पाठशाला Pathshala · ग्राहक Grāhak, The customer · Lesson 26 · Scale
Customer research at scale: a hundred conversations a quarter
At fifty people the founder can no longer hold every customer call. A research operation with its own panel, honest incentives, plain consent and one synthesis rhythm keeps the whole company close to users.
Pathshala, The Founder Library · 11 October 2026 · 7 min read
In the first year every founder talks to customers because there is nobody else to do it. By fifty people the calls have quietly stopped, and the company is building from a memory of what customers said two years ago.
This lesson sets out how to rebuild customer contact as an operation rather than a habit: a panel you own, incentives that do not buy the answer, consent that meets the DPDP rules, a hundred conversations a quarter spread across the people who build, and one synthesis rhythm that turns them into decisions. A figure works the recruitment arithmetic backwards from the target.
What changes at fifty people
Three things break at once. The founder is no longer on every call, so what customers say no longer reaches the person who decides. Product managers and designers multiply, and each runs a study when a feature needs one, so research arrives in bursts tied to launches. And the same twenty friendly customers get asked everything, because they are the ones who reply.
The fix is a cadence rather than a project. Teresa Torres, who coined the term continuous discovery, sets the bar plainly in Product Talk: good product discovery teams engage with customers at least weekly, interviewing and testing assumptions week over week. Jakob Nielsen made the economic case for many small studies a quarter of a century ago: his analysis finds five users uncover about 85 per cent of the usability problems in a design, and recommends three studies of five over one study of fifteen.
A hundred conversations a quarter is that cadence at the size of a fifty-person company: six or seven people who build, each holding about two conversations a week. It is not a sample for a survey. It is enough contact that nobody in product goes a fortnight without hearing a customer in their own words.
Recruitment: a panel you own
Recruiting is where research at scale usually dies. Nielsen Norman Group’s ResearchOps 101 lists finding, screening, booking and paying participants as the first of six areas of research operations and the most urgent pain point for overloaded teams. Solve it once with a panel: a table of customers and prospects who have agreed to be contacted for research, with their segment, plan, city, language, last contact date and how they prefer to be reached.
Fill it from four places. An in-product invitation shown to a sample of active users. The CRM, by segment, for accounts you need and do not hear from. Recent leavers, whose stories the [churn interview](/library/churn-interview-learning-from-those-who-left) collects. Lost deals, who know exactly why they chose someone else. Cap contact at once a quarter per person and twice a year per account, or the panel tires and the answers start to sound rehearsed.
Screen for behaviour, not opinion. A screener that asks “do you manage inventory?” recruits everyone; one that asks “how many purchase orders did you raise last week?” recruits the people you need. Recruit by language and city deliberately. A panel drawn only from English-speaking users in Bengaluru and Mumbai will describe a product that most of your users do not have; the lesson on [research in Tier 2 and 3 India](/library/research-in-tier-2-and-3-india) covers the field logistics.
Incentives that do not buy the answer
An incentive pays for time, not for a view. Fix it by the participant’s time and seniority, not by how much you want the study to succeed, and pay it after the conversation whatever was said. For consumers a gift voucher or a UPI transfer works; for business users many prefer a donation in their name or nothing at all, and some companies forbid employees from accepting payment, so ask.
Three rules keep incentives from distorting the research. Never tie payment to a product outcome, such as upgrading or posting a review. Never let the incentive be large enough that people lie their way through the screener; when one does, the screener question was too easy to guess. Keep a ledger of every payment with the participant’s name and date, so finance can account for it and so nobody is paid twice in a quarter. Ask your accountant how payments to individuals should be recorded before the first quarter, not after.
Consent under the DPDP rules
A research conversation collects personal data: a name, a phone number, a recording, often a screen with other people’s data on it. The Digital Personal Data Protection Rules were notified on 14 November 2025, and the government’s explainer says notices must be clear and easy to understand, must state the specific purpose for which data is collected, and that consent can be withdrawn at any time. The rules carry an eighteen-month phased compliance period. Checked October 2026.
