Apollo Finvest India Ltd Management Discussions

We Read the 700+ Bureau Score. We Also Read 700+ Other Signals.

MANAGEMENT DISCUSSION & ANALYSIS

Formal credit was built to read the past. That has always been its logic. That has always been its flaw.

The problem with the old way: it systematically excludes 750 million adults who have never had a formal loan¹ - not because they cannot repay, but because they have never been given the chance to prove it. And it misreads millions who do have records: a borrower active across multiple short-term credit platforms looks overleveraged by bureau standards. In this category, they are often the best customer you have.

Both problems have the same answer. Indias digital economy produced something the financial system hadnt bargained for: behavioral data at scale. UPI transaction histories. Real-time bank account data flows through the Account Aggregator framework. Employment records from EPFO. SMS data parsed for financial signals. Device intelligence. App usage patterns. Suddenly both kinds of borrowers had rich, legible financial lives the bureau had never been able to see - the ones with no history, and the ones whose history told the wrong story. The industry just needed to learn to read them.

In FY 2024-25, 3 out of every 4 (76%) personal loans in India were given by a fintech² - not a bank. But because each loan is small (a few thousand rupees, not a few lakhs), fintechs only account for 13% of the total money lent. Banks look at the money and see a small market. We look at the people and see the largest one.

The market serving the most Indians is the one traditional lenders have spent the least time understanding.

1. Who Built the Rs.1.20 Lakh Crore STPL Market - and How

(INDUSTRY STRUCTURE AND DEVELOPMENTS)

Rs.1.20 lakh crore³ in outstanding personal loans. A 33% CAGR⁴ over four years. Dozens of companies, millions of repeat borrowers. The STPL market stopped being a bet a long time ago.

True Balance started as a prepaid balancechecking utility - the app that told you how much balance you had left on your prepaid SIM - and grew it into a lending business serving millions of Tier 2 and Tier 3 borrowers. Branch International brought a globally-tested alternate data underwriting model to India, using smartphone signals to lend to borrowers no bureau could assess. Stashfin built a credit line layered on behavioral underwriting for the salaried segment. FatakPay, SmartCoin, mPokket, Kissht, Zype, Fibe, Freo, CASHe, MoneyView, and LendingPlate each found their angle on the same fundamental insight: Indias new borrower is real, creditworthy, and wildly underserved by the traditional system.

Collectively, fintechs and NBFCs have, till date, originated 87% of small-ticket personal loans in India⁵. This is not growth driven by marketing spend. It is growth driven by genuine product-market fit: millions of Indians who had never accessed formal credit discovering that digital STPL actually works for them.

The data infrastructure behind this growth is equally remarkable - and increasingly specific.

Perfios - incorporating Karza Technologies - aggregates real-time financial data from 700+ government and regulatory sources⁶, enabling instant bank statement analysis and income verification. CreditVidya uses AI on alternate data to score borrowers that bureau models refuse to see, helping lenders approve 15% more applicants without increasing risk⁷. FinBox reads the financial story hidden in a borrowers SMS inbox: when salary arrived, which EMIs were debited, whether utility bills were paid on time. The signals that formal credit has always missed - hiding in plain sight on every Indian smartphone. Bureau ID checks the phone before you check its data. A device flagged for fraud at another lender yesterday is one you do not want on

your book today. Signzy - the only Indian platform named in Gartners Market Guide for KYC Platforms for Banking 20 24⁸ - brings device fingerprinting and behavioral verification to the identity layer.

Sitting above all of this: the Account Aggregator framework, with 2.61 billion financial accounts now enabled for consent-based data sharing⁹ and AA- enabled loan volumes crossing Rs.1.6 lakh crore in FY25¹⁰. The AA framework converted data that once sat siloed in individual bank accounts into a shareable, consent-verified, machine-readable signal that any lender can act on in seconds.

60% of digital lenders have already deployed machine learning in 2025 to automate risk scoring, cutting approval timelines from days to minutes¹¹.

The RBIs Digital Lending Directions 2025¹² have brought the governance layer that will sustain this ecosystem through its next decade of growth. Regulation, in this case, is a filter. We welcome it.

2. Why the Timing Has Never Been Better - And What Could Still Go Wrong

(OPPORTUNITIES AND THREATS)

The Opportunity

Two things are converging in India that will not converge anywhere else on earth in this decade.

The first: 65% of Indias population is under 35. This cohort is entering the workforce right now - digitally native, without bureau history, earning their first salaries, and with exactly the kind of short-term credit need that STPL was built to serve. This is not a demographic niche. It is a wave.

