Discovery Is a Weekly Habit

The Core Insight

Teresa Torres ran product and design at AfterCollege, a job board for college students. Nearly every job board asked students what type of job they wanted and where, and most students had no answer to either question. Under that interface, 36 percent of visitors started a search. Her team asked about school, area of study, and graduation date instead, and faked the matching algorithm with saved searches hand-built per major. The prototype went live on a slice of traffic within days. Search starts went to 83 percent.

That progression is the book in miniature. The 2021 book defines the practice in four lines. Weekly touchpoints with customers, by the team building the product, through small research activities, in pursuit of a desired outcome. The reasoning is arithmetic. A team makes product decisions every day, so a team that interviews monthly makes a month of decisions without customer input.

Most product organizations treat discovery as a phase: research up front, in big studies, owned by leaders or an agency, followed by delivery. Torres argues discovery is a weekly habit run by the people who ship, and that small weekly research beats large occasional research.

The unit she prescribes is a product trio: a product manager, a designer, and a software engineer deciding together. She names the trade openly: every added person slows every decision, so balance inclusion against speed.

The Framework

One artifact carries the method, the opportunity solution tree. The desired outcome sits at the root. Below it sits the opportunity space, her collective term for customer needs, pain points, and desires. Below the opportunities sit the solutions under exploration, and below the solutions sit the assumption tests. She says opportunity rather than problem because desires count too: ice cream needs no fixing.

The habits chain into one loop.

• Negotiate one product outcome with your product leader and hold it for several quarters.
• Interview at least one customer every week and ask for specific stories.
• Map everything you hear as a tree of opportunities under the outcome.
• Choose one target opportunity by comparing sibling opportunities against each other.
• Generate 15 to 20 ideas for it and vote down to a set of three.
• Surface each idea's assumptions, map them by risk, and test the riskiest two or three.

Then measure impact, revise the tree, and keep interviewing. Marty Cagan's benchmark, quoted in the book, is that teams competent in modern discovery test on the order of 10 to 20 iterations per week. The tree is what keeps that pace legible.

Key Ideas

Trios Get Product Outcomes

Torres splits metrics into three levels. Business outcomes measure the business, lag, and need several functions to move: revenue, costs, market share. Product outcomes measure how the product moves the business and sit within the trio's control. Traction metrics measure usage of one feature, and a traction metric assumes the feature is already the right answer. Her rule assigns product outcomes to trios.

The dog-food subscription tails.com shows the mechanics. Retention in the first 90 days predicted long-term retention, and a 90-day feedback loop is too slow. The team cut the metric to 30 days, then to 5, which allowed weekly experiments. Interviews then surfaced the two real churn drivers: owners who never understood the value of tailor-made food, and dogs that disliked it. Both became measurable product outcomes.

Outcomes are a two-way negotiation. The leader brings business context and strategic intent. The trio brings customer knowledge and states how far it can move the number in a quarter, for example 10 percent in three months. Teams that set their own outcomes take more initiative, per research she cites. On complex work she starts teams on a learning goal, then graduates them to a performance number once strategies exist.

The cautionary case is Wells Fargo. Leaders pushed a defensible outcome, more accounts per customer, through impossible quotas and rich incentives, and bankers opened accounts customers never approved. The Consumer Financial Protection Bureau fined the bank 185 million dollars, and lawsuits cost billions more. Torres reads it as an outcome divorced from customer needs, and her repair is wording: grow customers who want to open more accounts.

Measurement got the same discipline at AfterCollege. The real outcome was students getting jobs, and hires happened off the platform. So the team emailed applicants 21 days after each application and asked what happened. Response rates grew from 5 percent to 14 to 37 by the time she left. Applications were easy to game. Measure the hard thing anyway.

Stories Beat Direct Questions

The keystone habit is the weekly interview, and the keystone skill is the question shape. Torres asks workshop participants what factors decide a jeans purchase, and fit comes back as the number one answer. Asked about her last actual purchase, one woman had bought on Amazon, sight unseen: a brand she liked, on sale. Torres runs the exercise in every workshop, and the gap between the two answers always appears.

People invent reasons without noticing. In Michael Gazzaniga's split-brain studies, patients picked a card using information their speaking hemisphere never received, and every subject fabricated a confident explanation. Torres carries a matching scar. Every recruiter she interviewed in 2007 wanted to source passive candidates, so her team built the tool. It flopped. Recruiters are measured on time to fill, and active candidates fill roles fastest.

The repair is a story request. Replace the criteria question with a prompt: tell me about the last time you bought jeans. Then excavate along the timeline: what happened first, what happened next, who was there. When the participant drifts into what they usually do, pull them back to the specific instance. Her golden rule hands the participant the wheel: they talk about what matters most to them. You steer twice only, by choosing which story to request and which parts to dig into.

The habit survives on automation. Her bar for recruiting is waking up Monday with an interview already booked, with zero effort spent that week. The easiest channel is a one-question ask inside the product, 20 minutes of feedback in exchange for 20 dollars, wired into scheduling software. An interview can run five minutes. Each interview ends as a one-page snapshot: a quote, quick facts, the story drawn as nodes, and opportunities in the customer's words. A feature request converts with one follow-up, asking what the feature does for them. A wish to speak a movie title aloud becomes a need to avoid typing long titles, which opens voice search and auto-complete both.

The payoff is switching speed. One team killed a target opportunity on Tuesday, picked a new one on Wednesday, and used interviews already booked for Thursday.

The Tree Turns Whether Into Which

Torres names the most common trio mistake: the whether-or-not decision. Stop everything and fix this problem, or build this executive's pet feature, framed as yes or no. A single option invites confirmation bias and hides opportunity cost. Every decision on the tree runs as compare and contrast instead: which of these needs matters most, and how else can we address it.

