पाठशाला Pathshala · हिसाब Hisāb, Unit economics · Lesson 06 · Start

Lending unit economics: yield, credit cost and the cost of funds

A lender does not earn the interest rate. It earns what is left after the money, the losses and the people are paid, usually single digits. The spread model, the NPA rules and how little room a startup has.

Pathshala, The Founder Library · 11 October 2026 · 10 min read

Lending looks like the simplest business in India. Borrow at one rate, lend at a higher one, keep the difference, and the demand is bottomless. It is also the business in which the revenue line is somebody else’s money going out of the door first, in which a cost that will not be known for a year is the largest one, and in which the margin that is left after everything is paid is thinner than in any other model in this track. This lesson is the arithmetic of that margin.

It covers the four lines of a loan book’s economics, yield, cost of funds, credit cost and operating cost, with the Reserve Bank of India’s rules on when a loan is bad; a benchmark from the country’s largest listed non-bank lender so that a startup can see what good looks like; a spread model to run your own book; and what changes if you are not the lender but the platform that brings borrowers to one.

The spread, in one line

Everything in lending is expressed as a share of assets under management, which is the money currently lent out. Yield is the interest actually earned on the book in a year, after waivers and after loans that stopped paying. Cost of funds is the interest paid on the money borrowed to lend. Yield less cost of funds is the net interest margin. From that come two more deductions: credit cost, meaning the provisions set aside and the loans written off in the year; and operating cost, meaning the people, the technology, the collections and the compliance. What is left is return on assets. Multiply it by one plus leverage, the ratio of borrowed money to the lender’s own, and you have return on equity, which is what the equity investor is buying.

The order matters because the costs arrive at different speeds. Cost of funds is known on the day the money is borrowed. Operating cost is known at the end of each month. Credit cost on a loan written today is not known until the loan has matured or failed, which for a twelve-month loan is next year. A lender that grows fast is therefore always reporting the costs of an old, small book against the yield of a new, large one, and its numbers look best in the quarter before they look worst.

What good looks like, from the one company that publishes it

Bajaj Finance is the largest non-bank lender in India and the most useful benchmark because it reports every line. Its investor presentation for the year to March 2026 gives a cost of funds of 7.54 per cent, loan losses and provisions of 2.09 per cent of average assets under finance, operating expenses of 33.3 per cent of net total income, gross NPA of 1.01 per cent and net NPA of 0.41 per cent, a return on assets of 4.3 per cent and a return on equity of 18.1 per cent. Read those as the frontier. Thirty years of data, a AAA rating that brings in bank money near the cheapest rate available to a non-bank, a collections machine in every district, and the result is four rupees of return on every hundred lent.

A startup has none of those advantages, and its book sits at the other end of every line. It borrows from banks and larger NBFCs at rates founders commonly put between twelve and sixteen per cent, a figure to treat as what the market says rather than a published statistic. It lends small, short, unsecured tickets to borrowers the large lenders declined, so its yield is higher, twenty-four to thirty-six per cent, and so is its credit cost. And it originates and collects by hand at a scale where every loan carries a large fixed cost. The spread between a startup’s yield and its costs is wider than Bajaj’s in the numerator and narrower in the result, and the figure below shows why.

Cost of funds, and the regulator’s thumb on it

A lending startup’s capital is equity, which is expensive and limited, and debt, which is cheaper and must be found. Debt comes from banks, from larger NBFCs on-lending to smaller ones, from debt funds and from securitising the book. Its price depends on the lender’s rating, its vintage and the regulator’s view of its asset class. In November 2023 the RBI raised the risk weight on unsecured consumer credit at banks and NBFCs from 100 to 125 per cent, excluding housing, vehicle, education and gold loans, and added 25 points to the weight on bank lending to NBFCs where the rating-based weight was below 100 per cent. Capital became dearer for exactly the loans most lending startups write, and the money that funds them became dearer at the same time.

In February 2025 the RBI restored the rating-based weights on bank exposures to NBFCs with effect from 1 April 2025. It did not reverse the 125 per cent weight on unsecured consumer loans, which stood as this lesson was checked in October 2026. The practical lesson is that cost of funds is a regulated variable, that the regulator moves it when it sees unsecured credit growing too fast, and that a lending startup’s model should carry a scenario in which its funding cost rises two points in a quarter for reasons that have nothing to do with the company.

Credit cost, and when a loan becomes an NPA

Credit cost is the line founders most underestimate, partly because they confuse it with the gross NPA ratio. Gross NPA is the share of the book that is more than ninety days overdue today. Credit cost is what the year’s bad loans cost, provisions and write-offs over the average book, and for a short-tenure lender it is routinely several times the NPA ratio, because loans are written off and leave the book before they can accumulate in it. A book with two per cent gross NPA and eight per cent credit cost is not a contradiction. It is a small-ticket lender.

The RBI fixes the clock. Its November 2021 clarification on income recognition and asset classification applies to all lending institutions including NBFCs: a loan is a special mention account from the first day it is overdue, SMA-1 past thirty days, SMA-2 past sixty, and a non-performing asset once overdue for more than ninety days, classified in the day-end process on the ninety-first day. An NPA goes back to standard only when the entire arrears of interest and principal are paid; paying the interest alone does not do it. The Scale Based Regulation of October 2021 brought every NBFC to the ninety-day norm by March 2026 and raised the minimum net owned fund for an NBFC-ICC to ₹10 crore by March 2027, which is the price of the licence a lending startup needs to lend on its own book.

For scale, the RBI’s Financial Stability Report of June 2024 put the NBFC sector’s gross NPA ratio at 4.0 per cent at end-March 2024 with capital adequacy at 26.6 per cent. That is the average across housing, vehicle, gold and corporate lending. A startup writing unsecured personal or small-business loans should expect to live well above it and should build its provisioning as if it will.

