The Fortress and the Flow: How Digital Public Infrastructure and the War for Liabilities Are Rewriting Indian Banking
1. Cold Open & The Upstream Belief: The Great Indian Balance Sheet Reset and the Frictionless Rupee
Start with a transaction that now repeats, in some version, several hundred million times a day across India.
A tea vendor in Jaipur takes a twenty-rupee payment. There is no cash, no change, no till. The customer scans a printed QR code taped to the counter, and the money moves between two bank accounts through the Unified Payments Interface β the national payments switch that in early 2026 was clearing more than 15.2 billion transactions a month, worth over βΉ20.6 lakh crore.1 The vendor pays nothing for the rail. The bank on either side earns nothing directly from that βΉ20.
What the bank earns is a record. Every one of those receipts lands in a ledger with a timestamp, and twelve months of them describe a business more honestly than any balance sheet the vendor could produce, because the vendor has never produced one. When that vendor opens a lending app and consents β through the Account Aggregator framework, a consent-based data-sharing protocol whose ecosystem had crossed 110 million cumulative account links by early 2026 β the lender pulls those statements directly, encrypted, in seconds.2 A pre-approved working capital line of βΉ25,000 can be sanctioned before the next customer arrives. No land deed. No gold. No guarantor. No branch visit.
The vendor is a composite drawn from the mechanics of the system rather than from a single documented case. The mechanics are not. This is the standard operating model of lenders such as AU Small Finance Bank, which built its franchise underwriting exactly this customer β self-employed, semi-urban, cash-flow-visible, with no formal financial history β and which reported net interest margins near 5.50% and advances growing about 24% year on year in FY26.3
Now rewind eleven years. In 2015 the Reserve Bank of India ordered an Asset Quality Review that forced banks to stop pretending. Loans to steel mills, power projects and infrastructure holding companies that had been quietly evergreened for years were dragged onto the books as what they were, and system gross non-performing assets climbed from around 4.3% to a peak of 11.5% by FY2018.4 In that world a small business loan required a site visit, a valuer's report on immovable property, months of file movement, and a credit officer's judgement about a borrower whose cash flows were invisible by design. The economics were brutal in both directions: the loan cost too much to originate, and the collateral behind it turned out, in aggregate, to be worth far less than the paper said.
Between those two scenes sits the question this article is about. Something in India changed the physics of making a loan. The question for an investor is whether the profits from that change accrue to the institutions that made it, and if so, which ones.
Why we looked here
The belief that sent us into Indian banking is a claim about household behaviour rather than about banks, and it can be stated as a single falsifiable proposition:
India's transition toward formal financialization β driven by digital public infrastructure, per-capita GDP crossing roughly $2,500, and the economic integration of rural and urban India β is permanently re-routing household savings out of physical, non-yielding assets and into formal bank intermediation, payment rails and capital-market channels, creating an expanding, high-margin, low-credit-cost intermediation pool for well-capitalised lenders.
Permanently is the load-bearing word. India has had bursts of financial deepening before, and they have reversed. The evidence that this one is different comes from three streams that are collected independently of each other and, crucially, independently of the banks whose share prices are at stake.
The first is macro and demographic. Household gross financial savings recovered to roughly 11.2% of gross national disposable income in FY25βFY26, after several years in which inflation pushed households back toward gold and unorganised property.4 Underneath that aggregate sits an infrastructure fact: the Pradhan Mantri Jan Dhan Yojana programme had opened more than 520 million basic bank accounts holding over βΉ2.3 lakh crore β around $27.5 billion β by early 2026, most of them operationally active rather than dormant.5 These come from central bank deposit registries and ministry audits, not bank marketing decks.
The second stream is behavioural, and it matters most because it measures what people do rather than what they own. UPI volumes and Account Aggregator consent counts are recorded at the switch level and in open-banking API logs. Fifteen billion transactions a month is not a survey response.
The third stream comes from sibling industries β places where the same household rupee would show up if the belief were true. Monthly systematic investment plan inflows into Indian mutual funds pushed past βΉ24,000 crore, roughly $2.85 billion a month, in early 2026, and retail credit bureau coverage expanded past 440 million unique individuals.67 A country still hoarding cash and metal would run neither.
How the belief reaches a bank's income statement
Three transmission channels carry this shift into bank fundamentals, and they hit different parts of the P&L. The liability engine converts physical currency and informal savings into digital deposits: frictionless onboarding turns a cash-economy participant into a current-and-savings-account holder, the cheapest funding a bank can obtain. The asset engine replaces collateral with cash-flow data, so a lender prices risk on observed receipts rather than on the liquidation value of a pledged asset β collapsing customer acquisition cost, by industry estimates on the order of 60β70% for digitally originated retail and MSME credit, and changing which borrowers are lendable at all. The operating efficiency channel is the arithmetic consequence: paperless origination and digital servicing pull cost-to-income ratios down from the historical Indian norm of roughly 50β55% toward 38β44% at the technology-led private lenders.
The same belief implicates a set of adjacent industries this article deliberately does not chase: asset managers capturing the discretionary savings that overflow past deposit rates, life and general insurers riding credit-linked distribution, and non-bank finance companies operating at the high-yield boundary of the credit grid while borrowing wholesale from the very banks they compete with.
The decision context, stated plainly
This piece is written for a general institutional public-equity reader with a three-to-five-year thematic horizon, evaluating listed expressions on a benchmark-relative basis. It contains no position sizing, no price targets and no trade recommendations. The scope is listed Indian scheduled commercial banks β the domestic systemically important banks, the private challengers, the small finance banks and the public sector banks β with deliberate comparison to Brazil, the United States and Japan. Standalone non-bank lenders, pure asset managers, payment aggregators and insurers sit outside the boundary, appearing only where they supply, distribute for, or compete with a bank balance sheet.
The governing distinction throughout: a correct forecast about a society can still produce a poor security outcome, because the analyst chose the wrong layer of the value chain, the wrong company within that layer, or the right company at the wrong price. India's financialization could unfold exactly as described and still leave a portfolio of Indian bank equities underwater, if the profits are competed away, regulated away, or already embedded in the multiple.
What would prove us wrong
The upstream belief has three named falsifiers, stated here before the evidence so the article reads as a test rather than an argument.
If currency in circulation as a share of GDP rises above 12.5% for two consecutive fiscal years, Indians are moving back to cash and the shadow economy is reasserting itself; as of mid-2026 that ratio remained contained, around 11.5%.4 If household net financial capital formation falls durably below 5.0% of GDP, real incomes are not generating the savings the thesis requires. And if credit costs on digitally underwritten retail portfolios exceed 3.5% annually across a full 24-month cycle, then cash-flow underwriting has delivered faster lending rather than better risk selection.
Digitisation has removed the friction from finding and underwriting a borrower. That is the easy half. Removing friction on the asset side of a bank's balance sheet only sharpens the fight on the other side, where the money to lend has to come from. That fight is the central economic story of Indian banking in 2026.
Subscribe on LinkedIn2. The Tech Foundation: India Stack, UPI, and Account Aggregators Demolishing the Underwriting Gate
For most of the history of commercial banking, lending to a small borrower has been governed by what credit officers call the three Cs: collateral, character and capacity. Of the three, collateral did nearly all the work, because it was the only one that could be verified cheaply. Character required a relationship. Capacity required accounts. Collateral required a document and a valuer.
The consequence was a country-sized filter. If a borrower's wealth sat in forms the formal system could not perfect a charge over β a rented shopfront, inventory, a family's gold, an unregistered plot β that borrower was unbankable at any price, and went to a moneylender at rates that made the comparison academic.
A useful way to describe what changed: traditional underwriting took a single static photograph of a borrower's assets, perhaps once every five years, and lent against the photograph. Account Aggregator-based underwriting streams continuous video of the borrower's cash inflows and outflows. The lender sees seasonality, concentration, volatility, and β most valuable of all β the moment things start to go wrong, months before a payment is missed.
The analogy has a limit worth stating immediately, because the limit is where the credit losses live. Video of a bank account is not video of a business. It shows the flows that pass through the formal system and is blind to the ones that do not, which in an economy still partly informal can be a large share. It shows a borrower's receipts but not the three other lenders who saw the same receipts and lent against them the same week. And it can be gamed: circular transfers between related accounts can manufacture the appearance of turnover. Cash-flow data improves the selection of borrowers dramatically. It does not, on its own, solve leverage stacking or fraud, and both of those are where digital lending goes to die.
The four layers
India's digital public infrastructure is best understood as four stacked layers, each of which removed a specific cost from the lending process.
The identity layer came first. Aadhaar-based electronic know-your-customer verification collapsed onboarding cost from several hundred rupees of paperwork, courier and manual verification to a figure practitioners put in the low double digits per account. That is what made the Jan Dhan expansion economically survivable; opening 520 million accounts at βΉ600 apiece would have been a βΉ31,000-crore act of charity.5
The payments layer β UPI, operated by the National Payments Corporation of India β did something more subtle than displace cash. It made small-value economic activity legible. A merchant who accepts payment digitally generates an audit trail as a by-product of doing business, without filing anything. By early 2026 the system was running at over 15.2 billion transactions a month.1 What matters to a lender is less the volume than the consequence: a previously invisible category of borrower now arrives pre-documented.
The data empowerment layer is the Account Aggregator framework, coordinated through the industry body Sahamati. It is a consent-based protocol: a borrower authorises a specific lender to pull specific data for a specific period from banks, the goods-and-services-tax network, and other financial information providers. The aggregator itself is a blind pipe β it moves encrypted data and cannot read it. Cumulative linked accounts crossed 110 million, with monthly consent volumes running near 12.5 million by mid-2026.2
The credit protocol layer β the Open Credit Enablement Network, or OCEN β is the least mature and the most consequential. It standardises the APIs of loan origination itself, so that any application with a customer relationship can present a credit offer at the point of need, with a regulated lender's balance sheet behind it. We will return to why this is the single largest structural threat to bank economics in this story.
