03. Market Research

Most people misunderstand market research.

They think it means surveys, interviews, reports, and spreadsheets. They imagine it as something you do after you have an idea, to validate it. In practice, that’s mostly noise. Real market research happens much earlier, and it looks much simpler.

It looks like drawing a map.

Market Research Is Cartography

The most useful form of market research is not asking users what they want. It’s understanding where you are. Founders often jump straight into building because they’ve found a problem that feels real. The problem is that real problems exist everywhere. What matters is not whether a problem exists, but where it sits in the larger landscape. Before you design a solution, you should be able to answer a basic question: What does this entire industry look like if I step back and view it all at once?

This is harder than it sounds. Industries are usually described as lists: competitors, segments, subsegments. Lists hide structure. What you want instead is geometry. When you force an industry into a 2D or 3D space, something interesting happens. You are forced to choose axes. And axes are opinions. They encode how value is created, how money flows, and how users experience pain. Choosing the wrong axes produces a useless map. Choosing the right ones makes opportunities obvious.

Example 1: Education

Take education. At first glance, it’s messy: schools, colleges, coaching centers, online courses, degrees, certificates, content platforms. But if you choose three simple axes, the chaos collapses into structure.

1. Age group served (Who?): Pre-school (< 5 years), K12 (5-18 years), Test Prep 1 (JEE, NEET, CET) (15-18 years), UG and PG (17-22 years), Test Prep 2 (CAT, UPSC, PCS, Bank, Job Interviews) (22-25 years), Upskilling (25-35 years) (6 possibilities).
2. Distribution model (How?): B2B, B2B2C, B2C, Marketplace (4 possibilities).
3. Outcome type (What?): Learning, Career, Operational, Discovery (4 possibilites).

Now imagine this as a 3D grid. Every education company occupies one or more cells in this space. Some cells are overcrowded. Some are empty. Some are small. Some are expanding rapidly. Empty cells are not automatically good. Crowded cells are not automatically bad. What matters is why a cell looks the way it does. You don’t need revenue numbers yet. Just seeing the space clearly already gives you an edge.

Good axes usually satisfy three conditions:

1. They are easy to explain to a non-expert.
2. They correspond to real economic behavior.
3. They don’t change every year.

If your axes require a paragraph to justify, they’re probably wrong. If they’re fashionable terms, they’ll age badly. If they’re internal jargon, they won’t generalize. The best axes sound boring.

What the Map Reveals

Once you draw the map, several non-obvious questions become easy to ask:

• Why are some combinations missing?
• Which cells have many companies but weak products?
• Which cells have strong demand but awkward distribution?
• Which outcomes are users paying for reluctantly, and which eagerly?
• Where does regulation distort the map?
• Where does technology make new shapes possible?

Without the map, these questions never arise. This is why market research done late feels useless. By then, founders are emotionally attached to a single point in the space. The map threatens their beloved idea, so they avoid drawing it. Good founders do the opposite. They delay attachment and invest heavily in understanding the terrain.

Another reason why founders skip this step is because it feels unproductive. There’s nothing to demo. No code gets written. No users get interviewed. It doesn’t feel like progress. But this is the phase where most leverage exists. A small insight at the mapping stage can save years of execution in the wrong direction. Conversely, a brilliant product in a poorly understood cell of the map often dies quietly.

Example 2: HRTech

Consider recruitment. One obvious axis is the employee lifecycle: Hiring phase, Tenure in the company, Offboarding/Exit.

This alone is insufficient. You need at least one more dimension. Possible second axes include:

• Who pays: Employer, Candidate
• Signal source: Resume-based, Portfolio-based, Network-based, AI-inferred

Pick one. Plot it. Suddenly, you’ll notice patterns:

• Heavy tooling around hiring, very little around exit
• Sophisticated software for engineers, crude processes for frontline roles
• High spend where hiring is rare, low spend where churn is constant

Again, none of this requires talking to users yet. It requires thinking clearly.

The Idea Generation Mechanism

I don't consider myself a naturally insightful person when it comes to generating original ideas. Therefore, I think if I can create a system, a structured process, that can aid me in the discovery and identification of innovative concepts. The correct order, that works for me, is:

1. Map the industry
2. Notice tensions, gaps, and asymmetries
3. Form hypotheses
4. Then generate ideas

If you start with ideas, market research becomes justification. If you start with mapping, ideas emerge naturally. The best startup ideas often feel obvious after the map is drawn. Before that, they’re invisible. That’s the point.

Market research is not about proving your idea is good. It’s about making sure you’re standing in the right place before you start walking.

SEAT Analysis

Ideas flow better when you do structured analysis of the industry.For each area on the industry landscape, you should ask questions in four sets: Structural, Execution, Outcome, and Timing (SEAT). It's more of an MBA-style business exercise, than engineering or product exercise.

