India Venture Capital Report 2017 — The Transplant Problem

India's venture market had grown substantially through the mid-2010s, funded largely by foreign capital applying models proven in the US and China. 2017 is when the limits of that approach became clear.

The central problem was arithmetic rather than executional. As the 2017 Asia-Pacific report describes, revenue per user is bounded by income. A business model requiring a customer to spend a certain amount annually is viable only where customers can afford that amount. Where they cannot, no amount of execution makes the model work.

Many Indian consumer businesses had been built on models imported from markets with several times the income per head. The models had been adapted — lower price points, different features — but frequently not enough, because the required adaptation was not a discount but a different economic structure.

The symptom was widespread discounting. Businesses acquired customers by subsidising transactions, funded by venture capital. Growth was rapid. The question, asked with increasing seriousness through 2017, was what happened when the subsidy stopped.

The answer separated two kinds of business:

• Those where the discount had accelerated adoption of a genuinely valued service. Customers stayed.
• Those where the discount was the value proposition. Customers left.

Distinguishing them is straightforward in principle — measure retention after discounts end — and was frequently not done, because doing it required ending the discounts, which meant accepting slower growth in a market where growth was what raised the next round.

Underneath this, a structural development was changing the landscape: public digital infrastructure. India was building identity, payments and data-sharing as public utilities rather than private platforms, which changes where value accrues in financial services in a way that invalidates theses transplanted from elsewhere.

The income arithmetic

The constraint deserves setting out precisely because it is frequently treated as a growth problem rather than as a structural bound.

Revenue per user cannot exceed what users can pay, and what they can pay is bounded by income. This is not a marketing constraint or a product-market fit problem. It is arithmetic.

Costs do not scale down proportionally:

• Some costs are broadly global. Engineering talent competes internationally. Cloud infrastructure is priced globally. Content licensing frequently is too.
• Some costs are local and genuinely lower. Customer service, delivery, local operations and office space scale with local wages.
• The mix determines viability. A business with mostly local costs can work at lower revenue per user. One with mostly global costs cannot.

The consequence is that a model imported from a higher-income market requires more than a price adjustment. It requires a different cost structure or a different revenue structure — usually both.

The models that worked in India shared characteristics:

• Very large scale with small per-transaction economics, monetising volume rather than value.
• Advertising or commerce margin rather than direct payment from users.
• Financial services, where lending and payment economics can be attractive at modest transaction sizes.
• Business-to-business models serving companies rather than consumers, since business spending is less income-constrained.
• Services exported to higher-income markets, capturing global revenue on local costs — the strongest version of the arithmetic and the basis of the IT services industry.

A model built for a market with several times the income per head is not a model that needs adapting. It is a model that needs replacing, and the difference is usually discovered after several years of funding.

What discounting actually measures

Discount-funded growth was the defining feature of Indian consumer venture in this period, and it is worth being precise about what it does and does not demonstrate.

What a discount demonstrates. That customers will use a service at a subsidised price. This is real information — it establishes that the service is usable, that the operational model functions, and that demand exists at some price.

What it does not demonstrate. That customers value the service at its unsubsidised price. That is the question the business exists to answer, and a discount actively prevents answering it.

The critical measurement is retention after the discount ends, and it separates two situations that look identical while the discount is running:

• The discount accelerated adoption of a valued service. Customers tried it because it was cheap, discovered they valued it, and continued at full price. The subsidy was a customer acquisition cost — expensive but genuine investment.
• The discount was the value proposition. Customers used it because it was cheap. When it stopped, they stopped. The subsidy purchased usage, not customers, and the spending is a loss rather than an investment.

Why this was frequently not measured: ending discounts means accepting slower growth, and in a market where growth determined the next round's valuation, slowing growth was expensive in the near term. So businesses continued subsidising, and the question went unanswered until capital availability forced it.

A related error compounded it. Businesses reported customer acquisition cost including the discount, and lifetime value based on behaviour observed while the discount was running. Both numbers are wrong in the same direction, and their ratio — the standard measure of unit economics — is therefore wrong twice.

The general principle: any metric measured under a subsidy describes the subsidised state, not the underlying business. This applies to discounting in India in 2017, to free tiers, to promotional pricing, and to any situation where the customer is not paying the price the business needs them to pay.

The addressable market is not the population

A specific error in market sizing was common and deserves stating plainly.

The error: treating the population as the addressable market. India's population is very large, and a small percentage of a very large number is still a large number, which makes almost any business look viable in a slide.

Why it is wrong: a service requiring meaningful spending is addressable only to the segment with that spending capacity. That segment is far smaller than the population, and it is the correct denominator.

The correct approach is to size by spending capacity:

• Identify the annual spending the business requires from a customer.
• Estimate how many households have that discretionary capacity, which requires income distribution data rather than population data.
• Adjust for the addressable share of those households — geographic reach, digital access, category relevance.

The resulting market is frequently one or two orders of magnitude smaller than a population-based estimate, and it is the number that determines whether the business can reach the scale its funding assumes.