For research that translates into four habits. Send a one-paragraph consent note before the call, in the participant’s language, saying what will be recorded, why, who will see it and how long it will be kept. Ask again on the recording. Store recordings in one place with access limited to the research team. Delete them on the date you promised. The lesson on [the DPDP Act](/library/dpdp-act-what-it-requires-of-your-product) covers the product side; research is the part teams forget.
At an 8 per cent reply rate, half passing the screener and 70 per cent showing up, a hundred conversations needs about 3,600 invitations. Double the reply rate, which a warm panel easily does against a cold list, and invitations halve. Six interviewers carry about 1.3 conversations a week each, which is sustainable. Two interviewers carry nearly four, which is a part-time job and the reason most research programmes collapse in their second quarter.
Research at scale is a recruitment problem first and an interviewing problem second. Own the panel and the conversations follow.
Synthesis: from a hundred calls to five decisions
A hundred conversations that live in a hundred personal notes documents are worth less than twenty in one place. Give every call the same record: who, segment, date, the job they were doing, verbatim quotes and a handful of tags from a shared codebook. Tag within twenty-four hours, while the call is fresh. The lesson on [turning twenty interviews into one decision](/library/twenty-interviews-to-one-decision) sets out the coding method; at scale the change is that one person owns the codebook and merges duplicate tags every month.
Name an owner for the operation. At fifty people this is rarely a full-time researcher; it is a product manager or designer with a third of their time given to the panel, the incentive ledger, the repository and the quarterly readout. When that person becomes overloaded, the first full-time research hire should be in operations rather than in running studies.
Once a quarter, hold one readout: ninety minutes, every product lead and the founder in the room. Show the tags that rose and fell since last quarter, play five short clips, and end with no more than five decisions, each with an owner and a date. A readout that ends in “interesting” has wasted a hundred people’s time.
A worked example: a lending app in Mumbai
A fifty-person company runs a credit app for self-employed borrowers. Its founders interviewed customers every week in the first year; by the third, the last study was six months old and had been run by one designer for one feature. The product team set a target of a hundred conversations a quarter across seven interviewers: four product managers, two designers and the head of risk.
The first quarter missed. Invitations went to a cold list of all borrowers, replies ran at 3 per cent and show-ups at half, so the team held 41 conversations, most of them in English and in Mumbai. In the second quarter it built a panel from an in-app invitation in Hindi, Marathi and English, offered a ₹500 UPI transfer after the call and sent a WhatsApp reminder the evening before. Replies rose to 14 per cent and show-ups to 80 per cent, and it held 104. The readout produced four decisions. The largest came from borrowers in smaller cities who described repaying through a relative’s UPI account, which the app’s collection flow did not allow for.
The quarterly research calendar
Week one: set the quarter’s questions with product leads, no more than three, each tied to a decision on the roadmap. Refresh the panel and check every contact cap. Weeks two to twelve: each interviewer holds two conversations a week and tags within a day; the owner checks the reply, show-up and tagging rates every Friday. Week thirteen: merge tags, hold the readout, record the decisions, pay any outstanding incentives and delete recordings that have reached their date.
Track four numbers on one line each quarter: conversations held, share of interviewers who held at least ten, days from call to tag, and decisions made. If decisions are falling while conversations rise, the questions were not tied to anything the company was going to decide.
The funnel rates in the figure are illustrative. The DPDP position was checked in October 2026; nothing here is legal advice.
Sources
- Teresa Torres, Product Discovery Basics: Everything You Need to Know, Product Talk — Good discovery teams engage with customers at least weekly, interviewing and testing assumptions week over week.
- Jakob Nielsen, Why You Only Need to Test with 5 Users, Nielsen Norman Group (2000) — Five users find about 85% of usability problems; spend the budget on three studies of five.
- Kate Kaplan, ResearchOps 101, Nielsen Norman Group (2020) — Six areas: participants, governance, knowledge, tools, competency, advocacy; recruitment is the most urgent pain point.
- Press Information Bureau, Digital Personal Data Protection Rules, 2025 (explainer, 17 November 2025) — Rules notified 14 November 2025; clear notices stating purpose; consent withdrawable any time; eighteen-month phased compliance. Checked October 2026.