पाठशाला Pathshala · ग्राहक Grāhak, The customer · Lesson 08 · Start

Field research: watching customers instead of asking them

People describe the job they think they do. Sit beside them for two hours and you see the job they actually do, workarounds and all. How to run a shadowing session in a kirana, a clinic or an office.

Pathshala, The Founder Library · 11 October 2026 · 7 min read

A pair of antique binoculars lying on a detailed road map — Photograph: Ylanite Koppens · Pexels

Ask a kirana owner how he keeps track of credit and he will say he writes it in the khata. Sit on the stool beside him from seven to ten one evening and you will see the khata, three chits, two phones, a call to his son and twenty minutes spent matching UPI alerts to names. The product is in the part he did not mention.

This lesson explains why watching beats asking, where to sit, how to ask for the session, what to write down and how to turn two hours of notes into a finding. It ends with a weekly habit.

Why watching beats asking

The [customer interview](/library/the-customer-interview-done-properly) is the right first tool. It finds the problem, the person who feels it and the money attached to it. It is a weak tool for the shape of the work, because people describe their routine as they believe it to be, with the friction sanded off. The steps they no longer notice are exactly the steps a founder needs to see.

Usability research learned this early. Jakob Nielsen’s rule, set out in 2001, is watch what people actually do, do not believe what they say they do, and be most sceptical of what they predict they will do. His evidence is blunt. Across 113 pairwise comparisons of interface designs that he and Jonathan Levy analysed, the correlation between how well people performed and which design they said they preferred was 0.44: stated preference explains only a fraction of actual performance.

The method has a name. Contextual inquiry, as the Nielsen Norman Group describes it, is an ethnographic field study built on a master and an apprentice: the customer is the craftsman doing their real work in their real place, and the researcher is the apprentice who watches, asks and learns. Its four principles are context, partnership, interpretation and focus. In plain words: go to where the work happens, let them lead, check your reading with them, and know what you came to learn.

Companies that took it seriously built it into the calendar. Intuit’s founder Scott Cook, according to a 2004 account in the ABA Journal, used to wait in software stores and ask buyers of Quicken whether he could follow them home and watch them set it up. The practice became the company’s Follow Me Home programme, and its point, in the words of a former manager quoted there, was getting engineers out for one-on-one contact with customers.

Where to sit: the kirana, the clinic, the office

Choose the stretch of work where your problem lives, and go when it is busiest, not when it is convenient for you. A kirana or a trading shop: the evening, when credit is given, collected and reconciled, or the morning delivery from the distributor. Sit on the customer side of the counter for the first hour and move behind it only when invited. A clinic or a diagnostic lab: the morning rush at the front desk, from opening until the first doctor’s lunch. Stand where you can see the screen and the register at once. Patient records are private, so watch the process and never copy a name. An office: one complete cycle of the work, such as a payment run, a month-end close or a hiring week, beside the person who does it rather than the manager who describes it.

Two hours is the minimum that gets past the performance; most people tidy up their behaviour for the first thirty minutes and then forget you are there. The NN/g guidance allows anything from an hour or two to several days. For an early founder, two to three hours in each of six places beats a full day in one.

Asking for the session

The ask is small, specific and honest. Not “can I observe your business”, which sounds like an audit, but “could I sit with you on Thursday evening while you do the khata? I am trying to understand how credit really works in a shop like yours. I will not sell you anything and I will not get in the way.” Offer something useful in return: help with the work, a summary of what you saw, a cup of tea. Ask the owner, then ask the person whose work you will watch, separately. A munim or a receptionist who was volunteered by the boss will perform for you.

Bring a notebook, not a laptop; a laptop on the counter changes the room. Take photographs only of objects and only with permission: the khata page with names covered, the chit, the shelf of files. Never record customers or patients.

What to write down

Use a page with five columns and fill one row for every step. Time, to the minute, because duration is the number you will price against. Step, in a verb: writes, flips back, phones, matches, re-enters. Tool, whatever is touched: the khata, a chit, a phone, the software, a calculator, a person. Workaround, marked whenever the step exists only because something else does not work: a paper diary kept beside the software, a call to confirm a payment the app should have confirmed, a spreadsheet that tracks what the accounting package already knows. Interruption, whenever the work stops for something else, and for how long.

Ask questions only between steps. “Why did you flip back three pages there?” after the customer has left the counter, not while he is counting change. The NN/g guidance is to ask about the usual steps and the unusual variations and why each happens, and to avoid over-validating your own interpretation during the session because it nudges what the person does next. At the end, read your timeline back to them and let them correct it. That is the partnership principle, and it is also where the best sentences come out: “haan, wo chit wala kaam toh roz hota hai.”

Customers describe the job they think they do. The workaround they have stopped noticing is the job they will pay you to remove.

A worked example: the khata in Lucknow

A founder building a credit-tracking tool for kiranas has done fifteen interviews. Every owner said the same thing: credit goes in the khata, collection happens at the end of the month, the problem is customers who do not pay. The founder’s product is a reminder app for the customer.

An elderly shopkeeper sits at a desk in a small grocery shop stocked with onions and packaged goods. — A shopkeeper at his desk in a small grocery shop in India. The ledger the owner describes is rarely the whole system: look for the chits and the second phone. Photograph: Aravind P.S · Pexels

Then she spends one evening on the stool beside a shop owner in Aliganj, seven to ten. Writing credit in the khata takes twelve minutes across the evening. But four other things take longer. He flips back through pages to find a customer’s old balance before giving more credit, nine minutes. He calls his son twice to ask whether a UPI payment has come into the son’s account, six minutes. He scrolls the UPI alerts on his own phone trying to match a sender’s name to a khata name, fourteen minutes, and gets two wrong. He keeps a separate chit for people who promised to pay tomorrow, because the khata has no column for a promise.

Half the evening’s bookkeeping is reconciliation, not collection. The customers mostly do pay. The owner cannot tell who has paid, because the money arrives on two phones under names that do not match the khata. The founder changes the first version of the product from reminders to matching: a screen that pairs each incoming UPI credit with a khata name and asks the owner to confirm with one tap. Nobody had asked for it in fifteen interviews. Nobody had to.

Turning a session into a finding

Within twenty-four hours, while you can still read your handwriting, write three things. First, the timeline: every step, end to end, with minutes. Second, the workarounds, each with its time and the reason it exists. Third, one surprise: the thing you did not expect, in one sentence. If there is no surprise, the session was either too short or you watched the wrong stretch of work.

After three sessions in similar places, lay the timelines side by side. A workaround that appears in all three is a pattern. A workaround that appears once is a habit of one person. The [next lesson](/library/twenty-interviews-to-one-decision) covers how to code and count across many conversations; field notes go into the same table as interview notes, with one difference: what you watched outranks what you were told.

There is a second dividend. Paul Graham’s essay on doing things that do not scale describes founders who installed their product on the customer’s laptop on the spot instead of sending a link. A founder who has sat through a payment run can do the same: offer to run next month’s with the customer, on the prototype, beside them. Field research and the first sale are often the same afternoon.

One session a week

Put one shadowing session in the calendar every week for the first six weeks of a new product or a new segment, and one a month after that. Before each session, write the stretch of work you will watch and the one question you want answered. During it, fill the five columns and ask only between steps. Within a day, write the timeline, the workarounds and the surprise, and send the customer a short thank-you with what you learned. Every fourth week, lay the timelines side by side and circle the workaround that keeps appearing. That circle is the next thing to build or the next thing to test.

Ask permission for everything, record nothing private, and leave the place as you found it. The people who let you watch are doing you a favour.