पाठशाला Pathshala · ग्राहक Grāhak, The customer · Lesson 13 · Build
Customer development does not stop at launch
The interviews that found the product usually stop the week it ships. Keep five customer conversations a week and a log the whole team reads, and the company keeps learning as it grows.
Pathshala, The Founder Library · 11 October 2026 · 7 min read
Most founders talk to customers obsessively until the product ships and then, without anyone deciding it, stop. Sales calls replace interviews, dashboards replace conversations and within a year the company knows its customers mainly through support tickets and a few loud accounts.
This lesson sets out a cadence that survives launch: five conversations a week, recruited automatically, attended in rotation and written into one log the whole team reads. It ends with the Monday ritual that makes the log worth keeping.
Why discovery stops the week the product ships
Steve Blank’s definition explains why it should not stop. A startup, he wrote in 2010, is an organization formed to search for a repeatable and scalable business model. Shipping the first version answers one question in that search: will anyone use this. It leaves open the questions that decide the next two years. Which customers stay. What they would pay more for. What made the ones who left leave. Who else has the same problem and has not heard of you.
Three forces end the conversations anyway. The calendar fills: after launch there is always a bug, a hire or an investor update that feels more urgent than a call with a customer who is not complaining. Data feels like a substitute: a dashboard shows what people did, and it is tempting to believe it also shows why. The founder becomes the only channel: early interviews were done by one or two people, and when they get busy nobody else has the habit or the permission.
Paul Graham’s essay on doing things that do not scale has the warning in one line: your initial model of users is always inaccurate, even if you are one of them. The same essay notes that when a company is big enough to resort to focus groups it will wish it could still go to its users’ homes and offices and watch them use the product. The point of a cadence is to make sure you never have to wish.
Five conversations a week, and who they are with
Teresa Torres defines continuous discovery as weekly touch points with customers by the team building the product, in pursuit of a desired outcome. Two words in that definition matter. Weekly means a rhythm that does not depend on there being a question; the question arrives when the rhythm is already running. The team building the product means engineers and designers, not only the founder and the sales lead.
Five a week is the house rule of thumb for a company with a product in the market and fewer than fifty people. It is enough to hear a pattern within a month and small enough to survive a bad week. Fix the mix so the sample does not drift towards the friendliest customers:
Two recent sign-ups, within their first month, to learn what they hired the product for and where the first week hurt. One heavy user, to learn what they would miss most and what they have built around you. One customer who went quiet or downgraded, to hear the problem before it becomes a cancellation; the [churn interview](/library/churn-interview-learning-from-those-who-left) handles the ones who have already gone. One prospect who chose someone else or chose nothing, to learn what the market thinks you are.
Each conversation takes thirty minutes and follows the discipline of the [first interviews](/library/the-customer-interview-done-properly): ask about the last time something happened, not about what they would do; ask what it cost; do not pitch. The difference after launch is that the customer has used the product, so the past you ask about includes last Tuesday in your own software. Ask them to share their screen and show you.
Recruiting that runs without you
Cadences die at recruiting. A founder who has to email ten customers every Friday to fill five slots will stop by the third month. Torres’s advice is to automate recruiting, and her guide names three ways that work.
Ask inside the product at a natural moment: right after a customer finishes a task, completes a payment or closes a support conversation, show a one-line invitation to a thirty-minute call with a self-scheduling link. Let customer-facing staff route people: a support agent or account manager who hears an interesting story sends the same link. Protect the slots: put five recurring thirty-minute blocks in the calendar of whoever leads interviews that week, and treat them as you would a board meeting.
For Indian customers adjust the channel and keep the principle. A WhatsApp message from a named person, sent within an hour of a support conversation closing, fills more slots than an email. Offer the call in the customer’s language and at the hour their business allows: a kirana owner is free after the lunch rush, a clinic owner before the evening OPD. A small thank-you, such as a month’s credit on their plan, is fair; cash for opinions brings the wrong people.
The shared log
A conversation nobody else reads is a private memory. The log turns it into company knowledge. Keep one sheet or one database, not a folder of documents, with one row per conversation and the same fields every time:
Date, customer and segment. Who led and who listened. Three verbatim quotes, the customer’s words, not a paraphrase. One surprise: the thing you did not expect, written in a sentence. Codes: two or three short labels from the shared list the [synthesis lesson](/library/twenty-interviews-to-one-decision) describes, so the rows can be counted. Link: to the recording or notes, kept only with the customer’s consent and deleted on the schedule your privacy notice promises; the [DPDP lesson](/library/dpdp-act-what-it-requires-of-your-product) covers what that notice must say.
Fill the row within a day, while the conversation is fresh. The person who listened writes it, not the person who led, because the listener took the notes and because writing is how a listener learns. Ten minutes a row is enough.
Who should hear a customer, and how often
The log helps everyone; hearing a customer’s voice helps more. An engineer who has watched a shopkeeper squint at a screen while serving a queue makes different decisions about font size and loading time than one who has read a ticket about it. The rota below shows how fast that exposure spreads. Rotate one colleague into each call as a silent listener and change who it is every time.
Read the figure at the default settings. A team of fourteen with five conversations a week and one listener on each has everyone through the rota within three weeks, and every person hears a customer at least once a month. Drop to two conversations a week and it takes seven weeks; grow the team to forty without adding listeners and some people go two months without hearing anyone. The fix is cheap: add a second listener, not a second interviewer.
Two cautions. More than two listeners turns a conversation into a panel and the customer stops talking. And a listener is not a participant: they keep their camera on, introduce themselves in a sentence and save their questions for the last five minutes.
Launch answers whether anyone will use the product. Only customers can answer the questions that come after it, and they answer only the people who keep asking.
What the cadence is for, as the company grows
The questions change with the company and the cadence should follow them. In the months after launch the conversations are about activation and the first month: what customers expected, where they got stuck, what made them come back. Later they are about expansion and price: what else they use, what they pay for it, what they would pay you to replace it. Later still they are about new segments: the customers who found you by accident and the ones who looked and left.
Set one question per quarter, written at the top of the log, and choose the mix of five to serve it. Keep a fifth of the time open for whatever customers want to talk about. The best surprises arrive in that open fifth, and a cadence that only answers your questions will not hear them.
The Monday ritual
Thirty minutes every Monday, with everyone who builds the product. Read the week’s rows aloud, quotes first; five rows take ten minutes. Count the codes for the month so far and note any that crossed three mentions. Name one decision or one question the rows raise and give it an owner and a date. Check the rota: who listens this week, and are the five slots booked.
Once a month, read the month’s surprises together and ask which of them should change the quarter’s question. Once a quarter, count how many conversations actually happened against the sixty-five the cadence promises. Below forty, the cadence is failing and the fix is recruiting, not willpower.
Five conversations a week is two and a half hours of founder time. It is the cheapest research a company will ever buy, and the only kind its competitors cannot copy.
Sources
- Steve Blank, What’s A Startup? First Principles, 25 January 2010
- Paul Graham, Do Things That Don’t Scale, July 2013
- Teresa Torres, Continuous Discovery (glossary), Product Talk — Weekly touch points with customers by the team building the product.
- Teresa Torres, Automate Your Recruiting (glossary), Product Talk — In-product opt-in, routing by customer-facing teams, self-scheduling, protected calendar slots.