AI That Impacts Business
When Sudipta Roy described AI as the force behind the fourth industrial revolution, it wasn’t rhetoric. For large NBFCs operating across rural, semi-urban, and urban India, AI is no longer an experimental layer it is becoming core infrastructure.
But what does “AI at scale” really mean inside a 40,000+ employee lending organization?
It doesn’t start with chatbots.
It starts with re-architecting the business.
- From Fragmented AI to AI-Native Architecture
One of the key observations was that many BFSI institutions approach AI in a fragmented way: a credit model here, a chatbot there, maybe a generative AI pilot on top.
That rarely delivers transformation.
Instead, the organization rebuilt itself into four intelligence stacks:
- Customer Intelligence – acquisition & field engagement
- Credit Intelligence – underwriting & risk engines
- Portfolio Intelligence – continuous monitoring
- Service Intelligence – collections & cross-sell
Each stack was rebuilt ground-up with AI as a base layer.
The philosophy was simple:
You cannot bolt AI onto legacy thinking. You must design for it.
- “It Takes a Village to Raise AI”
AI implementation at enterprise scale isn’t a data science project, it’s an organizational project.
When the flagship credit engine (Cyclops) was built, it required coordinated effort from:
- Decision science teams
- Credit & business teams
- Legal & compliance
- IT & InfoSec
- Model risk management
More than 50 models now run in parallel during underwriting, pulling data from 20+ APIs and processing decisions in seconds.
The lesson?
Without governance, observability, and cross-functional alignment, AI will not scale even if the models are powerful.
- Credit Intelligence: Reducing Risk at Entry
In lending, risk reduction begins at onboarding.
The AI-powered credit stack combines:
- Machine learning & ensemble models
- Fraud detection layers
- Geo-intelligence
- AML repositories
- Real-time API integrations
The impact has been measurable. In the two-wheeler portfolio, gross non-starters dropped significantly after successive model iterations.
But an important nuance emerged:
- Version 1 delivered improvement
- Version 2 delivered sharper precision
- Continuous iteration delivered real performance
AI modeling is not a one-time deployment. It is a cycle of learning and refinement.
- Portfolio Intelligence: Monitoring 28 Million Customers Autonomously
Traditional lenders are good at onboarding customers.
They are less effective at continuously monitoring them.
The automated portfolio engine (Nostradamus) was built to change that.
Instead of manual Excel reviews, the system now:
- Scans millions of accounts continuously
- Uses macro and micro economic models
- Detects behavioral anomalies early
- Runs stress-testing simulations
- Flags risky accounts into “attention trays”
For example, if crop prices fall in a specific district, the system can assess how that impacts borrowers in that micro-market.
Machines monitor in real time.
Humans intervene selectively.
That shift moves risk management from reactive to predictive.
- Underwriting with AI Co-Pilot: From Days to Hours
SME underwriting is document-heavy and time-consuming. Bureau reports can run into hundreds of pages. Bank statements require manual analysis. Human bias can creep in.
The AI co-pilot (Helios) now:
- Analyzes bureau, GST, and banking data
- Flags anomalies
- Highlights risk indicators
- Suggests personal discussion questions
- Generates structured credit assessment notes
The result?
- Earlier underwriting timelines: 2–6 days
- Current top-end SME underwriting: ~14 hours
- Target: sub-4-hour decisioning
AI here does not replace underwriters. It enhances their decision quality and speed.
- Collections Intelligence: When Economics Change
AI-powered voice collection calls cost a fraction of human-led calls.
While human collections may show slightly higher effectiveness, AI dramatically improves cost efficiency. Over time, as models improve:
- Routine follow-ups can be automated
- Human collectors can focus on complex cases
- Cost-to-income ratios improve structurally
In high-volume NBFC environments, this cost shift can directly impact profitability.
- Beyond Models: Building an AI Mindset
Perhaps the most powerful shift wasn’t technical, it was cultural.
AI training was extended to:
- Finance teams
- Legal teams
- Non-technology managers
Managers were required not just to learn AI, but to build AI-based tools.
In one instance, the legal team automated report analysis reducing a 3–4 hour task to 15 minutes.
When AI becomes organization-wide literacy rather than a tech function, transformation accelerates.
- AI as a Business Strategy
At scale, AI in NBFCs impacts four core levers:
- Productivity – faster underwriting & automation
- Growth – better targeting & credit precision
- Resilience – real-time monitoring & anomaly detection
- Cost Efficiency – optimized collections & operations
Lending profitability is a balance of:
- Risk
- Speed
- Cost
- Scale
AI influences all four simultaneously.
- The Bigger Picture
Generative AI and foundational models may capture headlines. But enterprise value does not come from “cute outputs.”
It comes from:
- Strong data governance
- AI-native architecture
- Continuous iteration
- Organization-wide adoption
The next 24–36 months in BFSI will not be about who experiments with AI. It will be about who institutionalizes it.
Because in large NBFCs: AI is no longer an innovation. It is infrastructure.
