latency eliminated, the firm participates in a larger share of auctions, including fast-moving deals that previously expired before assessment was complete.

07 · Why it matters

Why semantic reasoning, not just AI.

Conventional AI approaches can surface patterns. But they cannot reliably reason about why those patterns matter, or apply that reasoning consistently against a shifting regulatory backdrop.

Colrows constructs a semantic graph - a structured representation of the domain in which concepts, relationships, rules, and heuristics are explicit nodes and edges, not latent parameters. The AI agent is anchored to this graph, ensuring every inference is traceable, auditable, and grounded in the firm's own institutional reasoning.

The result is a system that combines the scale of ML with the correctness of structured legal and institutional knowledge - producing outputs that survive both market scrutiny and investment committee review.

Conventional AI vs. Colrows

Learns from data patterns	Reasons from encoded knowledge
Opaque inference	Traceable, auditable reasoning
Brittle to regulatory change	Graph updated, agent inherits
Black-box outputs	Explainable, bid-ready rationale

From months to hours: NPA portfolio evaluation with semantic AI.