The second: the infrastructure to serve this cohort credibly - AA-enabled bank feeds, EPFO employment records, device intelligence, SMS parsing, real-time alternate data - exists today, in production, at national scale. Five years ago, lending to a first-time borrower without bureau history meant accepting unquantifiable risk. Today it means deploying a model on data richer than anything a bank has ever had on its own customers.

These two things converging in the same decade is the opportunity. It will not repeat.

The Threats (The Honest Ones)

The most underappreciated risk in STPL is not credit risk. It is over-leverage - India has no comprehensive cross-platform visibility

into how many lenders a single borrower is simultaneously active with. A borrower with a clean repayment record on your book may be running live loans across four other platforms you cannot see. As the STPL ecosystem matures, this invisible over-leverage will produce the industrys next NPA cycle - and it will surprise every operator who was only watching their own portfolio.

The second risk is distribution commoditization. As affiliates, aggregators and marketplaces grow in power, the lenders product differentiates less and price differentiates more.

CAC on borrowed distribution can spike overnight with a single policy change from a platform partner. Building owned distribution - brand, direct installs, referral loops - is not a marketing exercise. It is a strategic hedge.

Both risks have the same antidote: know your borrower better than any platform or competitor does. The data flywheel is not just an underwriting advantage. It is the only durable defence.

3. Eight Years of B2B Research. One App.

(SEGMENT-WISE PERFORMANCE: B2B)

Most consumer lending apps are built on hypothesis. Apollo Cash had a different starting point: eight years and over Rs.900 crore deployed as capital partners to some of Indias best - and most instructive - digital lenders.

Co-lending, term loan, and warehousing partnerships gave us a vantage point that no market research can replicate: real portfolio data, real underwriting decisions, real collections performance across different borrower segments and risk philosophies. We have seen credit models that looked elegant in theory and fell apart under real borrower behaviour.

We have seen the precise moment when a lender who prioritised disbursement over underwriting started paying for it - in their NPA numbers, quarters later.

That library of observed failure - and observed excellence - is what the B2B business built for us.

It is also what Apollo Cash is built on.

Both lines operate on the same belief: the best credit decision is the one that reads the borrower most accurately. Not the one that approves the most. Not the one that rejects the most risk. Accuracy is the product. Everything else is downstream.

4. Apollo Cash: The Year Reality Became Our Product Manager

(SEGMENT-WISE PERFORMANCE: B2C)

In October 2025, we launched Apollo Cash, our direct-to- consumer STPL app, to the public.

What followed is the kind of education that no amount of planning can buy - only shipping can.

Real borrowers in real cash-flow situations, on real phones, making real decisions about whether to trust a new app with their financial life. You cannot simulate this. You can only build it, and then, listen very carefully.

We listened. We iterated on the onboarding journey, on borrower segment targeting, on underwriting calibration. Every signal from real user behavior went back into the product - and the product got meaningfully better for it.

What makes this iteration possible is not instinct - and in STPL, instinct is often exactly wrong. An overleveraged borrower is a red flag to human judgment; the data might show three years of on- time repayment on every short-term loan they have ever taken. Most lending decisions in this industry are still made the old way: a rule of thumb, a credit committee, a gut call.

Apollo Cash does not operate on hunches. Every decision is tested: every new risk rule, every product feature, every marketing campaign runs as an A/B test before it runs at scale. Real-time dashboards mean results come back the same day, not next quarter - so when something works, we double down fast, and when it doesnt, we catch it and roll back before it compounds. This is also why our team does not spend its time building Excel models and pivot tables: the dashboards do that. The team is freed up for judgment. This operating DNA is not new to Apollo - it is the same discipline that ran our B2B partnerships business, where real-time portfolio visibility meant we always knew, faster than most, where credit quality was drifting. We carried it into Apollo Cash. The data and the tests do the talking, not the room.

By March 2026, Apollo Cash had crossed 1,00,000+ organic downloads. More importantly - this is the number we actually track - our install-to-disbursal conversion rate in FY2025-26 ran at 4x the industry benchmark of 3-5%¹³. High conversion is not just a UX metric. It means the right users are finding the product, because the channel, onboarding design, and underwriting are all calibrated to the same borrower. That alignment does not happen by accident.

Every rupee spent on acquisition in FY2025-26 was evaluated through one metric: Cost per Disbursal (CPD). Not cost per install. Not cost per application. Disbursal - because an install is curiosity, an application is intent, but a disbursal is proof. That is the first moment a borrower is real.

