पाठशाला Pathshala · ग्राहक Grāhak, The customer · Lesson 20 · Build
Research in tier-2 and tier-3 India without flying everywhere
The next hundred million customers live in towns a founder cannot visit every month. Reach them by phone in their language, through partners they trust and a WhatsApp panel, and travel only where a call falls short.
Pathshala, The Founder Library · 11 October 2026 · 7 min read
Most Indian startups research their next market from a co-working space in Bengaluru, interviewing people who look like their team. The customers they plan to win next speak another language, pay another way and trust different people, and most of them are a long train ride away.
This lesson sets out a research system that reaches those customers without a founder spending half the month travelling: desk work first, then vernacular phone interviews, local partners, a standing WhatsApp panel, and visits planned for what only a visit can show. A figure compares the cost and the calendar of flying everywhere with a mixed plan.
Why the metro sample misleads
The users who will drive growth are mostly not English-first. A 2017 study by KPMG in India and Google counted 234 million Indian-language internet users against 175 million English users in 2016, and expected nine of every ten new users over the following five years to be Indian-language users. In its survey 68 per cent of internet users considered local-language content more reliable than English content. The numbers have moved since; the direction has not.
A sample of metro English speakers therefore answers a different question from the one you are asking. They describe their own habits: cards, apps, delivery in ten minutes, reviews read in English. A shop owner in a district town, a homemaker in a tier-3 city or a farmer’s son at college in a tier-2 one may share none of those habits. Each segment needs its own small sample, and the [lesson on choosing your hundred million](/library/india-1-2-3-choose-your-hundred-million) is the place to decide which ones matter.
Start at the desk: data before travel
Before calling anyone, look at what the data already says. Your own orders, sign-ups and support tickets by pin code show where demand exists without marketing. Public aggregates help too: PhonePe Pulse publishes quarterly digital payment data with state and district scoreboards, which shows where digital spending is growing fastest and which districts behave unlike their neighbours. Census and state economic surveys give household size, language and literacy by district.
The output of the desk stage is a short list: six to ten towns grouped into two or three clusters that differ in a way that matters to the product (language, income, whether the buyer is a shop or a household). Every later step samples from those clusters. Without them, research drifts towards whoever is easiest to reach.
Write the script once in plain English and have a native speaker of each language rewrite it, not translate it. The word for a subscription, a loan or an app differs by town, and a question that sounds neutral in English can sound like a sales pitch in Marathi or a test in Tamil. Pilot the script on two calls per language and change whatever made people hesitate. Keep the recruitment message short and specific: who you are, why you are calling, how long it takes, what you will send as thanks, and that nothing will be sold. Send it on WhatsApp from a number with a name and a photograph, because an unknown number from another state is the first thing a cautious customer ignores.
Vernacular calls, done properly
The phone interview is the workhorse. Three rules decide whether it works. The interviewer speaks the language natively, not through a translator; a translated interview loses the hesitation, the joke and the word the customer uses for your category, which is often the most useful thing in the call. The call is short and at the customer’s time: thirty minutes, booked in advance, often in the evening or on Sunday. The thank-you is real and immediate: a small amount sent by UPI at the end of the call, which settles the question of whether the interviewer can be trusted.
Use the same interview discipline as anywhere else: past behaviour, not opinions about the future, as the [lesson on Indian politeness](/library/mom-test-for-indian-politeness) sets out. Ask the customer to describe the last time they bought something like your product: where, from whom, how they paid, who they asked first. Record the call only with consent. The Digital Personal Data Protection Rules, notified on 13 November 2025, bring the Act’s core consent obligations into force on 13 May 2027 and require a consent notice to itemise the personal data and state the purpose. Build the habit now: say in the customer’s language what you will record, why and for how long, and keep the recordings out of shared drives. Checked October 2026.
How many calls? Jakob Nielsen’s usability rule for distinct user groups is three to four people per group when there are two groups and three each when there are more. Research for discovery needs more than usability testing, but the principle carries: a few calls per cluster find the patterns, and more calls in one cluster add less than a few calls in a new one.