Operating cost, which ticket size decides

Operating cost in lending is mostly per loan, not per rupee: the KYC, the credit check, the disbursal, the reminders, the field visit when reminders fail. Suppose it costs ₹800 to originate and collect one loan from first application to final instalment. On a ₹10,000 six-month loan that is eight per cent of principal, and because the money is out for only half a year it is sixteen per cent of average assets on an annual basis. On a ₹1 lakh twelve-month loan the same ₹800 is under one per cent. This is why ticket size and tenure decide whether a lending model can exist at all, and why so many small-ticket lenders show healthy yields and negative returns.

The figure opens on one book, worked. A startup in Gurugram lends ₹50 crore to small shopkeepers at a yield of 28 per cent, earning two per cent in processing fees on top. It funds the book with ₹12.5 crore of equity and ₹37.5 crore of debt at 14 per cent, so leverage is three times and interest costs 10.5 per cent of the book. Losses run at seven per cent and operations at eight. Per ₹100 lent: 28 plus 2 in, less 10.5 for funds, less 7 for losses, less 8 for people, leaves 4.5 before tax, about 3.4 after, and a return on equity of 13.5 per cent. Respectable. Now read the fourth tile. Pre-tax return reaches zero when credit cost reaches 11.5 per cent. The whole business sits in four and a half points of loss rate, and one bad festive-season cohort can move losses by that much.

Move the sliders the way the world moves them. Cost of funds up two points: return on equity falls by a third. Leverage up to five times to chase growth: equity return rises but the book now earns nothing if losses pass ten per cent instead of eleven and a half, and the regulator sees the same thing you do. Yield down four points because a competitor arrived: the return is gone. No other model in this track is this sensitive to a single assumption that cannot be known in advance.

A lender does not earn the interest rate. It earns what is left of the interest rate after the money, the losses and the people are paid. Usually that is single digits.

If you are not the lender

Most lending startups in India do not hold the book. They are lending service providers, originating and servicing loans that sit with a bank or an NBFC, and earning a fee. The economics look safer: no cost of funds, no credit cost, a fee of a few per cent of what is disbursed. Two things complicate that. The partner will usually ask the platform to share the losses, and the RBI’s Digital Lending Directions, 2025 cap any default loss guarantee at five per cent of the amount disbursed from that portfolio, require the lender to invoke it within 120 days of a default, and leave the recognition of NPAs and the provisioning with the regulated lender regardless of the guarantee. A platform earning three per cent of disbursements and guaranteeing five is in the lending business on worse terms than a lender, with a capped fee and an uncapped obligation to stay within the cap.

The second complication is that the fee is set by the partner and the partner reads the same vintage curves you do. A platform whose borrowers go bad at nine per cent will find its fee renegotiated, its guarantee called and its partner gone, in that order. The lesson for a lending service provider is that credit cost is still its number even when the loans are on somebody else’s balance sheet, and the spread model above should be run as if the book were its own, because in every way that matters it is.

A monthly ritual, in forty minutes

On the fifth working day of each month do four things. First, draw the vintage curves: for every month’s disbursements, the share of principal more than thirty days overdue at each month of age, one line per cohort, so that a new cohort going bad faster than the last shows up at month two rather than month nine. Second, recompute cost of funds as the weighted average across every lender you owe, with the next repricing date beside each. Third, divide the month’s operating cost by the number of live loans and compare with the previous month. Fourth, put the four lines into the model above and read the headroom tile.

Then apply the rule. If headroom is above four points and the newest vintage curve sits on or below the older ones, the book can grow and the question is how fast your funding allows. If the newest curve sits above the older ones, stop growing that segment this month, not after the quarter closes, because every loan written while the curve is rising is a loss that has not yet arrived. Lending is the one business in which the first sign of trouble is also the last cheap one.

Nothing here is legal, tax or investment advice. Lending in India is a regulated activity and the Reserve Bank’s directions change; the ones cited were checked in October 2026, and a lawyer who has taken a company through NBFC registration is the cheapest part of starting one.

Sources

1. Bajaj Finance Limited, Investor Presentation for the quarter and year ended 31 March 2026, April 2026 — Cost of funds 7.54 per cent; loan losses and provisions 2.09 per cent of average AUF; opex to net total income 33.3 per cent; GNPA 1.01 per cent; RoA 4.3 per cent; RoE 18.1 per cent for FY26.
2. Reserve Bank of India, Scale Based Regulation: A Revised Regulatory Framework for NBFCs, October 2021 — Net owned fund of ₹10 crore by March 2027; 90-day NPA norm for all NBFCs by March 2026.
3. Reserve Bank of India, Prudential norms on Income Recognition, Asset Classification and Provisioning: Clarifications, November 2021 — SMA-0/1/2 buckets; NPA at more than 90 days overdue in the day-end process; upgrade only when all arrears are paid.
4. Reserve Bank of India, Regulatory measures towards consumer credit and bank credit to NBFCs, November 2023 — Risk weight on unsecured consumer credit raised from 100 to 125 per cent; 25 points added on bank credit to NBFCs.
5. Reserve Bank of India, Exposures of Scheduled Commercial Banks to NBFCs: Review of Risk Weights, February 2025 — Rating-based weights on bank exposures to NBFCs restored from 1 April 2025; consumer credit weights not covered.
6. Reserve Bank of India, Reserve Bank of India (Digital Lending) Directions, 2025, May 2025 — Default loss guarantee capped at five per cent of the amount disbursed; invocation within 120 days; NPA recognition stays with the regulated entity.
7. Reserve Bank of India, Financial Stability Report June 2024, press release, June 2024 — NBFC GNPA ratio of 4.0 per cent and CRAR of 26.6 per cent at end-March 2024.