6.14 Thin-file underwriting — when formal data is sparse

Why this matters

A meaningful share — by some industry estimates over half — of India’s SME credit demand sits in thin-file territory: borrowers without clean, recent, complete formal data. The shapes vary:

• A roadside chemist with a current account and inconsistent GST filings, no Tally, no commercial bureau record on the entity, but 8 years of business vintage on a single anchor manufacturer’s records.
• A garment manufacturer whose bookkeeping is in a paper ledger, whose proprietor’s CIBIL is clean but whose entity has no commercial bureau footprint.
• A new e-commerce seller with 9 months of strong marketplace settlements but no GST history (below threshold) and no bank-statement history (new current account).
• A second-generation distributor running on the family’s reputation, no formal accounting separation between business and household.
• A services SME billing one corporate client for 3 years consistently, but the bank statements show only quarterly billings — not the rhythm classic SME underwriting expects.

These borrowers are not unbankable. They have repayment capacity and intent; what they don’t have is the data shape the standard scorecard was built for. A platform that underwrites them well captures a structural advantage; one that declines them out of hand cedes the wedge to lenders with more pragmatic frameworks.

This page sets the philosophy and decision framework. The sub-pages below catalogue unconventional data sources (6.15), reference and field-verification protocols (6.16), graduated lending (6.17), anchor-backed thin-file (6.18), UPI / POS transaction-based (6.19), and marketplace-seller (6.20) underwriting.

The thin-file taxonomy

Different kinds of “thin” demand different responses. Classify before underwriting.

Pattern	What’s missing	What’s present	Underwriting route
Tax-thin	GST not registered (sub-₹40 lakh turnover for services / ₹40 lakh – ₹1.5 Cr for goods depending on state)	Bank + bureau + business activity verifiable	UPI / POS / transactional underwriting + field; lower ticket
Bank-thin	New current account; <6 months of statements	GST + Tally + bureau may be fine	Borrower’s existing personal account if business is owner-operated; AA pull; anchor data
Bureau-thin (commercial)	Entity is unrated commercially; no commercial bureau footprint	Promoter’s consumer bureau is fine	Promoter consumer bureau + cash-flow data + reference checks
Bureau-thin (consumer)	Promoter has no prior credit history (new-to-credit)	Business has GST + bank	Cash-flow-based; reference checks; lower starting limit
Accounting-thin	No Tally / Zoho; paper or memory-based bookkeeping	Bank + GST adequate	Bank + GST + field verification only; CA-assisted basic P&L reconstruction
Both-thin (worst case)	Multiple data sources missing	Anchor relationship; references; field-verifiable assets	Anchor-backed or graduated lending only
Vintage-thin	Business < 24 months old; no historical pattern	Other signals strong	C-grade only; lower ticket; shorter tenure; faster monitoring

Decision framework for thin-file

                  Application received
                          │
                          ▼
        Run standard data ingestion (bureau, GST, BSA, AA, MCA)
                          │
                          ▼
            Data sufficiency score (0-100)
                          │
            ┌─────────────┴─────────────┐
            │                           │
       Score >= 70                Score < 70
       (standard path)            (thin-file path)
            │                           │
            ▼                           ▼
       Run standard            Identify thin pattern
       scorecard               (taxonomy above)
                                       │
                                       ▼
                            Apply thin-file scorecard
                            with unconventional data
                                       │
                                       ▼
                            Field verification +
                            reference checks mandatory
                                       │
                                       ▼
                            Smaller ticket + shorter tenure
                            + higher rate + tighter monitoring
                                       │
                                       ▼
                            Decision: APPROVE / REFER / DECLINE

The branching from the data sufficiency score is the central architectural choice. The platform must:

1. Compute a sufficiency score based on which data sources returned usable signal vs which returned nothing or stale data.
2. Route automatically to the appropriate scorecard.
3. Surface the thin pattern explicitly to the analyst so they know what extra diligence applies.
4. Apply appropriate risk premium in pricing and tenure / ticket caps.

Principles for thin-file decisions

1. Triangulate, never trust single source

In standard underwriting, GST + bank statement + bureau triangulate each other. In thin-file, every signal must be cross-checked against at least two independent sources because any single source could be unreliable (no audit trail).

Example: borrower claims ₹50 lakh monthly turnover. Standard would verify against GST 3B + bank credit. Thin-file: must triangulate UPI receipts + supplier confirmations + field observation of stock turnover + electricity consumption.

2. Start small, observe, grow

Graduated lending (6.17) is the safest thin-file pattern. First loan small (₹2 – 5 lakh), short tenure (30 – 60 days), tight monitoring. Internal performance becomes the most reliable signal for limit step-up.

