6.16 References and field validation protocols

References and field verification are the qualitative spine of thin-file underwriting. Done right, they cost ₹500 – ₹1,500 per loan and catch most fraud while validating the business reality. Done badly, they are ticked off by tired analysts looking at low-effort confirmations and add nothing.

This page is the protocol.

When references and field are mandatory

Loan profile	Field FI?	Reference calls?
Standard A-grade with clean GST + Tally + bank	Optional	Optional
Standard B-grade	Often	Optional
Thin-file C/D-grade	Mandatory	Mandatory
First loan to any borrower at ticket `>= ₹25 lakh`	Mandatory	Mandatory if ticket `>= ₹35 lakh`
Anchor-introduced borrower without anchor’s formal channel-finance programme	Mandatory	Mandatory
Repeat borrower with no DPD and full data	Skip	Skip

The platform’s policy engine (3.E) sets these triggers automatically; the manual review queue (3.F) surfaces FI / reference completion as a pre-sanction gate.

Reference checks

Selection

Borrower provides a named reference list with consent for contact:

3 customers — by name, business, mobile, relationship type (regular buyer / occasional / large-account).
3 suppliers — by name, business, mobile, relationship type.
2 neighbours — adjacent shops / businesses, by name, mobile.
1 banker — branch manager / RM at the borrower’s primary bank.

For very small thin-file borrowers, 2 + 2 + 2 + 1 may suffice. For larger thin-file, demand 5 + 5 + 2 + 1.

Pre-screening

Before calling:

Verify reference’s mobile number is active (test SMS).
Cross-check reference’s GSTIN (if business) against borrower’s GSTR-1 (customer) or 2A (supplier). Mismatch is a flag — borrower may have listed a related-party rather than a real customer.
Search reference’s name / business for online presence — fictitious references often have no digital footprint.

Structured call script

For each customer reference:

1. (Identity confirmation) Are you Mr/Ms [name], proprietor/representative of [reference business name]?
2. (Relationship existence) Do you know [borrower business name], specifically [borrower's owner name]?
3. (Tenure) How long have you been buying from [borrower]?
4. (Volume) Roughly what's your monthly purchase value from them?
5. (Frequency) How often do you buy — weekly, monthly, on need?
6. (Payment terms) Do they extend credit to you? Cash on delivery? UPI?
7. (Quality) Are you satisfied with their product / service / delivery quality?
8. (Reputation) Would you recommend them to other businesses? Why?
9. (Anything concerning) Any concerns or issues you've experienced?

(Note: the call is recorded and will be retained for audit. Confirm verbal consent at start.)

For each supplier reference:

1. (Identity confirmation)
2. (Relationship existence)
3. (Tenure) How long have you been supplying [borrower]?
4. (Volume) Roughly what's their monthly purchase value from you?
5. (Credit history) Do you extend credit to them? Typical credit period?
6. (Payment behaviour) Do they pay you on time consistently? Any history of delays?
7. (Disputes) Any disputes or rejected goods incidents?
8. (Anything concerning) Any concerns?

(Recorded.)

For each neighbour reference:

1. (Identity confirmation)
2. (Knowledge) How long have you been at [neighbour business address]?
3. (Borrower observation) How long has [borrower] been operating from the adjacent / nearby premises?
4. (Activity observation) Is the business open regularly? Customer footfall reasonable?
5. (Owner observation) Is the owner usually present?
6. (Reputation) Anything you've heard about them — positive or negative — in the area?
7. (Anything concerning) Any issues you've observed?

(Recorded.)

For banker reference (optional, often skipped):

1. (Account confirmation) Does [borrower] hold an account at your branch?
2. (Vintage) For how long?
3. (Behaviour) Any cheque / NACH bounces in the past 12 months?
4. (Profile) Roughly what's their account turnover pattern?
5. (Recommendation) Would you recommend them as a borrower?

(Recorded; banker may decline to share specifics due to confidentiality — capture decline as data.)

What you’re listening for

Confidence in answers — references who hesitate or are vague suggest unfamiliarity.
Independent corroboration — multiple references describe the borrower consistently.
Volume / tenure consistency with borrower’s own claims.
Voluntary positive remarks — unprompted positives are stronger than yes-responses.
Any “concerns” — references often hedge; explicit concerns are unusual and weighty.

What’s a red flag

All references say identical phrases — coached references.
References cannot recall basic details (volume, frequency) about the borrower.
References declined to be contacted post-handover but consented at borrower’s prompting — coached.
Reference is also borrower’s relative or shares the same address.
Mobile number disconnects after first call — fake reference.
References describe a different business than what borrower claims (different location, different product).
Hostile references — disputes mentioned even guardedly.

Disposition

The credit analyst dispositions reference checks as:

CONFIRMED — all references corroborate; no flags.
PARTIAL — most references confirm; one or two not reachable or with minor inconsistency. Proceed with note.
FLAGGED — material concerns; either re-do with new references or DECLINE.