1. Structural Analysis

1. Market Size: Who pays? Average ARPU Tolerance? TAM/SAM/SOM? Growth Rate?
2. Customer Insights & Spend Readiness: Behavioural Patterns? What they seek? When they buy?
3. Regulatory Intensity: Are approvals required for distribution? Are degrees/certifications restricted?
4. Sources of Variance: Regional, academic, social, economic, if any?
5. Competitive Density & Moatability: Number of funded players? Points of pairty (content, tech, GTM)? Switching costs? What moats are possible?

2. Execution Analysis

1. Distribution Efficiency: CAC vs ARPU? Sales cycle lengths? Channel power (schools, coaching institutes)? Digital readiness of the target segment?
2. Learning Modality Fit: Does the learner have the ability/discipline for self-paced? Or you need offline infra? How much does the outcome depend on mentorship or feedback loops?
3. Technology Leverage vs. Operational Complexity: Can AI/automation dramatically reduce cost of delivery? Can personalization improve outcomes non-linearly? Can tech deliver a 10× better experience over incumbents? Tech must somewhere act as a pillar to differentiation.

3. Actual Outcome Analysis

1. Outcome salience: How painful is the problem? How urgent is the problem? How much visible is the problem?
2. Outcome verifiability: Can you prove your product works? Scores or metrics? Observable productivity metrics?

4. Timing Analysis

1. Timing & Trends: Why now? What has recently changed? Tailwinds? Is this a pre-inflection, at-inflection, or post-inflection market?
2. Capital Intensity Fit: Does this space require long runway? Physical Infra? Regulated approval costs?

Indian Macros

The following data is something which we shall use for performing market analysis. For example, population demographic data is useful in estimating TAM. So, this is the page for India-specific data, to be readily used in guesstimates.

Population Demographics

The first question that comes is: what's the agewise distribution of the population? We have heard that India has a lot of young population. Interestingly, 50.94% population in India is less than 30 years old!

The next important question is, how much is the earning per household? The estimate was made as follows.

Step 1. The average number of people per household is 4.53. Now, if we sort the households based on the economic status, the number of people per household in top deciles would be a bit lesser than the bottom deciles. NFHS (National Family Health Survey) report has shown that the number of people in top 25% of households is around 3.8 to 4.2, compared to 4.5 to 5.0 for the bottom 25%.

Let \(\bar{h} = 4.53\) be the average number of people per household. Assume we divide the households in 10 deciles and \(h_i\) be the average number of people per household for the \(i^{th}\) decile. If we assume, that the number decreases linearly, then we can say \(h_i \approx a + (i-1)\frac{b-a}{9}\). We know that, \(\frac{\sum_{i=1}^{10}{h_i}}{10} = \bar{h}\). This gives, \((a + b) = 4.53 \times 2 = 9.06\). From NFHS data, we can estimate \(a = 5.4\) and \(b = 3.66\). This gives us the following estimate.

Step 2. We know certain facts about the India's income distribution. Gross National Income (GNI) is 3.85 trillion USD \(\approx\) 338.8 trillion rupees. Note that this number includes the corporate profits as well. The Gross Household National Income (GHNI) is 2.90 trillion USD \(\approx\) 258.27 trillion rupees ($2,000 income per capita). However, we will consider GNI for our calculations because the corporate profits eventually benefit the directors and shareholders.

Oxfam International in their report Income Inequality in India: Survival of The Richest shared that the bottom 50% of people take 15% of national income, bottom 90% take 43% of national income, top 10% takes 57%, top 1% takes 17% and top 0.1% takes 5%. This data is sufficient to fit the Lorenz curve. By calculating the area under this curve, we can estimate the income share of any percentile range of the population. By running this experiment, we found out the following.

These numbers are super close with the Blume's Indus Valley Report 2025!

Step 3. We can actually estimate the number of people in each age group for each economic category. A quick estimate would be \(n_{ij} = d_i \times n_j\) where \(d_i\) is the distribution percentage for \(i^{th}\) age group and \(n_j\) is the number of people in \(j^{th}\) economic category. However, this estimate would be slightly flawed as in poorer households, there are higher number of children due to high fertility rates and in rich households, old people are higher due to better medical facilities and lifestyle promoting longevity.

The crude idea goes like this: the national average of people per household is 4.53; if the household size is larger than this, it must be a kid. If it is smaller, it should favor senior citizens (a lot of empty nesters and seniors living alone in urban areas). So, we define kid's factor = average household size in that strata / national average household size. Similarly, senior citizen factor = 1 / kid's factor. This gives us the following factor values.

These factors form extremes. We should linearly spread them across age-groups.

While these correction factors capture the essence of our idea, multiplying them directly will result in the population to blow up because \(\sum_{i}{d_i \cdot k'_{ij}} \neq 1\), where \(d_i\) is the national distribution (proportion) of people for the \(i^{\text{th}}\) age group and \(k'_{ij}\) is the correction factor for the \(i^{\text{th}}\) age group and \(j^{\text{th}}\) economic strata. So, we derive the proportions as.

And using the population of each economic strata, we get the following results (number of people in lakhs).