Comparisons to European open banking are natural and partly misleading. Under Europe's second Payment Services Directive, banks were compelled to expose customer data to licensed third parties, and adoption was slow and litigated, partly because the incumbents supplying the data competed with the parties consuming it and had every incentive to keep the pipes narrow. India's design differs: the aggregator is a neutral, non-monetising intermediary on a common standard, and the incentive to obstruct is weaker because the same institution is usually both provider and consumer on different transactions. The limit of the comparison matters too. European open banking operated where credit bureaus were already deep and income documentation universal, so an API added little information. In India it adds an enormous amount, which is why the same protocol produces a much larger effect.
Who built what, and who is renting it
ICICI Bank is the clearest architectural protagonist among the large lenders. Its iMobile Pay application was opened to customers of other banks β unusual for an institution whose historic advantage was its own customer list β and had drawn more than 12 million non-ICICI users by FY26, working as a low-cost acquisition funnel into lending products.8 The financial signature showed up in FY26 as a cost-to-income ratio near 39.2% and return on assets around 2.38%, the highest among large Indian banks.8
Federal Bank, a mid-sized Kerala-headquartered lender, took the opposite route to the same rail. Rather than build a consumer super-app, it became the regulated balance sheet behind consumer-facing fintech applications including Fi, Jupiter and OneCard.9 The trade is deliberate: the fintech owns the relationship and the interface, Federal owns the deposit, the licence and the credit risk, and earns a spread it could not have bought that cheaply through its own branches. It works up to a point β FY26 return on assets of roughly 1.28% sat well below the large private banks, on a structurally lower margin near 3.18%.9 Renting distribution costs less than owning it and pays less.
Behind both models sits an unglamorous engineering fact. Core banking systems built in the 1990s and 2000s assumed a customer who transacted a handful of times a month. UPI turned that customer into one who transacts several times a day at values under βΉ100, while the same systems field real-time balance enquiries and settlement postings at national scale. The vendors supplying these platforms β Infosys with Finacle, Tata Consultancy Services with TCS BaNCS, Oracle with Flexcube β had to re-architect throughput and availability rather than features. Section 7 walks that supply chain properly.
What the digital rail is worth, in basis points
The investable claim here is narrow and measurable. Digital origination lowers two costs and one of them is contested.
Operating cost is the uncontested one. Technology-forward private banks now run cost-to-income ratios in the high thirties to low forties, against a historical system norm above 50%.4 For a bank with a 3.5% margin structure, a ten-point improvement in cost-to-income is worth roughly 30β40 basis points of pre-tax return on assets β enormous in an industry where the whole spread between an excellent bank and a mediocre one is about 100 basis points of RoA.
Credit cost is contested. The bull argument is that continuous cash-flow visibility plus near-universal bureau coverage structurally lowers loss rates, because lenders select better and intervene earlier. The bear argument, which has evidence behind it, is that Account Aggregator consent skews toward prime borrowers who already had documented income, so the marginal information gain is smallest exactly where the credit risk is largest β the informal, thin-file borrower the thesis is supposedly about.
A live empirical dispute sits inside this. Fintech lenders claim their machine-learning models underwrite personal loans better than bank scorecards; credit bureau data shows materially higher multi-lender stacking among borrowers sourced through fintech channels, the same individual holding several small unsecured loans from lenders each of whom could see the bureau file but priced as though they were alone.7 Both can be true β a model can rank risk well within its own applicant pool and still lend into a systemically over-levered cohort. The Reserve Bank took the second view and acted on it.
The share of retail and MSME originations flowing through Account Aggregator consent rails β roughly 15.2% by value in mid-2026 β is the cleanest single measure of whether the underwriting revolution is real or rhetorical, and one of the four crux indicators this article returns to at the end.2
Origination, then, is close to a solved problem. Which raises the question of what is being originated with. A bank cannot lend a data stream. It lends money borrowed from depositors, and in 2026 India ran short of them.
3. The Binding Constraint: The Credit-to-Deposit Gap and the War for Liabilities
Consider the arithmetic in front of the asset-liability committee of any fast-growing Indian private bank in 2026. Loans are compounding between 13% and 17%. Deposits are growing around 10%. Both cannot continue. The bank has three moves: raise deposit rates and give up margin, buy wholesale funding and give up margin and liquidity comfort, or slow lending and give up share and the growth multiple attached to it. There is no fourth move; the regulator will not permit one. This is the binding constraint of the entire theme, and the reason returns will not distribute evenly across the institutions participating in it.
Exhibit 1 β Indian scheduled commercial banks: credit growth vs deposit growth, FY2021βFY2026 Definition: year-on-year growth in outstanding non-food credit and aggregate deposits of scheduled commercial banks, as at end-March of each fiscal year; credit-to-deposit ratio is outstanding credit divided by outstanding deposits. Units: per cent. Geography: all-India. Source: Reserve Bank of India, Weekly Statistical Supplement and sectoral credit deployment data. Evidence status: observed official data.
| Fiscal year (end-March) | Credit growth (YoY %) | Deposit growth (YoY %) | Credit-to-deposit ratio (%) |
|---|---|---|---|
| FY2021 | 5.6 | 11.4 | 71.5 |
| FY2022 | 9.6 | 8.9 | 72.0 |
| FY2023 | 15.0 | 9.6 | 75.8 |
| FY2024 | 16.3 | 12.9 | 78.1 |
| FY2025 | 13.5 | 10.8 | 79.2 |
| FY2026 | 12.0 | 9.8 | 79.5 |
Read that table aloud and the story tells itself. In FY2021, in the depths of the pandemic, deposits grew twice as fast as loans β households hoarded, banks parked the surplus with the central bank, and the ratio fell to 71.5%.10 Then the recovery arrived: credit ran at 15β16% for two years while deposits struggled to reach 13% in the best year of the cycle. Every year since FY2022 loans have outgrown deposits, ratcheting the ratio up about eight points to 79.5%. In FY2026 the gap was 2.2 percentage points β undramatic in any single year, and compounded across five it has consumed the system's entire liquidity cushion.
Why the deposits went missing
The intuitive explanation β that Indians stopped saving β is wrong, and getting it wrong leads to the wrong conclusion about whether the problem self-corrects. Household financial savings recovered.4 What changed is where they went.
The same formalization that fed banks their borrowers handed households an alternative to bank deposits. A saver who ten years ago chose between a fixed deposit and gold now chooses between a fixed deposit, a systematic investment plan executed from a phone in ninety seconds, a sovereign gold bond and a direct equity account. A large part of the βΉ24,000 crore a month flowing into SIPs is money that would previously have sat, sullenly, in a savings account at 3%.6 The banking system built the rails that disintermediated its own cheapest funding.
Real deposit rates did the rest. For much of 2022β2026 savings account rates sat below inflation, term deposits offered a thin real return, and the tax treatment of interest income compared unfavourably with capital gains on equity funds for exactly the affluent urban saver a private bank most wants.
The regulator tightens the same screw
The Reserve Bank of India then did something that made the constraint harder rather than softer, and understanding why is essential to understanding the sector.
Under draft liquidity coverage ratio guidelines circulated in 2024 and refined through 2026, the RBI proposed raising the assumed run-off factor on retail deposits linked to internet and mobile banking from 5% to 10%.4 The logic is uncomfortable and correct. A depositor who must visit a branch to withdraw is a slow depositor. A depositor with a banking app can move an entire balance to a competitor in under a minute, at three in the morning, on the basis of a rumour. Deposit runs in the digital era are faster than any liquidity buffer sized on twentieth-century assumptions.
The consequence is direct. A higher assumed run-off requires more high-quality liquid assets β government securities yielding less than loans β against the same deposit base. Industry estimates put the margin cost at roughly 8β15 basis points, though bank-by-bank impact depends on how digital deposits are classified in final rules not fully settled at the time of writing.11 The banks most affected are the technology-forward private lenders whose whole strategic story is that their deposits are mobile-first. The regulation taxes the thing the theme celebrates.
Three banks, three positions in the same war
HDFC Bank created its own constraint. The 2023 merger of its parent, the mortgage lender HDFC Ltd, into the bank produced a combined balance sheet above βΉ38 lakh crore and one of the ten largest banks in the world by assets.12 It also imported a roughly $60 billion loan book funded largely by wholesale borrowings rather than deposits, pushing the merged credit-to-deposit ratio to around 100% against a pre-merger norm near 85%. Everything since has been organised around one objective: growing deposits fast enough to bring that ratio down without shrinking the loan book. In FY26 the bank reported advances near βΉ25.2 lakh crore against deposits of about βΉ23.8 lakh crore, with deposits growing 14.4% against advances of 12.0% β deliberately running assets slower than liabilities, and adding upwards of βΉ3 lakh crore of deposits a year.12 The cost shows in a net interest margin of about 3.45%, well below what a bank of its quality historically earned.12
State Bank of India sits on the other side of the same war and barely feels it. SBI held roughly βΉ50.2 lakh crore of deposits in FY26 β about 23% of the entire Indian system β with a current-and-savings-account ratio near 41%.13 Its cost of funds is structurally lower than any private competitor's because a substantial share of its deposits arrive not through pricing but through mandate and habit: government salary accounts, defence and public-sector payrolls, pension distribution, tax collection. When the price of deposits rises, SBI's competitors bid. SBI mostly does not have to.
IDFC First Bank is the challenger paying full retail price for its funding. It grew customer deposits about 26% year on year in FY26 against advances growth of 21%, with retail deposits above 78% of the total β an unusually high-quality mix by composition.14 It bought that mix by offering savings and term rates at the top of the market and by building branches, which is why its cost-to-income ratio sat near 69.5% and its return on assets near 1.15% despite a headline net interest margin of about 6.10%.14 IDFC First demonstrates the central asymmetry of this industry: high margins earned on expensive deposits and costly distribution are not the same asset as ordinary margins earned on cheap deposits.
Three worlds forward
The scenarios worth holding are causal paths, not percentage adjustments.