Local partners who already have the trust
Cold calls to strangers in a small town go unanswered. Introductions do not. Every town has people whom your customers already trust and who talk to dozens of them a week: the chartered accountant who files their returns, the owner of the coaching centre, the pharmacist, the agent who sells their insurance, the distributor who supplies the shop. One of these partners can recruit ten interviews in a week that a founder could not get in a month.
Treat partners as part of the research team. Brief them on whom you need and whom you do not. Pay per completed interview, not per name. Listen to a sample of every partner’s interviews in the first month, because a partner who recruits only friends or coaches answers will bias everything downstream. Keep the best two or three partners per cluster for the year; their knowledge of the town compounds.
WhatsApp communities as a standing panel
Once a cluster has been interviewed, keep the people close. A WhatsApp community, which since November 2022 has let one admin send updates across several groups of up to 1,024 members each and run polls inside the chat, turns a few hundred customers into a panel you can ask a question on Monday and hear back from by Wednesday. Voice notes in the customer’s language carry more than typed answers, and people who would never fill a form will talk for two minutes.
Run the panel with rules. One question a week at most, so it stays a privilege and not a chore. Separate groups by cluster so that a loud town does not drown the others. Share back what you changed because of their answers. And remember that panel members become friendlier to the product the longer they stay: rotate a third of the members each quarter and keep fresh interviews running alongside.
At eight towns and eight interviews per town, visiting every town at ₹45,000 a visit costs ₹3.6 lakh and four weeks of a two-person team. Visiting two towns and phoning the other six at ₹1,200 an interview costs about ₹1.5 lakh and three weeks, and the founders are in the office for most of it. The saving matters less than what it buys: the same budget can now cover three clusters instead of one.
Fly to learn what only a visit shows. Call to learn everything else, in the customer’s own language.
When you must go in person
Some things never come through a phone. How the shop is laid out and where your product would sit. Who stands at the counter and who sits behind it. What the household does with the phone at eight in the evening. The [lesson on field research](/library/field-research-watching-instead-of-asking) covers how to watch rather than ask. Plan visits after the calls, not before: the calls tell you which two towns will teach you most, and the visit tests the patterns the calls found rather than starting from nothing.
Make each visit dense. Two days, eight to twelve interviews in homes or shops arranged in advance by a partner, an evening with the partner going through what you saw, and a written note before the train leaves. A visit that produces only photographs and a good feeling has been wasted.
The monthly field rhythm
Each month: refresh the desk view, orders and sign-ups by pin code and any new public data. Complete a quota of phone interviews per cluster, for example five each, by native speakers, with consent recorded and payments made. Ask the WhatsApp panel one question and post back one change it caused. Listen to three recorded calls as a team in a weekly slot, with subtitles or a colleague translating live. Decide whether a visit is needed next month and to which town, from what the calls could not explain.
Each quarter, check the clusters against the business: are the towns you research the towns where growth is coming from? If not, move the research to where the customers are.
Data protection timelines were checked in October 2026 and change; take advice before recording or storing customer interviews at scale. Nothing here is legal advice.
Sources
- KPMG in India and Google, Indian Languages: Defining India’s Internet, April 2017 — 234 million Indian-language and 175 million English internet users in 2016; nine in ten new users expected to be Indian-language users; 68% find local-language content more reliable.
- PhonePe Pulse, quarterly digital payments data with state and district scoreboards
- Khaitan and Co, ERGO: Digital Personal Data Protection Rules, 15 November 2025 — Rules notified 13 November 2025; core consent provisions from 13 May 2027; notice to itemise data and purposes. Checked October 2026.
- Jakob Nielsen, Why You Only Need to Test with 5 Users, Nielsen Norman Group, 18 March 2000 — Three to four users per group for two groups; three per group for three or more.
- WhatsApp, Communities now available, 3 November 2022 — Admin updates across groups, groups of up to 1,024 users, in-chat polls.