Continuous Discovery: How to Talk to Customers Every Week
Most teams research in bursts, then build on stale assumptions for months. Here is how to run Teresa Torres's weekly customer habit without it eating your roadmap, and how to keep what you learn.
Share
Continuous discovery is the habit of talking to customers at least weekly, by the team building the product, through small research activities aimed at one outcome. Teresa Torres, who set out the practice in Continuous Discovery Habits, treats weekly as the minimum. Making it last takes three things: recruiting that runs itself, a protected weekly slot, and a place where every insight stays linked to the decision it informs.
Key takeaways
Torres defines continuous discovery as weekly touchpoints with customers by the team building the product, in pursuit of a desired outcome (Product Talk).
You can start with one 20–30 minute conversation a week; Torres says teams often complete three or four interviews in the first month and then don't want to stop (Product Talk, 2024).
Recruiting is where the habit dies. One team emailed about 5,000 free-plan users and got about three interviews; an in-app invitation booked six in its first week (Product Talk, 2024).
The bigger risk is forgetting. A weekly habit produces 50 or more conversations a year; without a system they decay into anecdotes.
AI is good at synthesis and scheduling, and already widely used for analysis. It is not a substitute for the conversation itself.
What is continuous discovery?
It replaces the research project with a research habit. Instead of a six-week study before a big bet, the team talks to customers, tests prototypes and runs small experiments every single week (Product Talk).
Three conditions make it work:
The team talks to customers directly. Not through a research report, a persona deck or a sales team's paraphrase. Second-hand insight loses the detail that changes a decision.
Activities are small. A 20-minute interview, one prototype test or one quick survey. It has to fit alongside shipping or it won't happen.
It is anchored to an outcome. You aren't collecting interviews for a library. You are chasing a specific result, such as faster activation or lower churn, and using customer contact to find the path.
It sits naturally beside delivery: discovery decides what's worth building while delivery builds it. Our guide to dual-track agile covers how to run the two side by side.
Why talk to customers every week instead of every quarter?
Because the gap between learning and building is where waste hides. If you only talk to customers quarterly, you spend up to three months making decisions with no fresh feedback. Weekly, you still remember what you shipped and why, and you can change course before a wrong assumption turns into a quarter of work.
The cost of building without that feedback is well documented. Pendo's analysis of 615 product subscriptions found that 80% of features in the average product are rarely or never used (Pendo, 2019). And 43% of the 385 VC-backed startups CB Insights could diagnose after shutting down since 2023 cited poor product-market fit (CB Insights, 2026). Neither number proves weekly interviews would have saved them, but both show how often teams build for customers they've stopped listening to.
How do you set up a weekly customer interview habit?
Start smaller than you think. Torres's advice is to find the tiniest first step and improve on it week by week, rather than design the perfect programme first (Product Talk, 2024).
Pick one outcome. "Raise week-one activation from 30% to 40%" gives every interview a purpose. "Learn about our users" doesn't.
Protect a 45-minute slot every week. Twenty to thirty minutes for the interview, 15–25 for notes (Product Talk). If the calendar is full, cancel one low-value meeting and put the customer there.
Automate recruiting so a participant appears without anyone chasing (details below).
Ask for stories, not opinions. "Tell me about the last time you…" beats "Would you use…?" every time. Our customer interview template has a 20-minute script.
Write a one-page snapshot straight after each call: quick facts, one memorable quote, a sketch of the story, and the needs and pain points you heard.
Add what you heard to an opportunity solution tree. The outcome sits at the root, customer opportunities below it, then solutions, then assumption tests (Product Talk).
Run one small assumption test a week on the most promising solution. Our guide to assumption mapping shows how to pick it.
Review monthly. Which opportunities grew? Which assumptions died? What changed on the roadmap because of it?
Here is what the habit can look like for a small team. It is a template, not a rule:
When
Activity
Time
Monday
Check the week's booked interview; confirm the outcome you're probing
10 min
Tuesday
Interview plus snapshot, with at least two of PM, designer and engineer present
45 min
Thursday
Update the opportunity tree together; pick or adjust this week's assumption test
30 min
Friday
Check the assumption test result; note any decision it changed
15 min
That is about 1 hour 40 minutes a week, or roughly 4% of a 40-hour week for each person involved.
How do you recruit customers for interviews every week?
Make the calendar fill itself. Product Talk's playbook for automated recruiting comes down to five moves: invite people at a relevant moment, such as just after they use a feature or contact support; let them self-schedule with a booking link; use in-the-moment intercepts; let sales and support route interested customers to the same link; and block recurring calendar time so interviews don't get bumped (Product Talk).
Two published cases show the difference it makes:
Zonar Systems put an in-app pop-up on its home page offering a $35 gift card, linked to a booking page. It went live at midnight; almost a dozen interviews were booked by about 5 a.m., and the team has since run about 20 interviews with no manual recruiting (Product Talk, 2022).
HiveMQ had emailed about 5,000 free-plan users and got roughly three interviews. In-app invitations booked six in the first week. No-shows hit 50% in some early weeks, which the team cut by recruiting more experienced users, adding screener questions, sending reminders and offering only slots in the next few days (Product Talk, 2024).
Don't interview the same customer every week. You are looking for variation, so rotate through segments and keep a simple log of who you've spoken to and when.
How do you stop customer insights from decaying?