In the bear world, inflation proves stickier than expected, the RBI holds or raises policy rates above 7.25%, and households keep preferring real assets and equities to deposits. Deposit growth settles at 7.5β8.5% while credit demand persists, driving the system ratio through 85%. Banks bid for bulk and term money; cost of funds rises faster than loan yields reprice; top-tier private margins compress into a 3.10β3.35% band, and credit costs rise as unsecured vintages season into a weaker income environment. Private-bank returns on assets fall toward 1.2% and public-sector banks toward 0.6%. The winners are the institutions that never needed to bid; the losers are high-credit-to-deposit private banks without branch density.
In the base world, deposit growth accelerates modestly as real incomes improve and the RBI eases reserve requirements, credit compounds at 12β13.5% β roughly 1.2 times nominal GDP growth β and deposits catch up to 11β12%. Margins hold at 3.65β4.10% for top private lenders, system gross NPAs stay near 2.0% and credit costs near 0.6%, with private RoA about 1.9% and public-sector RoA about 1.1%. The technology-efficient large private banks compound book value in the mid-to-high teens and the sector performs approximately as priced.
In the bull world, deposit growth breaks 14% as formalization pulls new savers in, credit runs at 15.5β17%, and the cost and credit-cost savings from cash-flow underwriting arrive in full. Margins expand to 4.30β4.65% for the best-positioned lenders, credit costs fall to 0.35%, and private RoA reaches 2.5%. High-yield challengers with operating leverage still to harvest outperform, because their earnings are geared to cost ratios that have furthest to fall.
Notice which variable discriminates between the three worlds. Credit demand is present in all of them; what varies is the supply of deposits. In modern Indian banking, asset creation has become a technology problem that is largely solved. Liability gathering remains a zero-sum contest over a pool that grows only as fast as household savings allow.
The next question is what that contest is actually worth. How large is the profit pool being fought over, and where does it come from?
4. The Profit-Pool Waterfall: From βΉ320 Lakh Cr GDP to the βΉ3 Lakh Cr Bank Profit Engine
Follow a single rupee of Indian economic activity down to a bank shareholder and you learn more about this industry than any ratio can teach. The journey passes through nine drains, and at the end roughly one paisa in a hundred of national output survives as bank net profit.
India's nominal GDP in FY26 ran above βΉ320 lakh crore β approximately $4.1 trillion.4 A portion of the income it generates is saved in financial form, pooling into the system deposit base of roughly βΉ215 lakh crore.10 That is the raw material of the industry, and everything downstream depends on how much of it a bank attracts and at what price.
The first drain is the regulator's, and it happens before a bank can lend a rupee. Indian banks must hold a non-interest-bearing cash reserve ratio with the central bank and a statutory liquidity ratio in government securities. Together these lock up roughly 22β23% of deposits, on the order of βΉ48 lakh crore, in instruments yielding materially less than loans.4 Economically this is a compulsory low-return asset allocation that dilutes the return on every deposit raised β which is why the marginal deposit is worth less than the headline lending spread suggests, and why the deposit war hurts more than it looks.
What remains is a deployable credit pool of about βΉ170 lakh crore. Lent across retail mortgages, auto and personal loans, credit cards, MSME working capital, corporate term lending and trade finance, and combined with fee income from third-party distribution, transaction banking and treasury, this generates a gross interest and fee income pool of roughly βΉ19.5 lakh crore.
Now the drains resume. Interest paid to depositors and wholesale lenders consumes about βΉ11.2 lakh crore β the single largest line in the industry, and the one the deposit war inflates. Operating expenses, meaning salaries, branch leases, technology and marketing, take about βΉ3.8 lakh crore.
What survives is pre-provision operating profit of roughly βΉ4.5 lakh crore. This is the number that matters most in bank analysis, because it is the earnings the bank generates before deciding how much to admit about its loan losses. Provisions absorb about βΉ0.8 lakh crore at the current credit cost of 0.5β0.6%, which is a multi-decade low. And the residue is the system net profit pool of βΉ2.8β3.1 lakh crore, roughly $34β37 billion.4
Two things about that figure deserve emphasis. It is unusually large relative to history because the provisioning line is unusually small β at FY2018 credit costs the same pre-provision profit would have produced a fraction of the net profit, and in several years none. Any analysis extrapolating 2026 credit costs indefinitely is extrapolating the best conditions in fifteen years. And the pool is being fought over by institutions with very different claims on it.
Exhibit 2 β Indian banking system market share by ownership group, 2026 Definition: share of aggregate scheduled commercial bank deposits and credit outstanding by ownership group. Units: per cent of system total. Geography: all-India. Period: as at 2026. Source: Reserve Bank of India, Handbook of Statistics on the Indian Economy and Financial Stability Report.4 Evidence status: observed official data.
| Ownership group | Share of system deposits (%) | Share of system credit (%) |
|---|---|---|
| Public sector banks | 57.5 | 53.2 |
| Private sector banks | 37.2 | 41.8 |
| Small finance and foreign banks | 5.3 | 5.0 |
The gap between the two columns is the whole competitive story in two numbers. Public sector banks hold 57.5% of deposits but only 53.2% of credit β they gather more money than they lend, and the surplus flows into government securities and the interbank market. Private banks hold 37.2% of deposits and 41.8% of credit β they lend more than they gather, and fund the difference by paying up. Now compare that to profits: public sector banks capture roughly 42% of system net profit while holding nearly 58% of deposits, and listed private banks capture roughly 52% of net profit on 37% of deposits. Private lenders convert every rupee of deposits into about 1.9 times as much profit as their public-sector counterparts, on returns on assets of 1.8β2.4% against 1.0β1.2%.
That gap has a cost explanation and a mix explanation. On cost, public sector banks run cost-to-income in the high forties to low fifties β State Bank of India reported about 52.8% in FY26 against ICICI Bank's 39.2% β driven by wage settlements, pension liabilities and a branch estate sized for a pre-digital country.138 On mix, private banks skew toward higher-yielding retail and MSME assets, while public sector banks carry more corporate and infrastructure credit priced off benchmark rates at thin spreads, plus a heavier priority sector lending burden they cannot always meet profitably.
That last point is a regulatory transfer that shows up nowhere in the headline ratios. Indian banks must direct 40% of adjusted net bank credit to priority sectors β agriculture, MSMEs, affordable housing, weaker sections β with shortfalls parked in NABARD's Rural Infrastructure Development Fund at low yields.4 For a bank with a natural franchise in those segments the mandate is costless; for one whose customers are urban and affluent it is a tax collected in basis points, through underpriced lending or through the RIDF penalty box. Indian bank margins therefore cannot be compared naively with those in unmandated markets.
Where the profit pool has moved
Two migrations have reshaped who earns what.
The first ran from wholesale corporate lending to granular retail and MSME credit. In 2015 corporate credit contributed above 60% of the large banks' lending profit; by 2026 that share had fallen below 30%. The move was forced rather than chosen. Corporate lending destroyed an extraordinary amount of capital in the 2015β2019 cycle, and the survivors rebuilt around assets that were smaller, more numerous, better diversified and 200 to 400 basis points higher-yielding on a risk-adjusted basis.
The second migration is still under way: from pure interest spread toward a fee stack. As margins normalise, the best Indian banks increasingly earn from distributing other people's products β mutual funds, insurance, wealth management β and from transaction banking and trade finance. ICICI Bank and Kotak Mahindra Bank have pushed non-interest income above 28% of total income.815 Axis Bank made the most explicit purchase of a fee pool, acquiring Citigroup's India consumer business and roughly three million affluent card and wealth clients, lifting its credit card share to around 14%.16 Fee income is worth more than its rupee value suggests because it consumes almost no capital: a rupee of commission requires no risk weighting, no provisioning and no deposit to fund it.
The reinvestment problem nobody puts in the pitch
Here is where bank analysis diverges from ordinary industrial analysis, and where thematic investors most often get hurt.
A bank's reported profit is not distributable cash. Growth consumes capital at a fixed ratio: to grow risk-weighted assets 14%, a bank must grow equity roughly 14%, which means retaining most of what it earns. Indian banks in this cycle retain on the order of 70β80% of earnings simply to fund 13β15% credit growth. A bank earning 17% on equity and growing loans 14% returns very little to shareholders in cash; the return arrives as compounding book value, and converts to shareholder value only if the market keeps paying a premium to book.
The corollary is sharp. The fastest-growing banks in this theme are the likeliest to need external equity. IDFC First Bank, growing advances above 20% on a 1.15% return on assets, generates internal capital of roughly 10β11% against balance-sheet growth above 20%.14 The gap closes through dilution. Kotak Mahindra Bank has the opposite problem: capital adequacy above 20.5% and Tier-1 above 19% means excess capital earning a low return, one reason its 14.5% return on equity understates the underlying business.15
One arithmetic consistency test is worth running on the sector's collective ambitions. If the system grows credit at 12% while every large private bank plans 14β17% and the small finance banks plan 24%, public sector banks must grow at 9β10% and cede two to three points of credit share a year. Some of that is happening. Sustained for five years it would require public sector banks to accept it β and they are recapitalised by a government that also sets their growth targets. Not every private bank's plan can be right.
The profit pool, in short, is large, has migrated toward the private lenders, and is currently flattered by the lowest credit costs in fifteen years. Whether that last condition is a structural achievement or the top of a cycle is the most important open question in the sector β and answering it means going back to the wreck.
5. Capital Cycles, AQR Scars, and the Structural vs. Cyclical Divide
In 2015 the Reserve Bank of India, under Governor Raghuram Rajan, began an Asset Quality Review. The mechanism was simple and merciless: supervisors identified specific borrower accounts across the system and instructed every bank with exposure to classify them consistently. A loan that one bank was carrying as standard while another had recognised it as stressed could no longer be carried as standard.
What that surfaced was a decade of accumulated fiction. Loans to steel producers, thermal power projects, road developers and infrastructure holding companies β extended during the 2007β2011 capex boom on the assumption that Indian growth would validate almost any project β had stopped performing years earlier and had been kept alive through restructuring, refinancing and forbearance. System gross NPAs went from around 4.3% to 11.5% at the FY2018 peak.4 Over βΉ10 lakh crore of credit was reclassified as impaired. Public sector bank equity was substantially wiped out and replaced by the government. System return on equity fell below 4%.