Selection works top down. Compare the top-level opportunities against each other, pick one branch, and ignore every other branch for the rest of the assessment. Repeat on the winner's children until you reach a leaf, because value ships as a series of small solved opportunities. Four lenses drive the debate: opportunity sizing, market factors, company factors, and customer factors, importance crossed with satisfaction. She refuses scoring formulas. The judgments are relative, and a formula treats an ill-structured problem as a well-structured one and stops the thinking.

The pick stays cheap because it is reversible. Choosing a target opportunity commits days or weeks of exploration, a two-way door in the Bezos sense. Torres cites Bullens and colleagues: people who framed a choice as reversible kept evaluating it critically afterward. Framed as irreversible, the same choice showed people only its positives. So time-box the selection to an hour or two, at most a day or two.

Generate Alone, Decide Together

Group brainstorming came from Alex Osborn's 1953 book Applied Imagination. For decades researchers compared brainstorming groups against the same number of people working alone, and the groups lost on volume, diversity, and originality. The causes are social loafing, conformity, production blocking, and downward norm setting, where the group sinks toward its weakest member. Groups still feel productive because members get stuck less often, which explains the advocates.

Her protocol alternates. Generate alone, share across the team, then return to solo generation, because exposure to others' ideas improves the next solo round. Continue until the target opportunity holds 15 to 20 ideas. Weed out the ones that miss the opportunity, then dot-vote with three votes per person. The stop point is a set of three, never one, so that testing stays a comparison. Quantity is the lever: volume correlates with diversity and originality, and the most original ideas arrive late in a session.

Assumptions Test Faster Than Ideas

Portland shows the cost of skipping this habit. The city spent tens of millions of taxpayer dollars on condominiums for displaced families, and the developer built mostly one- and two-bedroom units. Most displaced families had four or more members. Years past the target date, most units sat unsold, and the city let the developer sell on the open market. Nobody tested the sizing assumption.

Building three prototypes and A/B testing them takes longer than any quarter allows. Assumptions test fast and in parallel, and the speed itself fights escalation of commitment: less time invested means less love. Every idea gets a story map, drawn as if the solution already exists. Each step generates assumptions across five categories: desirability, viability, feasibility, usability, and ethical. A five-step map in her worked example produced 20 assumptions, and she gets 20 to 30 from a simple idea. Phrase each one so you need it true. Most trios have a blind spot for the ethical category, because their intentions are good.

Prioritization is a two-axis map, evidence against importance, placed relative rather than precise. Ten minutes per idea is enough. Test only the two or three assumptions in the top right corner, the leap-of-faith ones.

A strong test simulates one moment and evaluates behavior, with success criteria fixed before any data arrives. Criteria are counts, never percentages and never the word some: at least 3 out of 10 people choose the sporting event. In her streaming example, the first mockup test ran in a day or two, and 4 of 10 chose sports. The escalation was a promotion on the real home screen: a week of development, 500 participants in three days, success at 100 thumbs up. Small tests come first because most learning comes from failed tests, and small tests fail sooner. Stop when the remaining risk is bearable or the next test costs more than building.

Two disciplines round it out. Design tests likely to pass, with the friendliest audience, because a failure in the best case is unambiguous. And keep the claim modest: the goal is risk reduction rather than truth, and findings are evidence, never verdicts.

Practical Applications

Automate recruiting first. Add a one-question ask to the product, 20 minutes for 20 dollars, wired into scheduling software. Give colleagues on sales and support a trigger list and a script. The bar is a Monday morning with an interview already on the calendar, because a booked interview is easier to hold than to skip.

Change the question shape this week. Ask for the last specific time, walk the timeline, and block generalities. Convert every feature request by asking what the feature does for the customer, and file the need in their words. Close each interview by drawing the story and writing the snapshot.

Build the tree before touching the backlog. Draw the experience map as individuals, merge the drawings, and scope the map from the outcome. Add an opportunity only when it is a need, appears in more than one interview, and drives the outcome if addressed. Pick the target by comparing siblings, and time-box the pick to a day.

Before building anything, story map it, list assumptions across the five categories, and map them by evidence and importance in ten minutes. Write test criteria as counts before data arrives: how many people you test, how many must act. Run the smallest simulation that moves the riskiest assumption.

Handed an output instead of an outcome, work backward. Ask what the solution does for customers and what value shipping it creates, and those two answers form your first tree. Agree the expected impact in writing, instrument it, and run a post-release impact review. The strongest week to argue for discovery is the week a feature falls short.

Who This Is For

Product managers, designers, and engineers who ship weekly and decide by argument get the most, and the book works best read as a trio. Founders before product-market fit get the cheapest version, since the founding team already is the trio and the weekly interview costs one calendar slot. Product leaders get a manual for negotiating outcomes instead of assigning roadmaps.

Skip it if your team already interviews weekly and tests assumptions, because the book is her 12-week coaching program written down. Researchers wanting statistical rigor get method sketches, on purpose.

The evidence deserves its label. The case studies are her own coaching clients, told by people who bought the program, with no failure rate reported. The running streaming example is hypothetical, and the AfterCollege numbers are her recollection of her own product. The behavioral science is real and cited. The product claims riding on it are anecdote, so treat the mechanisms as durable and the rates as hypotheses.

The Decision

The entry habit costs one week and needs no permission. Book one 20-minute customer conversation before Friday. If no customer is reachable, book anyone who resembles one. Open by asking for one specific story, walk its timeline, and let the participant steer. Before the call ends, schedule the next one. Her bar for progress carries the whole system: make next week look better than last week.