3. Anchor over reference; reference over claim

The hierarchy of trust:

1. Anchor-confirmed transaction history (manufacturer / large corporate confirms 18 months of borrower’s purchases) — highest.
2. Verified reference (specific named customer / supplier confirmed in person or call-recorded) — high.
3. Borrower-declared claim with corroboration (Tally + photo of inventory + electricity bill consistent) — medium.
4. Borrower-declared claim alone — only sufficient with multiple corroborating signals.

4. Visit the business

Field verification is mandatory for thin-file approvals — not optional. The platform’s field-agent app must support thin-file-specific questionnaires that go beyond “address verified” to capture stock visible, employees seen, business activity observed, customer footfall, neighbour conformance, signage, vintage of fit-out.

5. Tighter monitoring, faster escalation

Thin-file borrowers may have legitimate cash-flow lumpiness that triggers false EWS in standard scorecards. The platform must:

• Apply thin-file-specific EWS thresholds.
• Triage thin-file alerts in a separate, faster queue.
• Escalate to field re-verification on first material drift rather than waiting for SMA-1.

6. Pricing reflects effort + risk

Thin-file underwriting is operationally expensive (field, references, anchor coordination) and credit-risk higher. Pricing premium of 100 – 250 bps over standard A-grade rates is normal. Borrower understands the premium reflects the work and the risk; the alternative is no credit at all.

What thin-file borrowers are NOT eligible for

• Large tickets at first loan (> ₹10 lakh typically blocked).
• Long tenures (> 90 days per draw) until track record exists.
• Standard-grade pricing.
• Auto-approval — every thin-file decision involves manual review.
• Co-lending pools that the partner has not pre-agreed to take thin-file (most banks insist on standard-grade for PSL claims).

What unconventional data sources help

Detailed catalogue at 6.15 Unconventional data sources. Headlines:

• UPI transaction data via AA (becoming richer as AA coverage of UPI grows).
• POS settlement data (Razorpay POS, mPOS, Pine Labs, Innoviti).
• Marketplace settlements (Amazon, Flipkart, Meesho, Swiggy / Zomato for F&B).
• GST e-invoice IRN verification (even for borrowers below the GST threshold who deal with above-threshold buyers).
• Wholesaler / anchor purchase confirmations (signed letter or anchor’s own portal).
• Logistics provider data (Shiprocket / Delhivery seller dashboards; e-way bill activity).
• Utility bills (electricity, gas) — proxies for plant / business activity.
• Property tax records — geographic asset signal.
• PoS / banking via bank passbook OCR for borrowers without netbanking.
• Photographic evidence of inventory, signage, factory floor.
• Reference calls with structured scripts (3 customers + 3 suppliers, with consent).
• SHG / JLG history for very small entities with microfinance lineage.
• Trade-credit history with named suppliers (verified).
• Footfall data mobile-derived for retail / hospitality.
• Social media presence — business pages, customer reviews, longevity.

Where this lives in the platform

• Ingestion module (3.D) adds adapters for unconventional sources.
• Decision engine (3.E) has a data-sufficiency check and routes to thin-file rule set when triggered.
• Manual review workflow (3.F) has a thin-file queue with extra-context display (field photos, anchor letters, reference scripts).
• Field agent app (4.10) supports thin-file-specific questionnaires.
• Portfolio monitoring (3.L) has tighter EWS thresholds for thin-file cohorts.

What can go wrong

• Adverse selection — thin-file is a magnet for borrowers who can’t get standard credit. Filter rigorously; don’t lower bars.
• Operational expense — thin-file underwriting is 3 – 5× the per-loan operational cost. Pricing must reflect.
• Manual overrides accumulate quietly — track approval rate and outcome by analyst; an analyst always approving thin-file is a red flag for either policy drift or pressure.
• Reference fraud — borrower coaches references; mitigate with surprise calls + recorded conversations.
• Anchor letter forgery — verify directly with anchor or via anchor portal where available.
• Cohort-level credit cost spikes — monitor thin-file cohorts separately; act fast on deterioration.

Outcomes that justify the approach

When done well, thin-file lending:

• Reaches ~30 – 40% of borrowers who would otherwise be declined.
• Shows credit cost only marginally higher (+100 – 200 bps) than standard A-grade — manageable with pricing premium.
• Builds repeat-borrower relationships at a stage when competitors haven’t reached.
• Creates anchor and channel partnerships (anchors and CAs love a lender that takes their referral seriously).
• Establishes the platform as MSME-genuine — important for regulator perception and bank co-lending appetite.

When done poorly, it sinks the book. Discipline is everything.

Related

• 6.15 Unconventional data sources — catalogue with technical detail.
• 6.16 References and field validation — protocols + scripts.
• 6.17 Graduated lending — start-small-grow-with-track-record.
• 6.18 Anchor-backed thin-file — distinct from generic SCF.
• 6.19 UPI / POS transaction underwriting — for cash-business borrowers.
• 6.20 Marketplace-seller deep dive — beyond the segment overlay.
• 16. Client diligence — the cross-cutting diligence playbook.