Causation matters here. The AQR did not create the bad loans; it revealed them, and in doing so achieved something no Indian regulator had previously managed β it made the losses simultaneous, forcing them to be dealt with together rather than rolled forward one bank at a time. The Insolvency and Bankruptcy Code, enacted in 2016, then gave lenders a resolution mechanism with a clock on it. Together the two interventions ended evergreening as a systemic strategy.
Exhibit 3 β Indian banking system asset quality, FY2018βFY2026 Definition: gross non-performing assets and net non-performing assets as a percentage of gross advances for all scheduled commercial banks; provision coverage ratio is specific provisions held against gross NPAs. Units: per cent. Geography: all-India. Period: fiscal years ending March. Source: Reserve Bank of India, Financial Stability Reports. Evidence status: observed official data.
| Fiscal year | Gross NPA (%) | Net NPA (%) | Provision coverage ratio (%) |
|---|---|---|---|
| FY2018 (AQR peak) | 11.5 | 6.1 | 52.4 |
| FY2020 | 8.2 | 2.8 | 65.1 |
| FY2022 | 5.8 | 1.7 | 70.9 |
| FY2024 | 2.8 | 0.6 | 76.4 |
| FY2025 | 2.3 | 0.5 | 78.2 |
| FY2026 | 2.1 | 0.4 | 79.5 |
Three lines, three separate improvements, and the third is the one most investors skip. Gross NPAs fell from 11.5% to 2.1% β a scale of repair few banking systems have achieved outside a sovereign crisis. Net NPAs fell further and faster, from 6.1% to 0.4%, which shows banks provisioning rather than borrowers merely curing. And provision coverage rose from 52.4% to 79.5%, meaning that for every rupee of admitted bad loan, Indian banks now hold nearly eighty paise of provisions against it.4 Alongside this, system capital adequacy stood at 17.2β17.4% against a regulatory requirement of 11.5%.4 By any historical standard, this is the cleanest and best-capitalised the Indian banking system has ever been.
Which is precisely why it deserves suspicion.
Reading the capital cycle
Bank capital cycles run in recognisable stages, and knowing which stage you are in matters more than any individual bank's quality.
Stage one, crisis and purge, 2015β2018. Forced recognition, provisioning surges, credit costs above 3.5%, equity destruction, and a collapse in the supply of credit as banks conserved capital. Public sector banks stopped lending to industry almost entirely.
Stage two, cleanup and consolidation, 2019β2022. Write-offs, state recapitalisation, and the merger of ten public sector banks into four larger entities. Strategy across the whole industry pivoted away from corporate capex toward retail lending β partly because retail was genuinely more profitable, and partly because everyone had just watched corporate lending destroy their peers.
Stage three, harvest, 2023β2026. Loss rates at multi-decade lows, capital abundant, returns on equity restored to 14β19% for the better institutions, and credit growth running well above nominal GDP for the first sustained period since the previous boom.
The uncomfortable historical observation is that stage three manufactures the next stage one. Capital cycles in banking turn not because banks make obviously bad loans but because they make ordinary loans at extraordinary volume into a segment untested through a downturn. The retail and MSME books built in India in 2023β2026 have never seen a serious unemployment shock, a monsoon failure combined with rural income stress, or a genuine property price correction. The credit costs in Exhibit 3 are real, and they are the credit costs of an untested vintage.
Separating what is permanent from what is passing
The single most useful discipline in this sector is sorting the forces into the ones that survive a cycle and the ones that do not.
Structural, and permanent: Account Aggregator adoption and the shift from collateral-based to cash-flow-based underwriting. The integration of the goods-and-services-tax database, which gave lenders a verifiable revenue signal for every registered business. Bureau coverage at 440 million individuals, which makes leverage stacking visible in a way it never was.7 The cash-to-digital transition in payments. And the regulatory architecture itself β higher risk weights on unsecured consumer credit, and the move to expected credit loss provisioning β which permanently changes how much capital a given loan consumes.
Cyclical, and temporary: the elevated credit-to-deposit ratio, the current round of deposit rate competition, net interest margin compression, short-term stress in unsecured retail vintages, and the policy rate cycle itself. These will oscillate. They matter enormously to two-year earnings and very little to ten-year value.
Two regulatory items sit awkwardly across the line. The expected credit loss framework β the move from provisioning after a default to provisioning for expected losses at origination, under Ind-AS 109 β is structural in effect but produces a one-time cyclical hit on transition. Banks with provision coverage already above 75β80% face modest capital impact; under-provisioned lenders face a real equity charge. Estimates of the transition cost range from 20 to 80 basis points of Tier-1 capital depending on the balance sheet, and the effective implementation date has itself been debated between FY26 and FY27.11 The honest position is that the framework's direction is certain and its timing and calibration are not.
The draft liquidity coverage ratio norms are the other. They are structural in that digital deposits genuinely are flightier than branch deposits and always will be. They are cyclical in that the final calibration is a negotiation.
Two banks that show what the cycle did
Punjab National Bank is the clearest expression of pure cyclical repair. It carried gross NPAs above 14% at the worst of the cycle; by 2026 that had fallen to around 4.2%, with provision coverage above 80%.17 Its return on assets of about 0.78% and net interest margin near 2.92% remain the weakest among the large listed banks, and its operating efficiency lags peers.17 What PNB owns is the second-largest branch network in India and a large, sticky, low-cost deposit base across the northern agrarian belt. What it has not yet demonstrated is the ability to convert that liability advantage into competitive returns. It trades at roughly 0.85 times adjusted book value, which is the market saying, reasonably, that it will believe the recovery when the earnings persist through a full cycle rather than through a benign one.
Bank of Baroda cleaned up faster and further. Gross NPAs of about 2.25%, provision coverage near 77.5%, return on assets holding above 1.0%, return on equity of roughly 16.1%, and a dividend yield near 4.5% β an unusual combination in a growth market.18 Its constraint is operational rather than financial: regulatory scrutiny of governance around its digital onboarding platform interrupted the customer acquisition engine at exactly the moment when digital acquisition became the main determinant of deposit growth. A public sector bank that cannot onboard digitally in 2026 is competing with one hand tied.
The system, then, enters the second half of this decade with the cleanest balance sheet in a generation, the best capitalisation on record, and a credit loss experience that is almost certainly better than the cycle-average it will eventually revert to. That combination sets up a competitive contest in which the participants are playing genuinely different games.
6. The Competitive Field & Parameter-Specific Leadership: Sovereign Moats, Tech Engines, and Micro-Yield Extractors
There is no single leader in Indian banking. There are leaders on specific parameters, for specific customers, in specific geographies, measured at specific dates β and the parameters point in different directions.
Three institutions make the point. State Bank of India operates from Nariman Point in Mumbai, an address that dates its lineage to the era when it was the government's bank. ICICI Bank operates from the Bandra Kurla Complex, the financial district built in the 2000s. AU Small Finance Bank is headquartered in Jaipur, because its customers are there. Each is winning at something the others are not seriously contesting.
Leadership on deposit scale and cost: State Bank of India
The parameter: total deposit franchise and cost of funds across all customer segments in India, as at Q4 FY26. The closest rival: HDFC Bank. SBI held approximately βΉ50.2 lakh crore in deposits, about 23% of the system, with a CASA ratio near 41% β four in every ten rupees sitting in accounts paying near-zero or low single-digit interest.13 HDFC Bank, the largest private bank, held about βΉ23.8 lakh crore, less than half.12
Why customers care: mostly they do not, consciously β which is the point. The advantage is inherited through institutional position rather than earned through product superiority. Why SBI leads rather than its rivals: the bank's lineage runs through the Imperial Bank of India and its nationalisation in 1955, which made it the default banking counterparty of the Indian state. Salary accounts for government employees, defence personnel and public sector undertaking staff default to SBI, and tax collection and pension disbursement flow through it. Layer on 22,500-plus branches in places where no private bank can justify the fixed cost, and you have a liability franchise that strategy did not build and strategy cannot dismantle.
Durability: high, with one qualification. The moat is a cornered resource β sovereign relationship plus rural physical presence β rather than a capability, which makes it exceptionally hard to copy and equally hard to extend. SBI cannot use its deposit advantage to win affluent urban wealth management, and its 52.8% cost-to-income ratio shows what happens when a network built for one purpose competes on another.13 Its YONO application, with over 75 million registered users generating more than 65% of digital retail originations, is the attempt to convert scale into digital efficiency; it works at the origination layer and has not yet moved the cost base.13 The one thing on the visible horizon that could erase the moat is a central bank digital currency letting households hold balances directly with the RBI, treated in Section 9.
Leadership on digital architecture and return on assets: ICICI Bank
The parameter: return on assets among large Indian banks, and the digital origination architecture producing it, as at Q4 FY26. The closest rival: Axis Bank. ICICI reported return on assets of about 2.38%, net interest margin of 4.25%, cost-to-income of 39.2% and gross NPAs of 1.40%; Axis reported 1.78%, 3.95%, 48.5% and 1.35%.816 Both are excellent banks, and the nine-point cost-to-income gap between them converts almost entirely into the 60-basis-point RoA gap.
Why customers care: approval speed and pricing, both of which follow from the architecture. A bank that underwrites from cash-flow data in minutes wins the customer who needs money this week.
Why ICICI leads: the answer is a reversal rather than a strategy document. ICICI was among the most damaged large private banks in the 2015β2018 corporate credit cycle. The rebuild reorganised the institution around risk-calibrated core operating profit rather than loan growth, and β the genuinely unusual decision β opened its digital front end to non-customers, accepting that much of the traffic would remain other banks' depositors. More than 12 million non-ICICI users on iMobile Pay is a customer acquisition machine with almost no marginal cost, feeding an automated risk-pricing engine, and retail and business banking now account for more than 68% of the loan book.8
Durability: high but not permanent. The capability is process power β accumulated skill in shipping digital products and pricing risk β hard to copy because it takes a decade of consistent management, and vulnerable to management change or to a competitor buying the same capability. Axis is the live test: it acquired Citi India's consumer franchise to obtain a segment and a fee stack it could not build organically.