The real bottleneck isn't talking; it's remembering. A weekly habit produces 50 or more conversations a year, plus tickets, calls and survey answers. Notes scatter across docs, Slack and people's heads, and three months later nobody can find the quote that should have killed a feature.
Four habits keep evidence usable:
One snapshot per interview, in one place. Same format, same location, every time.
Link every insight to its source. A theme without quotes and a count ("6 of 14 interviews") is an opinion.
Link every decision to its evidence. When you change the roadmap, record which interviews or tests drove it.
Separate observation from interpretation. "She exported the report to a spreadsheet every Friday" is an observation; "she needs a dashboard" is a guess.
AI helps most here. In a 2026 Nielsen Norman Group survey of 521 practitioners, 67% of those who analyse research data said they use AI often or always for it (NN/g, 2026). Use it to transcribe, tag and cluster, and insist that every summary links back to the original words. For a fuller method, see how to stop insight decay between interviews and tickets.
Can AI run continuous discovery for you?
AI can carry the busywork: scheduling, transcription, tagging, first-pass synthesis and drafting the next week's interview guide. It can also run short interviews. AI interviewers ask structured questions from a link, so people answer on their own time, which suits a weekly cadence when calendars are the constraint. Our guide to AI-moderated customer interviews covers when they beat a live call and when they don't.
What AI cannot do is the "team talks to customers directly" part. The same NN/g survey found 61% of practitioners had never used AI to interview users, and only 11% had used AI interviewers. And simulated "synthetic users" are not customers at all: treat their answers as hypotheses, never as evidence.
What does a weekly habit look like over eight weeks?
The following is illustrative, not a real company.
Say a five-person SaaS team wants to raise the share of trial accounts that invite a teammate in week one from 22% to 35%.
Weeks 1–2. An in-app invitation after a user's third session books two interviews a week. The trio runs four interviews and learns that admins fear inviting colleagues into a half-configured workspace.
Weeks 3–4. Four more interviews. Three of eight people describe that fear unprompted; two mention needing a manager's approval. Both become opportunities on the tree.
Weeks 5–6. Assumption test: a "share a read-only preview" link, built as a feature stub. 41 of 180 new trial admins click it in two weeks (23%).
Weeks 7–8. The team ships the preview link properly. Week-one invites reach 29%, short of the 35% target, so the approval problem moves up the tree for the next cycle.
Time spent: roughly the 1 hour 40 minutes a week from the table above, for each of the three people involved.
Where does the weekly habit break down?
Recruiting friction. If booking a call takes a week of back-and-forth, weekly becomes monthly becomes never. Automate it before anything else.
Only the PM talks to customers. The designer and engineer hear it second-hand and the detail is lost.
No outcome. Interviews without a target outcome produce interesting stories and no decisions.
Same friendly customers every week. Variation is the point.
Insight decay. Interviews happen, notes scatter, and nothing links them to what the team builds.
If you genuinely can't reach customers weekly, Torres's advice is to start with what you have: analytics, session recordings or an internal proxy such as support. They aren't perfect, but they are better than nothing (Product Talk, 2024). For a one-off deep question, a time-boxed discovery sprint or a two-week discovery loop can complement the weekly rhythm.
Where Specky fits
Specky is built for the remembering part. Every interview, support ticket and call flows into a Product Graph, so a point a customer made three weeks ago surfaces when it becomes relevant to today's decision, with its source attached. Alex, Specky's research agent, runs consent-first interviews from a shareable link and files transcripts, themes and sentiment back into the graph, so the weekly habit doesn't depend on calendars. The Opportunity Tree holds what you learn, and specs drafted in the PRD Editor cite the evidence inline.
Frequently asked questions
What is continuous discovery?
Continuous discovery is a product practice in which the team building a product talks to customers at least weekly and runs small research activities, such as interviews, prototype tests and quick experiments, in pursuit of a desired outcome. Teresa Torres set out the practice in her book Continuous Discovery Habits. The aim is a steady flow of small, evidence-backed decisions instead of occasional large research projects.
How many customer interviews should a product team do each week?
At least one. Torres's benchmark is weekly customer touchpoints, and a single 20–30 minute interview a week is a sound place to start. Teams that automate recruiting often settle at one to three a week. More is not always better: if you can't write up and use what you heard, add a synthesis habit before adding interviews.
Who should attend the weekly interviews?
The people who will build the solution: typically the product manager, designer and an engineer, often called the product trio. At least two should attend each interview, one asking questions and one taking notes. Hearing customers first-hand keeps detail that summaries lose and stops discovery from becoming one person's opinion relayed to the rest of the team.
What is an opportunity solution tree?
An opportunity solution tree is a visual map Teresa Torres uses to organise discovery work. The desired outcome sits at the root. Below it are opportunities (customer needs, pain points and desires found in interviews), then possible solutions for each opportunity, then the assumption tests that check which solutions will work. It keeps weekly learning connected to one outcome.
How is continuous discovery different from dual-track agile?
Dual-track agile describes the structure: a discovery track and a delivery track running in parallel. Continuous discovery describes the habit inside the discovery track: weekly customer contact by the team, small activities and a target outcome. You can run dual-track agile with quarterly research, but then it isn't continuous.
GDPR-aligned. EU workspace data is stored in Frankfurt, Germany. GDPR and EU data Built in Vienna, Austria. A product of influence solutions fm GmbH (FN 675374y). Imprint 14-day trial. Card required, and nothing is charged until day 14. 30-day refund after subscribing. See pricing