How to run continuous discovery interviews

The teams who say continuous discovery is "not realistic for our context" are almost never disagreeing with the principle. They agree that talking to customers every week would help. They have read the book, drawn the opportunity solution tree on a Miro board, and quoted Teresa Torres at a planning meeting. They also have a calendar, and the calendar is the part that breaks.

This is a working guide on how to run continuous discovery interviews when the live-call cadence does not fit your team. What the practice actually requires, why the prescribed weekly rhythm fails for most product trios on the second month, and the operational shape of running it async with voice notes so the cadence survives quarter two.

What continuous discovery interviews are

Continuous discovery interviews are a recurring product-research practice in which the same product team holds short, regular conversations with the people they are building for, not as a one-off study but as a standing rhythm. Teresa Torres' canonical definition is "at minimum, weekly touchpoints with customers, by the team that's building the product, where they conduct small research activities in pursuit of a desired product outcome." Four parts, and any one of them missing produces something that looks like discovery but is not.

The practice sits inside a larger framework. The product trio (product manager, designer, engineer) listens together. The output of each touchpoint is an interview snapshot that adds opportunities to an opportunity solution tree. The team picks one opportunity per cycle, generates solutions, and runs assumption testing on the solutions before deciding what to build. The interviews are the input to the tree, not the only research activity that happens, and the Interaction Design Foundation has a clean overview of the wider system.

The reason the literature focuses on the interview habit and not the rest of the framework is that the interview is the part that almost always slips. Trees do not draw themselves; opportunities do not surface from analytics. Without the touchpoint, every other part of the loop runs on memory.

Why the weekly cadence usually breaks

Most teams who try continuous discovery interviews fail in the second month, not the first. Week one and two are clean: the team has just finished the book, recruited a couple of friendly customers, and held two thirty-minute calls. By week six, two of the trio members are heads down on a release, the calendar Tetris has gone fragile, the recruiter has not had time to refill the pool, and the standing weekly slot becomes "we'll do it next sprint." The cadence does not collapse all at once. It thins.

The diagnosis is almost always the same. Live interviews are an expensive unit of research per minute of insight, and the unit cost is paid by the rarest people on a product team. A 30-minute call needs a confirmed slot from three trio members and one customer. That is four calendars, two time zones, one reschedule per round. The weekly version of that costs more time than the team has on its quietest week and impossibly more time on a launch week. Torres' own answer to this in talks is unsentimental: if the team has to hustle to find a customer every week, they will not do it. The recruiting pipeline has to be automated, and so does the conversation surface, or the habit dies on contact with a normal product calendar.

The voice-first async version of continuous discovery interviews changes one variable: the conversation moves off the calendar. Customers answer one or two short prompts on their own time, in their own voice. The team listens together at a fixed weekly slot that does not require anyone except the trio to be free. Recruiting moves from "find a customer who has thirty minutes Wednesday" to "send the prompt to the standing pool every Monday." The rest of the framework stays the same. The opportunity solution tree, the trio, the assumption testing, the snapshots: all unchanged.

"If they have to hustle to find a customer to talk to every week, they will not do it."

We cover the broader case for async in our note on async user research methodology. For continuous discovery specifically, the operational version is shorter: voice keeps the cadence alive past week six.

How to run continuous discovery interviews, step by step

Six steps. They are the same steps you would run for live continuous discovery, with two of them rebuilt for async. The order matters. Skipping step one is the most common failure mode and produces a habit that runs forever without ever changing what the team builds.

01 · Pick the assumption, not the topic

Most teams write a prompt before they write the question they are trying to answer, and end up with a habit that produces interesting transcripts and zero product decisions. Reverse the order. Each week, name the one assumption your team is currently betting the roadmap on. Phrase it as a falsifiable sentence: we believe new users abandon at import because the column matching is unclear, not we want to learn about onboarding.

The opportunity solution tree pins each assumption to one opportunity, and the assumption is what the prompt has to test. If your team cannot name the assumption in one sentence on Monday morning, the interview week is going to produce a transcript and no decision. Do the assumption work first, before opening the prompt builder. The companion piece on how to write user research questions covers prompt craft once the assumption is set.

02 · Frame one prompt that fits sixty seconds

The async unit is a single open prompt the participant answers in roughly thirty to sixty seconds of speech. That is one well-formed question, not a list. "Walk me through the last time you tried to import a CSV. Don't summarize, tell me the day" works. "What do you think about our import experience?" produces a clean two-sentence summary that is useless for testing the assumption.