Leadership on margin quality: a contested claim worth unpacking
The dossier framing that Kotak Mahindra Bank leads on net interest margin maximisation, with IDFC First Bank as challenger, is worth examining because on the raw number it is false. IDFC First reported a net interest margin near 6.10%; Kotak reported 4.80%.1415 Ranked on the headline metric, IDFC First wins by a mile.
Ranked on anything that reaches a shareholder, it does not. IDFC First converted its 6.10% margin into a return on assets of 1.15% and a return on equity of 10.8%. Kotak converted 4.80% into 2.15% RoA. The 130-basis-point margin advantage evaporated inside a cost-to-income ratio of 69.5% versus 46.8%, and inside a credit cost structure appropriate to high-yield unsecured and consumer-durable lending.1415
This is the most important comparability lesson in Indian bank analysis. Net interest margin measures gross revenue, not profitability. A high margin can mean pricing power, or it can mean lending to riskier borrowers with expensive money through an expensive network. Comparing NIM across banks with different asset mixes, funding structures and cost bases produces false rankings. The defensible statement is narrower: Kotak Mahindra Bank leads on margin converted to return on assets among mid-sized-to-large Indian private banks as at Q4 FY26, with capital adequacy above 20.5% and Tier-1 above 19% β the highest capitalisation in the peer set.15
Why Kotak leads: the institution grew out of a non-bank finance company and retains that heritage's instinct that a loan's price must compensate for its risk regardless of the volume implied, repeatedly declining segments where it judged pricing inadequate. The capital position is the visible artefact of that discipline. The vulnerability is real: Kotak operated under Reserve Bank restrictions arising from a technology audit, which constrained digital onboarding precisely when digital onboarding determined deposit growth. With leadership transition alongside, this de-rated the stock from roughly 3.5 times adjusted book to about 2.1 times.15 Excess capital is worth little until it can be deployed, and it cannot be deployed at scale without customer acquisition.
Leadership on semi-urban and MSME yield extraction: AU Small Finance Bank
The parameter: risk-adjusted yield on self-employed, semi-urban and small-business lending in northern and western India, as at Q4 FY26. The closest rival: IndusInd Bank, in adjacent vehicle-finance and microfinance segments. AU reported net interest margin near 5.50%, advances growth of about 24% and deposit growth of 28%, on an advances base near βΉ0.95 lakh crore; IndusInd, roughly three and a half times larger in advances, reported a 4.18% margin and 1.72% RoA.319
Why AU leads: domain underwriting that predates the digital rails. AU spent years as a vehicle and equipment financier lending to self-employed borrowers with no formal income documentation, where collateral was mobile and reputation was local. That produced underwriting heuristics and a field collection organisation that a large bank's centralised scorecard does not replicate; the India Stack made the capability cheaper to deploy rather than obsolete.
Durability: medium, and the risks are specific. AU is transitioning toward a universal banking licence, which if granted would lower its cost of funds β a 25-basis-point reduction is worth roughly βΉ280 crore of pre-tax profit at current scale β while simultaneously digesting the merger of Fincare Small Finance Bank and its differently shaped microfinance portfolio.3 At roughly 2.7 times adjusted book, the premium assumes both the licence and the integration go well.3
The rest of the field, and what each one is actually for
Federal Bank holds a genuinely differentiated position as regulated infrastructure behind consumer fintech applications, alongside a non-resident Indian deposit franchise worth about 18% of its liabilities β sticky, low-cost money sourced from the Gulf through relationships built over decades in Kerala.9 The economics are honest about themselves: a 3.18% margin and 1.28% RoA at roughly 1.3 times book.9 Its vulnerability is regulatory rather than commercial β if the RBI tightens rules on fintech co-lending or first-loss default guarantee arrangements, the partner-bank model's economics change overnight, and Federal has more exposure to that than any listed peer.
IndusInd Bank is a domain specialist in commercial vehicle finance β roughly 12% market share β and microfinance, both high-yield and highly cyclical.19 It reported a 4.18% margin and 1.72% RoA at roughly 1.5 times book, against a historical peak near 2.8 times.19 The discount reflects microfinance loss volatility, which spikes with rural distress and monsoon failure, and long-running questions about promoter holding structure. A 50-basis-point reduction in microfinance credit cost would add roughly βΉ700 crore to net profit β the mathematical statement of how much of this equity is a bet on rural weather and rural income.19
Below the large names sits a tier of regional lenders that will not lead anything nationally. City Union Bank lends working capital to small manufacturers and traders in Tamil Nadu on relationships built across generations of family businesses; Karur Vysya Bank runs a similar southern SME book with a disciplined asset quality record; Karnataka Bank is executing a slower repair from a weaker starting position. Each is a legitimate business bounded by its geography, and none is a vehicle for the national digitisation theme, because their advantage is precisely the local knowledge the digital rails commoditise.
Among the small finance banks beyond AU, Equitas Small Finance Bank built a book in micro-mortgages and small commercial vehicle finance, Ujjivan Small Finance Bank came out of microfinance and is diversifying into individual retail loans, and Bandhan Bank is the largest expression of the microfinance model, concentrated in eastern India. All three share one structural problem: a small finance bank pays more for deposits than a universal bank, competes for the same borrowers, and lacks the scale to absorb a credit shock without a capital event. The premium multiples this cohort once commanded β above 3.5 times book, on assumptions of uncapped high-yield growth β compressed as those economics asserted themselves.
The competitive map, then: SBI owns the sovereign deposit base, ICICI owns digital risk-pricing at scale, Kotak owns capital discipline, AU owns semi-urban underwriting, Federal rents its licence to fintechs, IndusInd owns vehicle and microfinance domain knowledge, and the regional banks own their districts. None of them owns the customer relationship at the point of transaction β and that is where the next threat comes from.
7. Value Chain, Suppliers, and the Open Credit Enablement Threat
When a customer taps "apply" in an Indian banking application, roughly two seconds elapse before a decision appears. Inside those two seconds, a chain of institutions is invoked in sequence, most of which the customer has never heard of and several of which the bank does not own. Understanding who sits where in that chain is how you work out where the profit will eventually settle.
Walk it from the customer inward.
At the touchpoint layer sits whatever the customer is actually looking at: a bank's own app, a branch, a fintech application, or increasingly a merchant's checkout screen. Ownership of this layer determines who controls the relationship.
Behind it sits the distribution and aggregation layer β fintech applications, marketplace lenders, and the emerging OCEN protocol that standardises how a credit offer is requested and returned. This layer is where customer intent is captured.
Beneath that is the data layer: the Account Aggregator protocol governed by Sahamati, which moves consented financial data between institutions, and the credit bureaus β TransUnion CIBIL, Experian and CRIF High Mark β which supply the borrower's existing obligations and repayment history to every scheduled commercial bank in India.7 Bureau data and aggregator data answer different questions. The bureau tells you what the borrower already owes. The aggregator tells you what the borrower earns. A lender needs both, and neither is optional.
Then the payment interoperability layer: the National Payments Corporation of India, which operates UPI, the RuPay card network, the Immediate Payment Service and the Aadhaar-enabled payment system.1 Every scheduled commercial bank connects to NPCI. There is no alternative switch. This is a regulated utility structure, and it is the most consequential chokepoint in Indian finance β an NPCI outage is a national event, and NPCI's pricing decisions on interchange and merchant discount rates directly determine whether payments are a revenue line or a cost line for banks. To date they have largely been a cost line, absorbed as the price of customer relationship and data.
Beneath the switch sits the core banking system β the ledger of record. Infosys supplies Finacle to State Bank of India, ICICI Bank, Axis Bank and Punjab National Bank; Tata Consultancy Services supplies TCS BaNCS to HDFC Bank, Bank of Baroda and IDFC First Bank; Oracle supplies Flexcube to Kotak Mahindra Bank, IndusInd Bank and AU Small Finance Bank. These relationships are confirmed in public vendor disclosures and bank filings.
Core banking is the stickiest supplier relationship in the industry, and the reason is switching cost rather than technical superiority. Replacing a core banking platform at a bank with tens of millions of accounts is a multi-year programme with a non-trivial probability of a visible failure, and bank boards that have watched peers suffer regulatory censure for technology outages are not enthusiastic buyers of that risk. The practical consequence is that vendors hold real pricing leverage on maintenance and enhancement, while banks hold leverage only at the point of initial selection or after a major failure. The dependency runs both directions: the vendors need these accounts because Indian banking is a reference market that sells their platforms globally, and a public failure at a marquee client is expensive.
At the bottom sits cloud and AI infrastructure β Amazon Web Services, Microsoft Azure and Google Cloud, which host analytics, model training, fraud detection and customer service systems for the private banks and their fintech partners in hybrid configurations. This relationship is credibly reported through hyperscaler financial-services case studies rather than through bank disclosure, and it should be read as partial rather than total: core ledgers in Indian banking have largely remained on-premises or in private infrastructure, with cloud used for the analytical and customer-facing tiers.
And at the base of the stack sits the thing none of the suppliers own: the bank balance sheet, with its licence, its deposits, its capital and its credit risk.
Where the money and the power actually sit
Now the analytical question. Across that chain, who earns what?
The core banking vendors earn a steady, low-volatility software toll β attractive economics, but a toll that scales with a bank's account count rather than with its profits, and is small in absolute terms relative to the bank's P&L. The bureaus earn a per-enquiry fee on an oligopoly structure with genuine data network effects: the more lenders report to CIBIL, the more valuable CIBIL's file is to each lender, which is a textbook scale-economy moat. NPCI, as a not-for-profit utility, deliberately does not extract economic rent from the payment rail; the surplus flows to consumers and merchants in the form of free instant payments. Sahamati likewise operates as a non-monetising standard-setter.