Field verification (FI) protocols

Standard questionnaire

The field-agent app (4.10) presents a structured form. Categories:

A. Identity confirmation

Borrower’s owner present at premises during visit (Yes / No).
Owner’s name, photo at premises captured.
Owner’s mobile number matches application — verify by calling on the spot.

B. Premises confirmation

Address matches application (Yes / No).
Signage visible: read out + photographed.
Signage vintage (paint condition, wear) — agent estimates.
Adjacent businesses identified for context.
Owned / rented (asked + supporting document if rented — rent receipt).

C. Operational confirmation

Business open and active at time of visit (Yes / No).
Customer footfall observed (if retail / hospitality) — quantitative if possible (> 5 customers in 30 minutes etc.).
Stock visible — photographed. Quantity / variety estimated.
Employees seen — count.
Equipment / machinery visible (for manufacturers) — photographed.

D. Business operations questionnaire (asked of owner)

Years of operation at this location.
Total years of business.
Primary products / services.
Main customer types (retail / wholesale / corporate).
Main suppliers — names of top 3.
Monthly sales rough estimate.
Number of employees.
Bank account used for business.

E. Cross-verification

GSTIN displayed (regulatory requirement for many SMEs) — photographed.
Other licences displayed (Drug Licence, FSSAI, factories) — photographed if applicable.
POS terminal / UPI QR present — photographed.

F. Neighbour conformance

Adjacent business owner asked: “How long has this business been here?” — agent records answer.
Any specific concerns from neighbour.

G. Final disposition

Agent’s overall assessment: business is real and active / questionable / suspect / fake.
Photographs uploaded (mandatory minimum 5: signboard, interior, owner, stock, neighbour view).
Geo-tag captured at arrival + departure.

Visit timing

Surprise visits preferred over scheduled — borrower may stage premises if forewarned.
For high-activity businesses (retail, restaurants), visit at peak hours; for offices / manufacturing, visit during working hours.
Repeat-borrower visits less critical but periodic (once a year minimum for active borrowers).

Visit follow-up

Reviewer disposition — internal reviewer (separate from agent) examines the visit report + photos + geo-tag and dispositions accept / reject / re-visit.
Reject reasons are documented (poor photos, geo-tag inconsistent, missing answers) — re-visit triggered.

Combining reference checks + FI into sanction decision

When both completed:

Reference outcome	FI outcome	Decision input
Confirmed	Confirmed	Strong positive
Confirmed	Partial / questionable	Refer; re-visit
Partial	Confirmed	Refer; re-check references with different selection
Partial	Partial	Likely decline; CRO override only
Flagged	Any	DECLINE
Any	Suspect / fake	DECLINE; internal blacklist update

Cost and time

Per loan:

Reference calls: 30 – 45 minutes of agent time; ₹200 – ₹500 cost.
Field visit: 60 – 90 minutes agent + 30 minutes review; ₹400 – ₹1,000 cost.
Total per thin-file loan: ₹600 – ₹1,500 and 2 – 4 hours of operational time.

Pricing premium of 100 – 250 bps over standard borrows easily covers this cost.

Recording and retention

All reference calls recorded with consent disclosure at start.
Recordings retained per record-retention (typically 5 – 7 years minimum).
Field visit photos + form + geo-tag stored in DMS, linked to loan, retrievable for audit.
Agent identity captured per visit; agent training certification verified at time of visit.

What the platform must enforce

Field-agent app cannot submit without all mandatory photos + geo-tag.
Reference call recordings must be attached to disposition.
Reviewer disposition required before sanction.
Agent’s active training status checked at visit time.
Compliance auto-QA samples 5 – 10% of calls + visits for review.

Fraud patterns to watch

Visit-stage borrowing of premises — borrower coordinates with neighbour to claim adjacent shop as own. Mitigate via independent neighbour ask: “Who occupies this shop?”
Phone-tree references — borrower hands phone to a coached person who answers in turn. Mitigate via independent dialing (don’t use a number forwarded by borrower).
Photo metadata stripping — borrower-supplied photos stripped of EXIF. Use only agent-captured photos with embedded metadata.
Multiple applications same agent — agent passing through too easily. Cross-check agent’s approval-to-decline ratio.
Reviewer fatigue — reviewer accepting weak FIs. Sample independent QA of reviewer decisions.

Cultural and linguistic nuances

India’s SME landscape is regionally diverse. The platform must:

Conduct calls and visits in the borrower’s preferred language.
Train field agents on regional / cultural conventions (e.g., visiting times, gender of agent for certain business types, holiday calendars).
Have agents who can read state-specific signage and documents.
Be aware that some businesses have legitimate seasonality patterns that may look like decline.

A standard scorecard tuned for a national norm can systematically under-grade specific regions; the field protocol is part of the corrective.

4.10 Field verification — vendor stack.
3.C KYC / KYB — KYC’s overlap with field FI.
6.14 Thin-file underwriting.
16. Client diligence — broader diligence framework.
17. Recovery deep-dive — recovery uses much of the same data.