A borrower who has never had a formal loan can open Apollo Cash and - in under 60 seconds - consent-share their live bank transaction history through the Account Aggregator framework: 2.61 billion accounts enabled, 64 lenders live in production loan journeys¹⁴. That AA feed, combined with EPFO employment verification, SMS data parsed for income and spending signals, device intelligence checking for fraud patterns and device trustworthiness, and behavioral signals from the application journey itself - produces a credit picture that is often more predictive than a bureau score built on a three-year-old loan repaid on autopay.

Financial performance for the year - including CRAR, maintained well above RBIs 15% minimum, and asset quality actively managed through the underwriting discipline in Section 5 - is reported in full in the financial statements. Key ratios that moved materially in FY2025- 26 reflect the deliberate shift from B2B colending to direct-to-consumer lending, not operational deterioration.

5. 3 Generations of STPL Underwriting: 2 Scale With People, 1 Scales With Data

(RISKS, CONCERNS AND INTERNAL CONTROLS)

Here is the most useful lens for understanding the STPL market: not by company, not by ticket size, but by how different operators decide who gets a loan.

Generation 1 - The Telephone Operators

We have walked these floors. Co-lending relationships take you into your partners operations: loan decisions made by humans on phone calls. A telecaller rings the applicant, asks about income and employment, makes a judgment call. Slow - approvals that take days because every decision waits for a human. Inconsistent - the same application gets different outcomes from different agents. Expensive - a human cost embedded into every disbursement. Impossible to scale without proportionally scaling headcount.

This generation built the first wave of microfinance and informal lending. It is now being disrupted by the generations below.

Generation 2 - The Bureau

Believers

Automated rules applied to two data sources: a CIBIL score and a PDF bank statement. Faster, more systematic, lower cost perdecision. But fundamentally limited. The CIBIL score only exists for people who have already borrowed formally - which, as TransUnion CIBILs own data shows, is only 27% of credit-eligible adults¹⁵. And a PDF bank statement from 90 days ago is already historical fiction: a snapshot of the past, not a signal of the present. Generation 2 is where most of the STPL industry still lives. It is better than Generation 1. It is not good enough.

Generation 3 - The Alternate Data Readers

AA-enabled live bank feeds¹⁶ instead of PDF statements. EPFO employment verification¹⁷ instead of salary slips. SMS data parsed for income regularity, spending patterns, and financial stress signals. Device intelligence from Bureau to detect fraud and assess device trustworthiness before a rupee moves. Behavioral signals from the application journey itself fed directly into the credit decision. Generation 3 does not just approve borrowers that Generation 2 rejects - it approves them more accurately, with lower NPA and higher repeat rates, because it is reading the present, not the past.

Apollo Cash is a Generation 3 lender.

Every underwriting decision is built on this stack. The competitive moat is not the brand or the marketing - it is the data flywheel: every loan disbursed teaches the model something that no external data source can replicate. Even defaults. A borrower who doesnt repay teaches us precisely where our underwriting was wrong - so the next borrower who looks like them never gets a loan. Every outcome, good or bad, sharpens the next decision.

On governance: the Company maintains audit and risk management frameworks commensurate with this complexity, reviewed actively by the Audit Committee. The RBIs Digital Lending Directions 2025¹⁸ compliance is built into the product from day one, not patched in after.

6. In STPL, the First Loan Is the Price of Entry. The Fourth Is the Prize.

(BUSINESS ECONOMICS AND OPERATIONAL PERFORMANCE)

Most lenders dont talk about this. We will.

Loan Cycle 1 is expensive. CAC, underwriting, KYC - and once onboarding funnel abandonment rates of 30-60% are factored in, the true cost per funded customer is substantially higher than most companies publicly report¹⁹. The first loan does not make money. This is understood. What is less discussed is what you are actually buying.

What you are actually buying is the right to observe a borrower under real financial pressure: did they communicate or go quiet Rs. Came back in three weeks or three months Rs. By Loan Cycle 3, you have stopped predicting them entirely - you are recalling them. That gap is the moat.

The first loan is a hypothesis. Every loan after that is proof.

This is where LTV - Lifetime Value - becomes the right frame for understanding the STPL business model. On Loan Cycle 1, LTV is negative. A small-ticket loan cannot absorb the marketing and servicing costs - and most people who look at the first- loan economics alone conclude the model doesnt work. They are looking at the wrong unit. LTV is not a single-loan calculation. By

Loan Cycle 3, the same borrower costs almost nothing to re-acquire, underwrites faster because you know them, and carries demonstrably lower credit risk because you have observed them under pressure. The interest income on Loan Cycle 3 is not just covering Loan Cycle 3 - it is retiring the investment made in Loan Cycle 1. By Loan Cycle 4, the relationship has compounded well past breakeven and into genuine, scalable value. That is what makes the fourth loan the prize.

