“Judge the Forehead, Then Lend the Money” – Relearning Risk in the Age of AI

A Risk Philosophy for a Digital Credit World

In my early years at Sundaram Finance, my seniors and mentors taught me a line I have never forgotten:

“Neththiya Paarthu Panam Kudukaradhu.”

Judge the forehead — then lend the money.

It was not poetry. It was philosophy — a risk doctrine wrapped in everyday language.

They meant:

• Numbers tell the story of the past

• Documents reveal the present

• Character decides the future

A young product manager from a rising fintech once told me proudly:

“Our algorithm is so strong, we can approve a loan in 30 seconds.”

I asked him, “Great… but who are you approving?”

He smiled wide — “People with clean files and great scores!”

I replied — “Those people usually don’t need a loan.”

Real lending begins where data becomes incomplete,bureau reports show history,and judgment — not speed — decides eligibility.

Because technology can reduce effort… but it cannot replace wisdom.

And this is exactly where India’s digital lending story stands today.

Today our industry has scorecards that approve loans in 30 seconds, data aggregators that read financial history in a minute, and AI that predicts behaviour before the first EMI.

And yet, the most important question in lending remains the same:

How will this borrower behave when things go wrong?

That answer has never existed in datasets. It has always lived in judgement.

Why ‘Man in the Loop’ Still Matters

Risk doesn’t interpret itself.

Data aggregation can tell you:

• what happened

• when it happened

• how often it happened

  
  

But only humans can answer why —and whether the past is a predictor of the future.

Algorithms can:

• fetch bank statements, GST, bureau files

• score patterns within seconds

• flag anomalies statistically

But only bankers can:

• calibrate risk appetite to reality

• ensure profile–product fitment

• translate score → structure

• understand intent behind repayment

• prevent good borrowers from becoming bad loans

Automated approvals may be fast. Automated understanding is dangerous.

True digital lending is not machine vs man —it is machine accelerates, man decides.

Digital lending succeeds when technology accelerates and humans safeguard.

That balance is exactly what RBI is trying to enforce.

RBI’s View: Speed is Good — Accountability is Better

That’s why RBI stepped in. Not to slow down innovation —but to anchor responsibility.

Here is the regulator’s logic in one straight line:

RBI Focus	Meaning	Why
Who owns the customer?	The Regulated Entity (Bank/NBFC)	Accountability cannot be outsourced
Who handles the money?	Direct bank → customer, and back	Control of funds = control of risk
Who handles the data?	Only with explicit informed consent	Privacy has dignity
Who sets the price?	Full upfront disclosure	No hidden lending
Who decides credit?	Bank/NBFC — not the app	Judgment matters
Who supports convenience?	LSPs help, but don’t influence risk	Sales ≠ underwriting
Who resolves problems?	Clearly visible channel to the bank	No confusion in distress

Simply put:

If technology makes lending faster, responsibility must travel at the same speed.

Why ‘Man in the Loop’ Still Matters

Risk doesn’t interpret itself.

Data aggregation can tell you:

• what happened

• when it happened

• how often it happened

But only humans can answer why —and whether the past is a predictor of the future.

Algorithms can:

• fetch bank statements, GST, bureau files

• score patterns within seconds

• flag anomalies statistically

But only bankers can:

• calibrate risk appetite to reality

• ensure profile–product fitment

• translate score → structure

• understand intent behind repayment

• prevent good borrowers from becoming bad loans

Automated approvals may be fast. Automated understanding is dangerous.

True digital lending is not machine vs man —it is machine accelerates, man decides.

If technology makes lending faster,responsibility must travel at the same speed.

Case Study: What One Card Teaches Us

One Card is a digital-first co-branded credit card —tech-stack by the fintech, card issued by a partner bank.

At its best, it showed:

• seamless onboarding

• instant virtual card issues

• strong app-centric control

• millennial-friendly UI

But it also revealed the fragility of ignoring data governance.

In 2024, some partner banks stopped issuing new co-branded cards after RBI flagged that One Card’s tech stack had too much access to customer transaction data — something only the issuing bank can lawfully control.

The regulator’s concern was fundamental:

If the fintech controls the decision engine + data + infrastructure, then who is truly lending? And who is accountable when things go wrong?

One Card has since worked with banks to restructure the model and restore compliance.

But it stands as a cautionary tale:

🏁 Speed delivered.

⛔ Ownership diluted.

⚠️ Accountability blurred.

Whilst I know that One Card and BNPL operate in different segments, their stories have common paths and some lessons to offer.

🧩 Common Lessons from One Card & BNPL Crashes

1️⃣ Speed without governance creates fragility

Products that prioritise instant approvals and frictionless UX often skip essential affordability checks, and collapse when delinquency rises.

2️⃣ Algorithms cannot shoulder accountability

When the decision engine lies with the fintech stack — not the regulated lender — responsibility becomes blurred, prompting regulatory intervention.

3️⃣ Overlapping credit creates invisible debt

Multiple small digital credit lines, across apps/cards, accumulate silently without borrower awareness — until defaults surface.

4️⃣ Data control = Risk control

Regulators act swiftly when fintechs get deeper customer data access than permitted — because who controls data controls risk.

5️⃣ Convenience-first design can harm financial discipline

One-tap approvals and deferred-payment options train users to borrow casually, without understanding obligations.

6️⃣ If compliance is an afterthought, survival is at risk

BNPL shutdowns and issuance halts show that regulatory non-compliance can end the business overnight.

One sentence that unifies all these lessons

Digital lending fails when convenience outruns responsibility.

Exactly the scenario RBI wants to prevent.

Where Paper vs Ground Still Diverge

Even today, across many digital journeys:

• Disclosures scroll past faster than they can be read

• Consent becomes a design trick — “one-tap accept”

• LSPs chase volume faster than banks track risk

• Dashboards measure speed, not understanding

• Collections come before comprehension

Convenience, if not checked, can quietly turn into complacency.

The Real Digital Lending Transformation

It’s not:

• AI scorecards

• fancy UI

• one-minute onboarding

• automated TAT dashboards

It is:✔ calibrated risk✔ fit-for-purpose product structures✔ transparent charges✔ human judgment where needed✔ clear accountability every step of the way

Digital lending succeeds when technology accelerates and humans safeguard.

One Final Thought

We must stop asking:

“How fast can we lend?”

The better question is:

“How responsibly can we lend at speed?”

Because banking is not a sprint. It is a judgment business — always has been, always will be.

Now your turn

Which part of the guideline do you think is least understood in the ecosystem?

👇 Comment with your thoughts — let’s decode this together.