Working Capital Lending for India's Underserved Micro-SMEs: Productize, Agencify, or AI-Fy?

Who does it: Micro-SME lending is sourced by three types of humans today. First, relationship managers at banks and NBFCs who manually call on shops, kirana stores, and small factories — they spend 60–70% of time on paperwork, not selling. Second, commission agents and dalals who know local businesses and take home 1–3% of the sanctioned amount as a referral fee — they are the de facto credit officers for the bottom of the pyramid. Third, сами borrowers (themselves) who walk into bank branches with a folder of physical documents, wait for hours, and get rejected because their ITR is too low or their bank statements show cash deposits the banker doesn't understand.

What they use: Phone is WhatsApp for document sharing (photos of bank statements, GST returns, electricity bills). Excel for tracking eligibility, EMI schedules, and pipeline. Physical register books in regional languages at the branch level. CIBIL bureau pulls are standard at larger NBFCs but the report is unreadable to most micro-SME borrowers — they don't know their score or why it matters.

Where time and money leak: The single biggest leak is document collection. A micro-SME borrower has GST returns, 12 months of bank statements from potentially two banks, a shop Act license, and PAN — these live on four different screens or papers. The RM or agent assembles this manually. This takes 3–5 days of back-and-forth. The second leak is manual statement analysis — a human reads 12 months of transactions line by line to spot salary credits, loan EMIs, cash deposit patterns. This takes 2–4 hours per file. The third leak is rejection without feedback — a bank rejects 60% of micro loan applications and the borrower never learns why, so they don't fix it and come back with the same file in six months.

---

Who profits from it staying manual:

Bank branch managers — they control the queue. Faster processing means fewer appointments, less work, and a lower headline number for their "contacts processed" metric. Some managers also extract informal fees from brokers for faster processing; automation removes this lever.

Commission agents / dalals — they earn 1–3% per sanctioned loan. If a platform makes sourcing instant and self-serve, their income disappears. They actively resist digital onboarding.

Traditional NBFC RMs — same dynamic as bank managers. They are evaluated on sanctions, not on speed or borrower experience.

Who is hurt: Micro-SMEs pay the price. Working capital gaps force them to borrow from informal moneylenders at 24–60% annual rates. A ₹2 lakh loan from a moneylender at 3% monthly works out to ₹6,000 in interest per month — a crushing cost that consumes most of the margin on the inventory the capital was meant to finance.

NBFCs that want to scale — their cost-to-originate per loan stays high (₹3,000–8,000 per file in manual processing overhead), which means they can't profitably lend below ₹5 lakh. Below that threshold, the interest income doesn't cover the processing cost.

Who would pay to change it: NBFCs and smaller fintech lenders — if an agent or tool reduces their cost-to-originate below ₹1,500 per file and improves their hit rate (approvals that don't go delinquent), they would pay a per-file fee or a success fee. The NBFC sourcing problem is real: they have capital to deploy but can't find enough qualified borrowers efficiently.

B2B marketplaces and SaaS platforms — for example, a platform serving kirana stores (like a modern trade distributor or a Pos biz solution) could embed a working capital offer and earn a referral fee. They have transaction data but not the credit infrastructure.

---

What it does on day one:

An AI agent (call it the Credit Qualification Agent) that takes a micro-SME's GST login credentials and bank statement PDF, then produces in 90 seconds:

• A standardized credit one-pager (not a score — a narrative assessment in plain English and Hindi)
• Flagged risk signals (consecutive months of salary > revenue, heavy cash deposits, existing loan EMIs consuming >50% of inflows)
• A recommended ticket size and tenure range
• The three things the borrower must fix before reapplying (if not approved)

Who pays and how:

NBFCs pay per assessment — a SHAPE of ₹150–300 per qualified credit brief delivered to their sourcing team or integrated API. Not per loan. Per brief. This is a sourcing tool, not a lending product.

A qualified brief = a micro-SME who consented to share their data, whose GST and bank data was analyzed, whose report was delivered to an NBFC's dashboard or CRM.

The NBFC pays because the agent converts their manual 5-day sourcing cycle into a 90-second first-pass filter, and their RM now calls only borrowers who have a real chance.

What it does NOT do on day one: No loan underwriting, no credit decision, no lending, no disbursement. The agent is a qualified lead generator, not a lender. This sidesteps RBI licensing entirely on day one.

---

Verified players:

LendingKart — uses GST data and bank statement analysis for small business loans (₹1–75 lakh). Founded 2014. They build their own credit models. They are both lender and originator.

Aye Finance — uses a proprietary scoring model called FLAMES (Financial, Lending, Asset, Management Evaluation System). They do in-person field visits combined with digital data. Lender and servicer.

NeoGrowth — embedded lending at merchant outlets; uses POS and transactional data from retail businesses. Lender.

Klub — works with SMEs for revenue-based financing; uses bank and marketplace data.

Capital Boost (Capzo) — offers digital business loans with minimal documentation, uses GST and bank statements.

Smile Loan (iviIT) — small ticket digital loans for micro-SMEs, less verified on scale.

Account Aggregator ecosystem (RBI-mandated data sharing): Players like Finvu, CAMS, NESL operate AA infrastructure that allows a borrower to share financial data (bank statements, GST returns, pension records) with a lender digitally, with borrower consent. This infrastructure exists and is live as of 2025. The legal framework is in place. The adoption on the micro-SME side is still low because borrowers don't know it exists.

What is missing: A tool that sits between the AA / GST data and the NBFC, does the interpretation work, and sells the qualified brief. The gap is not in data infrastructure — it's in the human-readable translation layer between raw financial data and a credit decision.

---

Fact 1: Micro-SMEs don't want formal credit, they manage without it.