But how do you ensure that your best users get to the 4th loan Rs. In STPL, retention is a precision (not a volume) problem.

A borrower who repaid 30 days ago is not best reached on a random day with a generic message. Reach out to them when their behavioral data signals a cashflow need: typically in the last week of the month, when spending patterns shift and the gap between income and outflow widens. A reactivation campaign targeting borrowers who repaid in the last 45 days, receive salary on the 1st, and show end- of-month spending signals will convert at multiples of any broad-based push.

Which brings us back to Loan Cycle 1 - and why the quality of the first acquisition matters so much. Companies that understand the value of Loan Cycle 3 stop optimising acquisition for volume and start optimising for quality of hypothesis. Our install-to-disbursal rate running 4x above industry benchmark²⁰ is a signal of this: the users completing our journey are real borrowers with genuine intent, for whom our speed is not a UX feature. It is a trust signal. A 2-hour approval says: we are still deciding. A 2-minute approval says: we trust you. For someone accessing formal credit for the first time, that difference is not about speed. It is about being believed. Apollo Cash is built to deliver that moment. Every time.

7. The Believers We Found: Hiring People Who Have Already Been Wrong at Scale

(MATERIAL DEVELOPMENTS IN HUMAN RESOURCES)

"It doesnt make sense to hire smart people and then tell them what to do. We hire smart people so they can tell us what to do."

- Steve Jobs

We take this literally. Not as a value statement on a wall. As an operating principle in every hire.

In a zero-to-one build, you dont just need smart people. You need believers - people who have done this specific thing before, at real scale, with real consequences, and who believe it is possible here, now, with us.

Smart people are everywhere. Believers are rare. What does a believer look like Rs.

In risk: someone who has watched a credit model fail in production - not in backtesting, in production ~ and can tell you exactly which signal they over-trusted, which borrower cohort surprised them, and what they rewrote. In marketing: someone who has stared at a CAC number that makes no sense and found the funnel leak anyway. In engineering: someone who has rebuilt a lending apps architecture with a traffic spike still in progress.

We have an informal test for every senior candidate: have you ever been badly wrong, at scale, with consequences Rs. The candidates who answer yes - specifically, without defensiveness, with a clear account of what changed - are the ones we want. You cannot build good judgment without having paid for bad judgment first.

You only need one believer to begin with.

One believer per domain changes what the room expects of itself. When the standard is a real person - someone who has actually done this, broken it, and rebuilt it - there is nowhere to hide. Mediocrity doesnt survive proximity to that. Which is why the hire has to be right. A wrong hire doesnt just fill a seat badly. They lower the bar for everyone around them. One per function. Then you compound them.

The other thing about believers: you have to give them real authority. Companies consistently hire someone exceptional and then ask them to present decisions upward for approval - which sends one unintentional message: we hired you to execute, not to decide. That is not Apollo. We hired them to decide. If we werent prepared to trust their decisions, we hired the wrong person.

In FY2025-26, we found believers across data science, engineering, risk, collections, and finance - each from deep fintech experience, each carrying what we think of as productive scar tissue from having built something similar and watched parts of it break. They walked in knowing what good looks like. We gave them the room to build it. Because the best decisions come from people who bear the consequences of those decisions.

8. The Hypothesis Is Proven. 1 to 100 - The Beginning.

(OUTLOOK)

As Munger observed, the best businesses improve their competitive position simply by operating - where each transaction makes the next one stronger. Data compounds. Trust compounds. Underwriting accuracy compounds. Distribution relationships compound.

Apollo Cash is designed to work exactly this way. Every loan disbursed sharpens the risk model. Every repeat borrower confirms the thesis. Every distribution partnership - marketplace, aggregator, API - adds reach without the cost of infrastructure. The flywheel turns slowly at first. Then, with enough data and enough trust relationships in motion, it becomes very difficult to stop.

We are not at the beginning of the flywheel. We are past it. The product is live and learning. The team is built and compounding. The underwriting model is running on real data. The infrastructure - AA, EPFO, alternate signals, device intelligence - is production-grade and getting richer with every loan.

The market itself is proven: a dozen thriving players - from True Balance to Branch to Fibe- each growing, each serving a fraction of 750 million underserved borrowers²¹ who have barely begun to discover that formal credit is accessible to them.

FY2026-27 is where we prove it at scale. The hypothesis of FY2025-26 is proven. The proving year begins now..