The unusual feature of the Indian stack is therefore this: the most powerful infrastructure layers have been placed, by policy design, outside the profit motive. That is why no private Indian company occupies the position that Visa and Mastercard occupy in the United States, or that a dominant payment app occupies in several other emerging markets. It is also the strongest single argument for why Indian banks retained their economics through the fintech era while banks elsewhere did not.
The banks, meanwhile, earn the spread and bear the credit risk, the capital requirement, the liquidity requirement and the regulatory obligation. That is the trade. It is a good trade when the bank also owns the customer.
The OCEN threat, stated precisely
Which brings us to the one development that could break it.
The Open Credit Enablement Network standardises loan origination the way UPI standardised payments. If it achieves comparable adoption, any application with a customer relationship β a payments app, a messaging platform, an e-commerce checkout, the software a neighbourhood retailer uses to manage inventory β can present a credit offer at the exact moment the customer needs money, with a regulated lender's balance sheet behind it.
The mechanism by which this destroys bank economics is not disintermediation of the balance sheet. Banks would still hold the loans; capital regulation guarantees it. The mechanism is the commoditisation of origination. If five banks bid to fund the same loan presented by the same platform to the same borrower, the platform captures the origination economics and the banks compete the spread down to the cost of capital plus credit risk. The bank becomes a balance sheet utility β profitable, regulated, and structurally unable to earn an excess return.
The observable milestone to watch is the share of MSME working capital originations reaching banks through embedded non-bank applications rather than through bank channels. If that crosses roughly 30%, the bargaining power has moved.
Two counterweights are worth stating fairly. First, the Reserve Bank has consistently intervened when non-bank entities accumulated bank-like power without bank-like regulation β the restrictions imposed on Paytm Payments Bank being the most visible instance β which caps how much of the value chain an unregulated platform can occupy. Second, origination is not the only thing banks own; the deposit relationship, the salary account, the mortgage and the wealth relationship are far stickier than a single working-capital loan, and platforms have shown little appetite for the regulated, capital-intensive work of gathering deposits.
Federal Bank's model is the live experiment in what a partner-bank future looks like, and its returns β 1.28% RoA against ICICI's 2.38% β are a reasonable first estimate of what a bank earns when someone else owns the customer.98 The consequence is a permanently lower multiple rather than a catastrophe.
Where value ultimately settles between the app and the balance sheet is the open question. The global evidence on that question comes from markets that ran the experiment first.
8. Public-Market Expressions & Expectation Wedges: Pure Plays, Value PSBs, and False Positives
In mid-2026 the Indian banking sector traded across a valuation range that would look implausible in most developed markets. ICICI Bank changed hands at roughly 2.7 times adjusted book value, State Bank of India near 1.2 times, and Bank of Baroda below book at about 0.95 times.81318 Three institutions in the same country, under the same regulator, the same rate cycle and the same credit environment, priced across a spread of nearly three times. That dispersion is where the investment question lives. The businesses genuinely differ, so the spread is not obviously wrong β but it embeds assumptions that can be extracted and tested.
Exhibit 4 β Indian bank financial read-through and valuation, FY2026 Definition: reported net interest income (βΉ crore), net interest margin, cost-to-income, gross NPA ratio, return on assets, return on equity and price to adjusted book value. Geography: India. Period: FY2026 (year ended 31 March 2026) results; P/ABV as at mid-2026. Source: company quarterly and annual disclosures; Indian exchange prices. Evidence status: reported financials observed; P/ABV is a point-in-time market price and adjusted book definitions vary modestly between banks.
| Bank | NII (βΉ cr) | NIM (%) | Cost-to-income (%) | Gross NPA (%) | RoA (%) | RoE (%) | P/ABV (x) |
|---|---|---|---|---|---|---|---|
| ICICI Bank | ~82,500 | 4.25 | 39.2 | 1.40 | 2.38 | 18.5 | ~2.7 |
| HDFC Bank | ~118,000 | 3.45 | 40.1 | 1.24 | 1.95 | 16.2 | ~2.2 |
| State Bank of India | ~168,000 | 3.05 | 52.8 | 2.10 | 1.12 | 15.8 | ~1.2 |
| Axis Bank | ~52,000 | 3.95 | 48.5 | 1.35 | 1.78 | 16.8 | ~1.8 |
| Kotak Mahindra Bank | ~27,500 | 4.80 | 46.8 | 1.20 | 2.15 | 14.5 | ~2.1 |
| IDFC First Bank | ~18,200 | 6.10 | 69.5 | 1.88 | 1.15 | 10.8 | ~1.5 |
| Federal Bank | ~9,200 | 3.18 | 51.2 | 2.05 | 1.28 | 14.2 | ~1.3 |
| Bank of Baroda | ~45,000 | 3.12 | 47.8 | 2.25 | 1.05 | 16.1 | ~0.95 |
Read down the return-on-equity column and something odd appears. SBI at 15.8% and Bank of Baroda at 16.1% post returns on equity comparable to HDFC Bank's 16.2% and above Kotak's 14.5% β yet trade at a third to half the multiple. The explanation sits in the RoA column. Public sector banks reach their returns on equity through leverage of roughly 13 times, against about 8 times for the leading private banks. Identical RoE built on materially more leverage is a lower-quality return: more fragile to a credit shock, more likely to require dilution when one arrives. The discount reflects the composition of the return plus a governance and policy-mandate risk that has repeatedly proven real.
What each expression requires, and what would break it
ICICI Bank is the purest expression of the digital-underwriting thesis among large banks, and its exposure is directly proven: retail and business banking above 68% of loans, RoA of 2.38%, cost-to-income of 39.2%, and an open-architecture app funnelling more than 12 million non-customers.8 At roughly 2.7 times adjusted book, the price requires sustained return on equity above 17% with no meaningful asset quality deterioration β continuation of the best operating performance in the industry. The variant case rests on operating leverage still to come from the API architecture pushing RoA toward 2.5%, and on the arithmetic that every 10 basis points of systemic credit cost reduction adds roughly βΉ1,280 crore to pre-tax profit.8 The first smart objection: at 2.7 times book you are paying a premium multiple against an above-mid-cycle credit cost, and 1.40% gross NPAs is not a number that survives a downturn. What would kill the company thesis is retail unsecured default rates rising through 2.5%, or RoA below 1.60% or gross NPAs above 2.80% for two consecutive quarters.
HDFC Bank presents the clearest identifiable expectations gap in the sector, and it is a timing gap rather than a quality gap. The fundamentals are not in dispute: roughly 19% of system deposits, 20% of system credit, a balance sheet above βΉ38 lakh crore.12 What is disputed is how quickly the post-merger credit-to-deposit ratio normalises from near 100% toward the historic 85% range, and therefore when the margin recovers from 3.45%. At about 2.2 times adjusted book β a discount to its own long-run multiple and to ICICI β the price appears to assume a slow recovery. The mechanical wedge is that every 10 basis points of margin expansion adds roughly βΉ2,500 crore to net interest income, the largest such sensitivity in the sector.12 If deposits keep compounding at 14.4% against advances at 12.0%, the ratio closes faster than that assumption implies. The objection: deposit growth of that magnitude is bought with term deposits, and a bank funding itself with expensive money does not recover its old margin merely by fixing a ratio. The kill criterion is deposit growth falling below credit growth for four consecutive quarters.
State Bank of India is the liability-moat expression, at about 1.2 times adjusted book excluding subsidiaries, with a 23% deposit share, a 41% CASA ratio, and enormous sensitivity: a 10-basis-point reduction in deposit cost lifts net interest income by roughly βΉ4,800 crore.13 The variant argument is that in a deposit war the institution that does not have to bid gains relative advantage, so SBI's margin should hold while private peers compress; there is also value in subsidiaries β SBI Funds Management, SBI General Insurance, the YONO platform β inside an entity valued largely on standalone bank earnings. The objection is immediate: the PSU discount has persisted for two decades and may correctly price the risk that the majority shareholder is also the policy-setter. A direction to lend into unremunerative priority projects, or a wage revision cycle, converts shareholder value into public policy with no recourse. The kill criteria are net NPAs above 1.25% or Tier-1 below 10.5%.
Axis Bank is a restructuring and integration story. The Citi India acquisition delivered around three million affluent card and wealth customers and a credit card share near 14%, feeding the fee-income migration described earlier.16 At roughly 1.8 times book, expectations sit meaningfully below ICICI's β either an opportunity or an accurate reading of execution risk. Every 20 basis points of cost-to-income improvement adds roughly βΉ850 crore to operating profit, and with cost-to-income at 48.5% against ICICI's 39.2% that runway is the whole thesis.16 The objection: acquired card portfolios have a habit of producing worse loss experience than the acquirer modelled, and integration synergies are announced more reliably than delivered. Credit costs breaking 1.50% of advances would settle it.
Kotak Mahindra Bank is the de-rated quality franchise, at roughly 2.1 times book against a prior 3.5 times on 2.15% RoA and the peer set's strongest capital position.15 The wedge, if it exists, is that regulatory remediation releases a customer-acquisition constraint on a balance sheet already carrying surplus capital: the capacity to grow is present and only the permission is missing. The objection is that undeployable excess capital is a drag rather than an option, and remediation timelines at Indian banks have historically run long. Deposit growth slowing below 10% year on year signals the constraint is binding rather than temporary.
IDFC First Bank is the operating-leverage bet. At roughly 1.5 times book on 1.15% RoA and 10.8% RoE, the market is paying for a cost ratio that has not yet fallen.14 The mechanics are stark: at 69.5% cost-to-income, revenue growth that outpaces cost growth compounds into pre-provision profit at a rate no already-efficient bank can match, and reaching 58% would transform the earnings. The objection has ended every previous version of this story: high-yield unsecured and consumer-durable lending produces its cost efficiencies in benign conditions and its losses in bad ones, and a bank growing above 20% on 1.15% RoA must return to shareholders for equity. The kill criteria are gross credit costs above 2.2% or failure to bring cost-to-income below 65% by FY27.