Three rules carry over from the live version and matter even more async, because there is no interviewer in the room to follow up:

• Anchor to a specific recent moment. "The last time you...", not "in general...". Memory generalizes; specific moments do not.
• Avoid the leading version of the question. "Was the import confusing?" presupposes the answer. "Tell me what happened the last time you imported" does not.
• Add one optional follow-up, not five. A short second question for the participants who want to keep talking, marked optional, is the right shape. Five prompts in a row is the shape that kills response rates.

The tighter the prompt, the cleaner the signal on the assumption. We get into prompt-craft tradeoffs in the longer guide to research questions; for the weekly habit, the rule of thumb is one prompt, sixty seconds, one assumption.

03 · Recruit a small standing pool, not a one-off cohort

In a one-off study you recruit ten participants for one moment. In continuous discovery interviews you recruit a small standing pool you can re-poll across weeks. Twenty to forty people, opted in, segmented by the personas your trio cares about, refilled every quarter so the pool does not staleness-bias the data.

The recruitment screener has to filter for two things, in order: relevance (right user segment, right context) and willingness to be re-contacted (a checkbox at signup, an explicit opt-in to a "research panel" or whatever your legal copy calls it). Without the second filter, the second week of the habit becomes a recruiting project again, and a recruiting project does not fit inside a normal product week. Torres calls this the difference between an automated pipeline and an artisanal one. The automated version survives.

A practical default: target a 20-person standing pool, send the weekly prompt to all twenty, expect six to ten async voice responses back per week. That is enough to listen to as a trio in a single hour and dense enough to surface a real signal on most assumptions.

04 · Set the cadence: weekly, async, recoverable

The cadence is the contract. Pick a fixed Monday "send" slot and a fixed Thursday "listen-as-a-trio" slot. The participant has Monday through Wednesday to answer in their own time, on their own device. The trio meets Thursday for an hour to listen together. That single shared hour is non-negotiable; everything else flexes. Move the listen hour rather than skip it.

Weekly is the right default. Bi-weekly drifts to monthly drifts to nothing. If your team genuinely cannot do weekly for the next quarter, do not pick fortnightly: pick weekly with a one-prompt minimum, and accept that some weeks will produce a thin transcript. A thin transcript on schedule beats a rich transcript six weeks late. The ones that produce nothing carry the most signal: a participant pool gone quiet on a prompt is information about the prompt.

05 · Run each week as a thread, not a study

Each week is a thread, not a study. The trio listens together, in the same room or call, with the audio playing out loud and the transcript on the screen. No-one writes a summary slide. Two artifacts come out of the listen hour: an updated opportunity solution tree (one or two opportunities added, removed, or moved) and one interview snapshot per participant whose response carried weight. That is it.

The temptation, especially for teams new to the habit, is to over-process: cluster transcripts, write up findings, present at the next planning meeting. Resist it for the weekly cadence. Synthesis happens at the assumption layer, not at the interview layer (next step), and writing a polished week-six findings deck is the activity that ate week seven. We cover the per-question coding pass in more depth in analyzing user interview transcripts; for the weekly habit, the lighter version is enough.

06 · Synthesize at the assumption, not at the interview

Once a quarter (not weekly), the trio reviews the standing assumption set and asks one question per assumption: do the last twelve weeks of interviews support, contradict, or fail to address it? Three buckets, one decision per assumption. Assumptions that have been contradicted are de-prioritized on the tree; assumptions that have failed to surface in the interviews need a different prompt next quarter; assumptions that are well-supported get promoted to actual product work.

That quarterly review is where continuous discovery becomes a research practice rather than a habit. Without it, the weekly listening drifts into entertainment: thirteen pleasant Thursdays of trio bonding that produced no roadmap moves. With it, every assumption on the tree has been examined against twelve weeks of customer voice before it survives or dies.

The product trio listening problem

The Torres framework is explicit that the trio (PM, designer, engineer) listens together. In the live-call version this is the most expensive part: three calendars, one customer, all on the same Zoom. Most teams quietly downgrade it to "the PM listens and reports back," which loses the entire point. Designers and engineers who have not heard a customer in the customer's own voice do not build with the same frame as ones who have.

