Sabarmati Small Finance Bank: Should the Zero-MDR Merchant Vertical Be Scaled or Shut?
Sabarmati Small Finance Bank runs 340 branches, 6.2 million customers and 480,000 UPI merchant acceptors. UPI person-to-merchant transactions carry zero MDR by regulation, so the merchant vertical earns nothing on payments; it costs ₹180 to onboard a merchant and only 38% are monthly active. Two years ago the bank built KiranaOS, an inventory and khata app, and piloted credit plus subscription with 12,000 merchants: an average ₹35,000 line at 22% APR, a ₹99 monthly subscription, and a 3.8% credit loss rate. The pilot's gross revenue works out to roughly ₹40 per active merchant per month against a servicing cost of about ₹55. The chief digital officer wants ₹90 crore to scale to 1.5 million merchants. The CFO wants the vertical shut, which saves ₹46 crore of annual run cost. The board notes that two competitors are already embedding credit at the point of sale through the same merchant base.
1. Test the merchant business model on unit economics rather than growth metrics. What would have to be true for it to work?
Read the worked answer
Current unit economics (per active merchant per month). Revenue ≈ ₹40 (mostly interest on a small drawn book); servicing cost ≈ ₹55; contribution ≈ −₹15 before onboarding cost of ₹180 spread over a short life. The model loses money on every active merchant, so the 480,000 base is a cost, not an asset.
Only four levers can flip the sign.
- Take rate: interest is capped by the ₹35,000 ticket; the realistic jump is a higher share of merchants drawing, not a higher APR (22% is already near the tolerable ceiling for kirana working capital).
- Servicing cost: ₹55 is branch-assisted cost. Self-service repayment, UPI autopay and automated collections must push it below ₹25; this is the lever most within the bank's control.
- Attach rate: subscription at ₹99 only works if KiranaOS replaces something the merchant pays for already - distributors' billing, a ledger book, a separate POS. Without that, churn kills it.
- Credit loss: 3.8% is acceptable only if the yield holds; at 22% APR with short tenors, a rise to 6% wipes out the contribution.
What must be true. Attach rate above 25% of active merchants, servicing cost under ₹25, loss under 4.5%, and a second revenue line (distributor-side monetisation, co-branded supply credit) contributing at least ₹15 per merchant. At ₹90 crore for 1.5 million merchants, the bank would be funding a negative-contribution engine at scale.
Diagnosis. The asset is real - a merchant graph and transaction data - but the current model monetises it through a product that is too small to carry the cost.
2. Design the scale-or-shut decision: pilot structure, success gates, and how you would avoid scaling on vanity metrics.
Read the worked answer
Decision rule. Do not approve ₹90 crore. Approve a capped ₹12 crore, two-quarter experiment with pre-agreed gates; ₹46 crore of run cost is not saved by hope and not spent on a hunch.
Pilot design.
- 40,000 merchants in four clusters, randomised against a control group, so onboarding and training effects can be separated from selection bias.
- Segments defined by monthly GMV and distributor dependence, not by geography alone.
- Two arms: credit-only, and credit + KiranaOS subscription with a distributor-billing hook.
Gates at month six (all must pass).
- Contribution per active merchant ≥ ₹0 (not revenue - contribution).
- Attach rate ≥ 25%; monthly active ≥ 55% (from 38%).
- Servicing cost ≤ ₹30 and falling with volume; onboarding cost ≤ ₹110.
- 90-day delinquency ≤ 3.5%; fraud ≤ 0.5% of disbursements.
- Cohort retention at month 3 ≥ 85%.
Avoiding vanity metrics. The trap is counting onboarded merchants, GMV or app downloads. Report cohort contribution margin, payback month, and active-rate decay curve; force the CDO to present the worst cohort each month. Insist on a counterfactual: what would these merchants' credit behaviour have been without the app?
Kill criteria. If contribution is still negative at month six with servicing above ₹40 and attach below 15%, shut the credit line, retain the data asset, and re-partner rather than build - the merchant graph can be monetised through a distributor-side or insurer-side product with a partner's balance sheet.
Why this is the judgement. The board is being offered a scale decision where it should be offered a learning decision; the option value sits in the data, the cost sits in the credit.