AU Small Finance Bank at roughly 2.7 times book requires two things to go right at once β the universal banking licence and the Fincare integration β on a franchise growing advances 24% at 5.50% margins.3 The objection is that a premium multiple on a small finance bank has historically been a bet on the absence of a microfinance credit event, and such events are periodic rather than random. Gross NPAs breaking 3.0% or post-merger RoA below 1.2% ends it.
Federal Bank, Bank of Baroda, IndusInd Bank and Punjab National Bank occupy the value end. Federal at 1.3 times is priced for its structurally lower margin, with the fintech partnership model as the optionality and the regulatory treatment of those partnerships as the risk.9 Bank of Baroda below book with 16.1% RoE and a 4.5% dividend yield is the sector's closest thing to a statistically cheap asset, gated on whether digital onboarding is restored before it loses retail deposit share.18 IndusInd at 1.5 times is a rural credit-cycle bet wearing a bank's clothes.19 PNB at 0.85 times is priced for the recovery to be cyclical, and the burden of proof sits with the bank.17
The crowding question, and two false positives
One positioning fact changes portfolio risk rather than company quality. HDFC Bank, ICICI Bank and Axis Bank together represent more than 30% of total foreign institutional investor allocation to Indian equities, with net foreign flows into the banking sector running at approximately +$1.2 billion year-to-date in 2026.11 A redemption cycle in emerging-market funds therefore produces correlated selling across all three regardless of their individual earnings, and an investor holding all three has less diversification than the ticker count suggests.
Two categories of apparent beneficiary do not survive scrutiny.
The first is the monoline unsecured lender, valued on total-addressable-market logic β the assertion that Indian personal loans and buy-now-pay-later credit can compound above 25% for a decade because penetration is low. The evidence points the other way. Bureau data shows multi-lender stacking concentrated in exactly this cohort, the RBI has already raised risk weights by 25 percentage points to slow it, and the growth rate consistent with stable loss experience is closer to 12β15%.74 A valuation requiring the higher number is priced on a growth rate the regulator has explicitly acted to prevent.
The second is the sub-scale small finance bank, sold as a high-yield asset book plus a banking licence and therefore a superior compounding vehicle. The economics say otherwise: it pays above-market rates for deposits precisely because it lacks brand and branch density, competes for borrowers against universal banks now equipped with the same digital rails, and lacks capital depth to absorb a credit cycle without dilution. High yield earned on expensive funding through a costly network is a spread that disappears when either input moves.
The competing frame comes from outside India, where the same disruption ran a decade earlier and produced a very different market structure.
9. Global Parallels & Future Game Changers: Nubank Lessons, CBDC Disintermediation, and AI Fraud
Three cities offer three different answers to the question of what happens to banks when technology arrives.
SΓ£o Paulo: what happens when the incumbents are bad
Nu Holdings β Nubank β is the most successful digital bank ever built, and the reason is not primarily technology. By early 2026 it served more than 100 million active customers across Brazil, Mexico and Colombia, with a cost to serve below $1 per active user per month, a net interest margin near 9.8% and return on assets above 4.0%.20 Those figures are extraordinary in absolute terms and require context to interpret.
Nubank grew into an oligopoly. Brazilian retail banking before its arrival was concentrated among a handful of institutions charging fees consumers found punitive, on service that gave customers reason to leave. A challenger offering a no-fee credit card and a decent app was counter-positioned against incumbents who could not respond without cannibalising their own fee revenue β the classic structure in which the attacker's advantage is the defender's unwillingness rather than inability. The 9.8% margin likewise reflects Brazilian rate levels and unsecured consumer spreads rather than a technology dividend; comparing it with an Indian bank's 4% as though the gap measured efficiency would be a category error.
The Indian counterfactual is the interesting part. India never produced a Nubank, and the reason is that the Indian state built the disruption itself. UPI made payments free and interoperable before any private platform could establish a proprietary network. Aadhaar made identity verification a public utility rather than a private data moat. The Account Aggregator framework made financial data portable without creating a data monopolist. Each of these decisions removed a layer of value that, in Brazil or the United States, a private disruptor could have captured. What remained for a private entrant was the deposit franchise and the credit risk β precisely the regulated, capital-intensive, slow-compounding parts.
The market fear of the late 2010s was that Paytm, PhonePe and their peers would disintermediate Indian banks. The outcome was the reverse: those platforms became distribution channels for regulated lenders, constrained by first-loss-guarantee caps and, in one prominent case, by direct regulatory action against a payments bank. India's public infrastructure protected its banks' economics more effectively than any amount of private defensive spending could have.
New York and Tokyo: scale and rates
JPMorgan Chase demonstrates a different lesson, and one Indian bank managements study closely: at sufficient scale, technology investment and physical distribution reinforce rather than substitute for each other. A bank that can spend at global-leader levels on technology while retaining branch density in its core markets builds a position that neither a pure digital entrant nor a pure branch incumbent can attack. The relevant Indian parallel is not any single bank but the emerging structure: the largest three or four private banks are pulling away from the mid-sized field on exactly this combination. We do not reproduce JPMorgan's disclosed financials here; the parallel that matters is architectural rather than quantitative.
Mitsubishi UFJ and the Japanese banks illustrate the opposite constraint. A banking system operating for decades in a near-zero rate environment discovers that deposit franchises are worth very little when there is no spread to earn on them, and that the value of a bank is often a leveraged bet on the direction of policy rates rather than on anything management does. The Indian relevance is a caution: SBI's deposit moat is worth what it is worth because Indian rates are positive and real. In a permanently low-rate India β an unlikely but not impossible future β the most valuable asset in Indian banking would become close to worthless.
Three developments that could reorder the value pool
Central bank digital currency and the deposit franchise. The Reserve Bank has run retail digital rupee pilots for several years. In their current form they are payment instruments with holding limits and no interest, which makes them a substitute for cash rather than for deposits. The mechanism that would matter is different: if the RBI ever permitted interest-bearing retail e-rupee balances, or raised individual holding limits substantially β the frequently discussed threshold being βΉ50 lakh per individual β then a household could hold a risk-free, sovereign-backed, instantly transferable balance directly on the central bank's balance sheet. In ordinary times that would raise banks' cost of funds as low-cost current and savings balances migrated. In a stress event it would be far worse, because a digital run toward the central bank has no friction at all.
Who loses: every bank whose valuation rests on a cheap deposit franchise, which in India means SBI most of all and the large private banks nearly as much. Who gains: nobody in the listed banking universe, though lenders funded predominantly by term deposits and wholesale markets would suffer relatively less because they never had the advantage in the first place. The observable milestones are specific β any RBI move toward paying interest on retail e-rupee holdings, or raising holding limits into the tens of lakhs. Neither has occurred. This belongs in the category of emerging signal rather than investable theme, and the correct posture is monitoring rather than positioning.
OCEN and embedded credit. Treated in Section 7, and worth restating here as the middle-probability, medium-timeline risk. Announced protocols and pilot integrations are not scaled commercial adoption; the milestone that would confirm the shift is embedded non-bank applications accounting for more than 30% of MSME working capital originations.
Generative AI and synthetic identity fraud. This is the most underpriced risk in the digital lending thesis, because it attacks the mechanism the thesis depends on. An instant loan approval system is an automated decision made without human contact on the basis of documents and data. Generative models have made synthetic identity construction and document forgery cheap, scalable and difficult to detect at the point of application. The attack does not need to defeat a bank's underwriting model; it needs only to defeat the verification layer that feeds it.
The mechanism by which this damages economics is worth spelling out. Fraud losses on instant digital loans appear in early-vintage delinquency β defaults in the zero-to-ninety-day window, where a genuine borrower would normally have made at least one payment. A sudden rise in that metric across multiple digital lenders simultaneously is the signature. The response, once detected, is to reintroduce human verification touchpoints: video KYC with live review, physical address confirmation, callbacks. Each of those restores the customer acquisition costs that the digital rail eliminated. The theme's central financial claim β that acquisition costs fell 60β70% and stay fallen β is the claim under attack.
The beneficiaries in that scenario would be lenders with existing customer relationships and verified histories, who need to onboard fewer strangers. The losers would be the digital-first challengers whose entire growth model is acquiring customers they have never met. It would also, ironically, restore some of the advantage of the branch network β the asset the whole industry has spent a decade treating as a cost.
India's public infrastructure has, so far, protected its banks from the disintermediation that reshaped banking elsewhere. What it has not protected them from is the two risks that come from the same direction as the benefit: a sovereign that could compete for deposits directly, and an automation layer that can be attacked by automation.
10. The Monitoring Dashboard & Crux KPIs: The Falsification Map
Everything above reduces to a hierarchy of observables. Structural: currency in circulation as a share of GDP and household net financial savings β annual, lagging indicators of a slow variable. Adoption: Account Aggregator consent volumes, monthly. Industry: the system credit-to-deposit ratio, weekly. Company: slippage rates, quarterly. Market: foreign institutional flows into banking β monthly, noisy and lagging.
Four are load-bearing, set out below with the mechanism that makes each lead rather than lag, the disagreement each discriminates, and the threshold that changes the conclusion.
Crux KPI 1: System credit-to-deposit growth delta
What it measures: the percentage-point difference between year-on-year system credit growth and deposit growth.
Why it leads: it sits directly on the binding constraint. Margins are an outcome; the funding gap is the cause. When credit outgrows deposits by more than about 300 basis points for a sustained period, banks must bid for bulk and term deposits, and the cost reaches net interest margin two to three quarters later. By the time compression appears in reported earnings, the delta has been signalling it for a year.
Which disagreement it settles: the bear argument is that Indian banks must permanently choose between margin and growth because household savings cannot fund 13% credit expansion; the bull argument is that deposit growth accelerates as real incomes rise and the RBI eases reserve requirements. Where it sits: industry level, coincident-to-leading.
Source, cadence and latest reading: Reserve Bank of India fortnightly scheduled bank statement; +2.2 percentage points as at mid-2026, being credit growth of 12.0% against deposit growth of 9.8%.10
Confirm or break: the thesis strengthens if the delta narrows below +1.0 percentage point. It breaks if the delta expands beyond +4.0 points for two consecutive readings, forcing either credit rationing or a step-change in funding costs; the system credit-to-deposit ratio breaking above 83% carries the same message.