Async voice notes change the trio listening cost without changing the principle. The recordings sit in a shared inbox. The Thursday listen hour is the only thing that needs to be on three calendars, and it does not need to coincide with a customer's calendar at all. We have watched teams who were running PM-only continuous discovery for a year switch to async and run their first real trio listen the following week, because the constraint that had been keeping it solo was scheduling, not appetite. Our broader case for what voice catches that text loses covers why hearing a customer in their own voice changes the trio's frame.

When continuous discovery interviews stop working

Three failure modes worth naming in advance, because they are the patterns we see at month four, not month one.

The pool gets stale. A panel of twenty customers is dense for the first quarter and tired by the third. Refresh roughly a third of the pool each quarter. Participants who answered every week for three months tell you something about your most engaged customers, not your representative ones.

The prompts converge on what the team already believes. Once the team has heard six weeks of evidence for an assumption, the temptation is to write next week's prompt to confirm it. Resist. Each week's assumption should be one the team is genuinely uncertain about. If the trio cannot name a real bet on the question, pick a different one.

The trio shrinks to a duo. Engineering or design quietly stops attending the listen hour. This is the early warning that the cadence is dying. The fix is operational: protect the hour like a planning meeting, not like a research meeting. If the listen hour is not a calendar fixture by month two, it will not be one at month six.

The wider voice-first user research playbook covers the cadence patterns we have seen across teams, and the failure modes are mostly operational rather than methodological. Continuous discovery interviews fail for the same reasons standing meetings fail. The shape that works is one that costs less to keep alive than to skip.

FAQ

What is a continuous discovery interview?

A continuous discovery interview is a short, recurring conversation between a product team and one of its customers, held as part of an ongoing weekly habit rather than a one-off study. The team uses each conversation to test an assumption on the opportunity solution tree, and the trio (product manager, designer, engineer) listens together. The async version replaces the live call with a voice note answered in the customer's own time, while keeping the weekly cadence and the trio listening intact.

How often should continuous discovery interviews happen?

Weekly, at minimum. This is the cadence Teresa Torres prescribes in Continuous Discovery Habits, and it is the threshold where the habit produces enough signal to move the roadmap. Bi-weekly drifts to monthly drifts to nothing within a quarter. If the team genuinely cannot run weekly with a live-call format, switch to async voice prompts before lowering the cadence; the operational cost of an async week is small enough that even a busy launch week can carry one prompt.

How many participants do I need for continuous discovery?

A standing pool of twenty to forty opted-in customers is enough for most product trios. Each weekly prompt typically returns six to ten voice responses, which is dense enough for a trio to listen to in one hour and broad enough to surface real signal on a single assumption. For thematic saturation across a quarter, the same logic Guest, Bunce and Johnson found for interview research applies: six to twelve responses on a homogeneous group is usually enough to read the pattern. Refresh roughly a third of the pool each quarter to avoid staleness bias.

What is the difference between continuous discovery and a one-off user interview study?

A one-off study is a short, intense burst of research scoped to a question, run end to end in a fortnight, and then stopped. Continuous discovery interviews are a standing habit: same standing pool, weekly prompts, quarterly synthesis, no defined end. One-off studies go deeper on a specific decision; continuous discovery keeps the team in steady contact with customer reality. Most teams use both, with the habit running in the background and one-off studies inserted when a major decision needs more depth. Our voice user interview guide covers the one-off interview shape.

Can continuous discovery work without live calls?

Yes, and for most teams it works better. Live calls are the cadence-killer for continuous discovery: the calendar cost falls on the rarest people on the team, and the habit usually dies in month two when calendars get tight. Async voice prompts move the customer side of the conversation off the calendar, while keeping the trio listen hour intact. The framework Torres describes is medium-agnostic; the prescription is weekly touchpoints by the team, not weekly Zooms.

What is the product trio in continuous discovery?

The product trio is the product manager, the designer, and the engineer who form the core decision-making unit on a product team. The Torres framework prescribes that the trio listens to customer interviews together, on the grounds that designers and engineers who have not heard the customer in the customer's own voice do not build with the same frame as ones who have. Async voice keeps the trio listening practice alive when scheduling three live attendees on every customer call is not realistic.

---

Continuous discovery interviews are not a research method that lives or dies on the choice of question. They live or die on the cadence, and the cadence lives or dies on the operational cost of one week. If running the habit live is what your team can afford, do that. If the calendar has been winning, switch the customer side of the conversation to async voice and keep the trio listen hour. The framework is the same. The unit cost drops by an order of magnitude. Talkful has a free plan that is enough for a first month of weekly prompts, and the end-to-end voice user research guide covers the wider habit once the cadence is in place.