Check it cheaply: Spend ₹5,000 on a WhatsApp survey. Create a Google Form, share it in three Vizag business association WhatsApp groups (Vizag has 16K+ member networks through Alok's network). Ask 50 micro-SMEs: "Have you ever applied for a formal business loan? If no, why not?" If more than 40% say "too complicated / didn't think I'd qualify / no documents," the idea survives. If 60%+ say "don't need it / manage fine," kill it.

Fact 2: The cost to acquire one micro-SME GST + bank data pair exceeds ₹800.

Check it cheaply: Run a 10-day ad campaign on Meta with ₹3,000 budget targeting small business owners in one city. Offer a free "credit health check" in exchange for GSTIN and bank statement upload. Measure: cost per completed submission. If >₹800, the unit economics don't work at a ₹300 per brief price point without dramatically better distribution channels. The fix is to go through NBFCs who already have borrowers walking in the door — but only if those borrowers have GST and bank data ready.

Fact 3: NBFCs won't pay for qualified leads for loans below ₹3 lakh.

Check it cheaply: Talk to one relationship manager at a regional NBFC branch in Vizag. Ask: "If I could tell you tomorrow which 10 walk-in borrowers have a real chance of approval, would you pay for that?" Then ask what they'd pay. If the answer is under ₹200 per qualified lead, the math is too thin. If they say "our sourcing cost is ₹5,000 per sanctioned loan, we'd gladly pay ₹300 for a qualified brief," the model survives.

---

Total budget: ₹25,000

Month 1 (₹10,000):

Week 1–2:

• Set up a simple web form (Google Forms or a cheap Carrd page) that collects GSTIN, 12-month bank statement PDF upload, and one phone number.
• Use the RBI's Account Aggregator sandbox (or Finvu's test environment) to pull real GST return data for 10 volunteer businesses.
• Manually run a Python script (or even Excel) to produce the credit one-pager — no AI needed yet. This is to prove the output format is useful.
• Recruit 10 micro-SMEs from your own network (or Alok's contacts) as volunteers. No money changes hands. They get a free credit health check.

Week 3–4:

• Approach one NBFC RM in Vizag (or via the Vizag Startups network) and show them the credit briefs. Ask: "If I could produce 20 of these per week, would you use them and what would you pay?"
• Document their feedback verbatim. Adjust the output format based on what the RM actually finds useful.

Month 2 (₹8,000):

Week 5–6:

• Build the lightweight agent: use an LLM API (cost ~₹2,000 for 500 assessments at current API rates) to parse bank statements and GST returns into the standardized one-pager.
• The agent uses prompt engineering, not fine-tuning. It follows a strict output template so the format is consistent.
• Test on the 10 volunteer businesses. Verify the LLM output matches what the manual analysis produced.

Week 7–8:

• Run a targeted WhatsApp campaign in one Vizag business group: "Get your free business credit health check — takes 5 minutes." Use a shared Google Sheet to track respondents.
• Goal: 30 micro-SMEs complete the process.
• Each completes a short WhatsApp survey at the end: "Was this useful? Would you pay ₹99 for this? Would you share this with two friends?"

Month 3 (₹7,000):

Week 9–10:

• Deliver credit briefs to 2–3 NBFC RMs (or fintech lenders) and offer the first 20 for free in exchange for a review.
• Collect their scoring: Was the data accurate? Was the format useful? Would they pay?

Week 11–12:

• Count the numbers:

- How many briefs delivered? - How many NBFCs said they'd pay? - What's the actual cost per brief (API + your time at ₹500/hour)? - What's the minimum price that covers cost + margin?

Pass mark:

• 25 micro-SMEs complete the process (not just start it)
• 2+ NBFC RMs say they would pay ₹150–300 per brief
• Cost per brief under ₹250 (including your time at a conservative estimate)
• If all three are true: proceed to build the actual product
• If NBFCs won't pay: pivot to charging micro-SMEs directly (₹99–199 per brief)
• If cost per brief exceeds ₹400: the agent isn't efficient enough yet — go back to prompt engineering

---

AGENCIFY first, PRODUCTIZE later, AI-FY in layers.

Start as an agency because the real asset isn't software — it's the NBFC relationship and the borrower trust. An agent working as a credit qualification service learns what NBFCs actually want in a brief (the format, the signals they care about), learns what borrowers will share willingly, and builds a track record that can then be turned into software. The AI is already inside the agent from day one (LLM parsing); it becomes the product when the agency's playbook is codified and the NBFCs are paying enough to fund a developer. Do not raise capital, do not build a lender's tech stack, and do not try to disintermediate the broker until you have at least 3 NBFCs paying you every month.

Availability confirmed against the .in registry (RDAP) on 2026-09-22. Prices and ownership read from our own intelligence tables. Nothing here is estimated.

Single-word, available now

• kapitals.in — available
• workings.in — available
• kapital.co.in — available
• kapitals.co.in — available
• workings.co.in — available

Already ours

• kapital.in · parked, free to use

Also available (compound)

• mykapital.in
• gokapital.in
• kapitalhub.in
• kapitalmart.in
• kapitalmandi.in
• kapitalbazaar.in
• kapitaldirect.in
• kapitalsupply.in
• kapitalconnect.in

Taken and developed — do not chase

• kapital.com · entropy 5.83
• capital.com · entropy 5.22
• capitals.co.in · entropy 5.06
• kapitalkart.in · entropy 4.61
• capitalmart.in · entropy 6.22
• capitalkart.in · entropy 6.37

---

Generated 2026-09-22 10:38 UTC. Topic from our research queue; no market-size figure appears here unless a source is named. The domain block above is read from our own intelligence tables and confirmed at the .in registry (RDAP); the model wrote the analysis, not the domain facts.