Crux KPI 2: Account Aggregator digital credit disbursement share
What it measures: the share of retail and MSME loan origination value processed through Account Aggregator consent rails, in per cent.
Why it leads: it measures whether cash-flow underwriting is being used, as distinct from being available. Every claimed benefit of the theme β lower acquisition cost, lower credit cost, higher return on assets β flows through this rail, and adoption precedes the cost savings, which precede the margin improvement, which precede the earnings.
Which disagreement it settles: the bull case holds that Account Aggregator integration permanently lowers operating costs by roughly 40 basis points and credit costs by 30 basis points; the bear case holds that consent-based sharing works only for prime borrowers who were already documented and adds little where the informal-sector risk sits. If adoption stalls in the mid-teens, the bear case is being demonstrated. Where it sits: adoption level, leading.
Source, cadence and latest reading: Sahamati ecosystem dashboard, monthly and quarterly; approximately 15.2% of retail and MSME origination value as at mid-2026, alongside monthly consent volumes near 12.5 million and cumulative account links above 110 million.2
Confirm or break: confirmed if disbursement share crosses 25% by end-FY27; broken if it plateaus below 18%, indicating friction in consumer consent or bank API integration the technology has not overcome. A stated limitation: this captures value through the formal AA rail and understates lending that uses cash-flow data obtained elsewhere. It is a proxy for cash-flow underwriting adoption, not a census of it.
Crux KPI 3: Net slippage ratio
What it measures: annualised fresh non-performing asset additions, net of write-offs and recoveries, as a percentage of opening gross advances.
Why it leads: it is the earliest point at which asset quality deterioration becomes visible in disclosed data. Gross NPA ratios lag because they are a stock affected by write-offs and recoveries; credit cost lags further because it reflects provisioning policy. Slippages are the flow, and they turn two to three quarters before the headline metrics.
Which disagreement it settles: whether post-COVID unsecured retail and MSME vintages produce a delinquency wave as they season into a weaker income environment, or whether bureau coverage and cash-flow monitoring hold slippages permanently below 1.5%. Nothing in this article's structural argument survives if slippages break out. Where it sits: company level, leading.
Source, cadence and latest reading: bank quarterly disclosures and investor presentations; approximately 1.10% annualised across top private banks and 1.65% for public sector banks in Q4 FY26.
Confirm or break: confirming if top private net slippages stay below 1.25%; breaking if they exceed 2.20% for two consecutive quarters, at which point the credit cost assumption embedded in every valuation in Exhibit 4 is wrong.
Crux KPI 4: Provision coverage ratio, excluding write-offs
What it measures: cumulative bad-debt provisions held on balance sheet divided by gross non-performing assets, in per cent.
Why it leads: it measures the buffer before credit losses reach equity, and it matters most at the moment of regulatory change. The transition to expected credit loss provisioning requires forward-looking provisions; institutions already above 75% coverage absorb it with minimal capital impact, while under-provisioned lenders take a real charge. Coverage today tells you who will be hurt by a rule change that has not yet fully arrived.
Which disagreement it settles: whether the reported balance-sheet strength of Indian banks is genuine or an artefact of a lenient provisioning regime about to be replaced. Where it sits: structural-to-industry level, leading with respect to the ECL transition.
Source, cadence and latest reading: RBI Financial Stability Report, biannual, plus bank quarterly filings; 79.5% systemwide for scheduled commercial banks in early-to-mid 2026.4
Confirm or break: confirming if system coverage holds above 75% through the ECL transition; breaking if it falls below 65%, indicating provisioning has not kept pace with the loan book.
Duplicated bets and hidden factor exposure
A portfolio of Indian bank equities can look diversified by name and be a single position by cause. Three overlaps matter. Holding ICICI Bank, Axis Bank and IDFC First Bank together concentrates exposure to the unsecured retail credit cycle β urban consumer default rates and RBI risk-weight policy β the same bet expressed three ways, on a rule the regulator has already shown willingness to change. Any combination of mid-sized private banks running credit-to-deposit ratios above 85% duplicates exposure to the liquidity and rate cycle; their business models differ, their funding vulnerability does not. And the concentration of foreign institutional ownership in the three largest private banks means a global emerging-market allocation decision moves all of them together, independently of anything happening in Indian credit.11 An investor who wanted Indian financialization and bought global EM risk appetite has the wrong instrument for the thesis.
Theme kill criteria and security kill criteria are frequently confused and are not the same. The theme dies if top-tier private net interest margins fall structurally below 3.20% while cost-to-income ratios fail to improve despite a digital transaction mix above 80%, which would demonstrate the digital dividend was competed away rather than retained; it also dies if system credit-to-deposit breaks 85% for two consecutive quarters, or if top private slippages exceed 2.20%. The security kill criteria are the company-specific tests named in Section 8, and a bank can fail its own test while the theme holds, or hold up while the theme fails around it.
Where the upstream belief stands
Return to the proposition that sent us here: that India's formalization is permanently re-routing household savings into the formal financial system, creating an expanding, low-credit-cost intermediation pool for well-capitalised lenders.
On the evidence available at the end of July 2026, the belief is holding. The asset side is confirmed beyond serious dispute β 15 billion UPI transactions a month, 110 million Account Aggregator links, 440 million bureau-covered borrowers, and gross NPAs at 2.1% against a peak of 11.5% eight years earlier. Cash is not returning; currency in circulation to GDP remains contained near 11.5%, and households are saving in financial form at roughly 11.2% of gross national disposable income.
It is fraying in one place, and it is the place the original framing understated. Formalization has been far more successful at creating borrowers than at creating depositors. The same digital rails that turned an invisible tea vendor into a bankable customer also turned a captive savings account holder into someone who can move βΉ5,000 a month into an equity fund from a phone. Five consecutive years of credit outgrowing deposits is the measurable consequence. The intermediation pool is expanding, as the belief predicted; the cheap part of it is not.
That is why the investment conclusion does not follow automatically from the social forecast. An investor could have been entirely right about India's financialization since 2021 and still lost money by owning the wrong layer β the asset originator rather than the liability holder, the high-margin challenger rather than the low-cost monolith. The theme is real. The scarce resource inside it turned out to be the oldest thing in banking: other people's money, cheaply obtained, and kept.
Glossary
CASA (current and savings account) ratio β the share of a bank's deposits held in current and savings accounts rather than term deposits. These accounts pay little or no interest, so a high CASA ratio directly lowers a bank's cost of funds. In India it is the single best predictor of which banks survive a deposit war with margins intact.
CRAR (capital to risk-weighted assets ratio) β a bank's Tier-1 and Tier-2 capital divided by its risk-weighted exposures. The RBI requires a minimum of 9% plus a capital conservation buffer. It determines how much a bank can grow before it must raise equity.
Credit-to-deposit (C/D) ratio β total outstanding loans divided by total customer deposits. Above roughly 80β85% a bank is funding loans with borrowings rather than deposits, which is more expensive and less stable. It is the industry's central constraint in 2026.
D-SIB (domestic systemically important bank) β banks designated by the RBI as too important to fail, currently State Bank of India, HDFC Bank and ICICI Bank. Designation carries additional capital requirements and, implicitly, a stronger official backstop.
ECL (expected credit loss) β a provisioning framework under Ind-AS 109 requiring banks to provide for losses expected over a loan's life at the point of origination, rather than after a default occurs. Its adoption in India will penalise under-provisioned lenders and barely touch well-provisioned ones.
GNPA / NNPA (gross and net non-performing assets) β loans overdue more than 90 days, expressed as a share of advances. Net NPA subtracts provisions already held. The gap between the two measures how honestly a bank has provisioned.
LCR (liquidity coverage ratio) β the stock of high-quality liquid assets a bank must hold to survive 30 days of severe stress, with a 100% minimum. The RBI's draft revision assumes digitally connected retail deposits run off twice as fast as branch deposits, which raises the cost of being a mobile-first bank.
NIM (net interest margin) β net interest income as a percentage of average interest-earning assets. It measures gross spread, not profitability, and is not comparable across banks with different asset mixes and cost structures.
OCEN (Open Credit Enablement Network) β a protocol standardising loan-origination APIs so that any application can present a credit offer backed by a regulated lender. If it scales, it moves origination economics from banks to customer-facing platforms.
PCR (provision coverage ratio) β provisions held against gross non-performing assets, as a percentage. It measures the cushion between a credit problem and a bank's equity.
PPOP (pre-provision operating profit) β revenue less operating expenses, before credit provisions. It is the cleanest measure of a bank's underlying earnings power because it strips out the most discretionary line on the income statement.
PSL (priority sector lending) β the RBI requirement that 40% of adjusted net bank credit go to agriculture, MSMEs, affordable housing and weaker sections, with shortfalls parked in low-yielding NABARD deposits. It is a real, invisible tax on banks whose natural customer base sits outside those sectors.
References
-
UPI Product Statistics β National Payments Corporation of India ↩↩↩
-
Account Aggregator Ecosystem Data Dashboard β Sahamati ↩↩↩↩
-
Investors: financial results and disclosures β AU Small Finance Bank ↩↩↩↩↩
-
Financial Stability Report β Reserve Bank of India ↩↩↩↩↩↩↩↩↩↩↩↩↩↩↩↩
-
Pradhan Mantri Jan Dhan Yojana progress β Press Information Bureau, Government of India ↩↩
-
Research and Information β Association of Mutual Funds in India ↩↩
-
Weekly Statistical Supplement and fortnightly scheduled bank statements β Reserve Bank of India ↩↩↩
-
Financial Results and Investor Relations β State Bank of India ↩↩↩↩↩↩↩
-
Nu Holdings Ltd. filings, including Form 20-F β U.S. Securities and Exchange Commission ↩
LinkedIn
WhatsApp
Reddit
Hacker News
Email
Kindle
RSS Feed
Spotify
Apple Podcasts
Amazon Music
Audible
YouTube