The distinction matters enormously for capital planning. A business funded on a population-based market estimate will raise capital sized to an outcome it cannot reach. The capital is deployed, growth is pursued toward a ceiling that arrives earlier than modelled, and the shortfall appears as a plateau that is read as an execution failure. It was a sizing error, made years earlier.

The same analysis applies to any large-population, lower-income market, and it is why per-capita and income-distribution data are more useful inputs than population figures.

Public infrastructure and where value accrues

The most consequential structural development in this period had nothing to do with any individual company.

India was building digital public infrastructure: a national identity system, an interoperable real-time payments layer, and frameworks for consented data sharing. These were built as public utilities — available to any participant on equal terms — rather than as private platforms.

Why this changes where value accrues:

In markets where payment infrastructure is privately owned, controlling the rails is a source of durable advantage. Card networks capture a fee on every transaction, and much of the value in payments accrues to whoever owns the network. A fintech thesis in such a market frequently involves either building rails or capturing a position adjacent to them.

Where the rails are a public utility, that advantage is unavailable. Any participant can access the payment layer on equal terms. Value must be created in services built on top — which means competing on product, distribution and customer relationship rather than on infrastructure ownership.

The consequences:

• Payment businesses face structurally lower margins, because the interchange economics that fund payment businesses elsewhere do not exist in the same form.
• The cost of building financial services falls dramatically, since identity verification, payment and data access are available as utilities rather than requiring investment.
• Competition intensifies, because barriers to entry are lower — which is the point of public infrastructure and a problem for anyone underwriting a moat.
• Value accrues to distribution and to lending, rather than to transaction processing.

A fintech thesis built on owning the rails does not transfer to a market where the rails are a public road. The advantage being underwritten is one the architecture does not permit.

This is a genuine structural difference, not a temporary condition, and it means investment frameworks developed in private-rails markets systematically misidentify where the durable positions are in India.

The models that do work

A report that explains why imported models fail owes an account of what works instead. The Indian companies that reached durable scale share identifiable characteristics, and they are consequences of the income arithmetic rather than exceptions to it.

Very large scale with small per-transaction economics. A business capturing a small margin on an enormous volume of transactions does not require any individual customer to spend much. The constraint moves from revenue per user to total transaction volume, which a large population does supply. The requirement is that the cost per transaction is genuinely small, which usually means the operation must be substantially automated.

Financial services. Lending economics work at modest ticket sizes because the revenue is a spread on the amount rather than a fee for a service. A small loan generates meaningful revenue relative to its cost of origination once origination is cheap — which the public digital infrastructure made possible. This is the clearest case where the infrastructure changed what was viable rather than merely what was cheaper.

Business-to-business. Business spending is not constrained by household income. A company selling software or services to Indian enterprises faces a market sized by corporate budgets, which is a different and less binding constraint. The same applies to businesses serving small merchants, where the spending is a business input rather than consumption.

Services exported to higher-income markets. The strongest version of the arithmetic: global revenue on a local cost base. This is the structure of the IT services industry and it inverts the constraint entirely — the income level that limits domestic revenue is the same level that provides the cost advantage.

Advertising at scale. Monetising attention rather than payment. This requires very large user numbers to produce meaningful revenue, which means it works for the largest platforms and not for most.

What these have in common is that none requires an individual customer to spend more than they can afford. Each finds revenue in volume, in a spread, in a business budget, or abroad. The models that failed all required the customer to pay an amount the addressable segment could not sustain — and no amount of capital, product quality or execution changes that.

What an allocator could act on

Compute the required annual spend per customer and check it against income distribution. This is the single most useful diligence question for any Indian consumer business, and it is answerable from the company's own unit economics and free World Bank income distribution data. If the required spend is above what the addressable segment can sustain, the model has a ceiling that will arrive regardless of execution.

Ask for retention after discounts end. Any metric measured under subsidy describes the subsidised state. Customer acquisition cost including the discount and lifetime value based on subsidised behaviour are both wrong in the same direction, so their ratio is wrong twice. The only informative number is what happens when the customer pays the price the business needs.

Distinguish the acceleration reading from the substitution reading. A discount that accelerated adoption of a genuinely valued service is a customer acquisition cost. A discount that was the value proposition is a purchase of usage. These look identical while running and are cleanly separated afterwards.

Do not import a payment-rails thesis. Where the rails are a public utility, the advantage that anchors fintech value elsewhere is unavailable by design. Value accrues to distribution and to lending. An investor underwriting infrastructure ownership in this market is underwriting something the architecture does not permit.

Weight the exit route heavily in the return model. In 2017 this was the binding constraint, and its subsequent development — described in the 2021 India report — is the most important change in the market over the following decade. An investment's achievable return depends on the exit route existing, and in 2017 it largely did not.

What 2017 established for India

• Income arithmetic bounds business model viability, and imported models require replacement rather than adaptation.
• Discount-funded growth measures price sensitivity, and any metric measured under subsidy describes the subsidised state.
• The addressable market is set by spending capacity, not population — an error that appears years later as a plateau.
• Public digital infrastructure removed the rails advantage, invalidating transplanted fintech theses.
• The exit environment was the binding constraint, and its subsequent deepening is the most important change of the following decade.