How to Run Product Discovery Fast, Cheaply and Honestly
How to run product discovery: start from the riskiest assumption, pick the cheapest test that could change your mind, timebox it and record the decision.
To run product discovery, list what must be true for an idea to work, test the most dangerous assumption first with the cheapest test that could change your mind, set a time limit, and write down what you decided. Done that way it is faster than the alternative, because the alternative is building.
Discovery has a reputation for being slow and academic. That is a description of discovery done badly. This post is the overview of the whole loop, and it points to the sibling posts that go deeper on each part.
How to run product discovery as a weekly habit
Think of discovery as one small loop, repeated. Each pass has six moves.
- Write the idea as a set of assumptions, not as a feature.
- Rank the assumptions by how damaging it would be to be wrong.
- Choose the cheapest test that could genuinely change your mind.
- Set a time limit and a date to decide.
- Run the test, talking to people or looking at evidence.
- Decide, and record what happened.
None of these moves needs a research department. They need discipline, a short list and a willingness to be wrong. The loop also assumes a few decisions were made earlier, such as who decides when something is ready, which are covered in what validated means.
Move one: turn the idea into assumptions
Discovery items should be phrased as questions or risks, not as solutions waiting for approval. "Will users complete setup without help" is a discovery item. "Build the new onboarding" is a delivery item that has skipped the question.
The distinction matters because a discovery track full of pre-decided features is just a delivery backlog with a longer approval process. It will produce exactly the confirmations you went looking for. Start from the risk, not from the thing you already want to build.
For any idea, the list of things that must be true usually includes these:
- People actually have this problem.
- They will change their behaviour.
- They will understand what you made.
- They will pay.
- It is technically feasible.
Move two: rank by damage, not by interest
Teams reliably test the assumption that is easiest or most enjoyable to look at. That is usually usability, which is comfortable and measurable. Meanwhile the assumption that would actually kill the idea, typically whether anyone cares, goes untested until launch.
Rank by which assumption would hurt most if wrong, and test that one first. The detail is in how to test your riskiest assumption.
Move three: pick a test you would obey
For each assumption, ask what the least expensive thing is that would move your belief. The guide lists several options.
| Kind of test | What it looks like |
|---|---|
| A conversation | Talking to people who match the target |
| A landing page | Describing the idea and seeing whether anyone responds |
| A prototype | Something people can react to before it is real |
| A manual version of the service | Doing by hand what the product would later do |
| A concierge run | Serving a handful of customers personally |
| Data you already have | Looking at what your own records can already tell you |
Then apply the real test: if this comes back negative, will we actually stop? If the answer is no, the test is theatre. Skip it and go straight to building, which is at least honest. A test you would ignore is worse than no test, because it uses time and manufactures false confidence.
Move four: set the limit before you start
Discovery expands to fill available time like any other work, and it has a particular hazard: it always feels responsible to learn a little more. So decide the limit first. A few days or a week is typical.
When the time is up, make a decision with what you have. If the honest result is that you still do not know, write down the single thing you would test next. Open-ended research is how a discovery track turns into an internal department that ships nothing.
Move five: talk to people without steering them
Many tests are conversations, and the most common way to waste one is to describe your solution and ask whether people like it. That reliably produces politeness. Ask about behaviour and history instead. The approach is covered in full in how to interview users without leading them.
Two habits are worth stating here. Treat enthusiasm as weak evidence, since people are generous in conversation and much less generous with their time and money. And speak to people who match the target, not whoever happens to be available.
Move six: decide and record
Every question ends in a decision. Sometimes the decision is to build. Sometimes it is to drop the idea, which is frequently the most valuable outcome because it returns a quarter of capacity to better uses. Sometimes it is to run one more specific test.
Whatever the decision, write it down in four lines: what we asked, what we did, what we found, what we decided. How to make that record findable is the subject of how to keep a discovery record.
What good and bad discovery look like
| Question | Discovery that works | Discovery that does not |
|---|---|---|
| What is on the list? | Assumptions and risks | Features waiting for approval |
| What gets tested first? | The most dangerous assumption | The most convenient one |
| What counts as a good test? | One that could change your mind | One you would ignore if it came back negative |
| How long does it run? | To a date, then a decision | Until everyone feels certain |
| What happens to killed ideas? | Recorded with what was learned | Forgotten, then proposed again |
How discovery fits around everything else
Discovery on its own is only half of the practice. It runs beside delivery, by the same team, and two things keep it honest.
The first is direction. Discovery attached to no objective becomes a wandering curiosity project, interesting and unfocused. The clean relationship is that objectives say what should be different, discovery finds routes that might achieve it, and delivery builds the routes that survive. An outcome-based objective is what permits discovery, because it commits to a result rather than a predetermined feature. Finding a cheaper route to the same outcome is a success, not a deviation from the plan.
The second is protected time. If the team only does discovery when there is time, it will do none, because the time never appears. How to defend that time is covered in how to split capacity between discovery and delivery.
Who runs the loop
The same team that builds also does the discovery. This is the structural decision the guide asks you to take above all others, and to repeat often, because the most common misreading is to create a discovery team and a delivery team. When that happens, findings arrive as a written specification, the people building had no part in forming the question, and everything that made the insight actionable is lost between the document and the build.
The builders do not need to attend every conversation. They need to be present enough that the reasoning is shared and not transmitted. A designer who heard the hesitation in a customer's answer will make better small decisions later than one who only read the summary.
Signs the loop has stalled
Discovery rarely fails loudly. It fades. Watch for these signals.
- Nothing has been killed. A track that never decides against building is not screening anything, so the queue downstream goes unexamined.
- Tests keep confirming the plan. A run of agreeable results usually means the items were features in disguise, or the tests were ones nobody would have obeyed.
- Questions have no date. Work with no timebox drifts, and drifting discovery becomes an internal research function.
- The record is empty. If nobody can say what was decided last month, the track may be running only in conversation.
- It happens only when there is time. Unprotected capacity goes to whatever is loudest.
A direct monthly question exposes most of these: what did discovery decide this month, and what did delivery ship? Two consecutive months with a blank on either side is the signal to act.
When discovery is ready to hand over
A question is finished when someone with the authority to say so decides that enough has been learned. The guide asks teams to name that person or small group, and to give them permission to say both "not yet" and "we know enough, stop researching". Calling time is as necessary as asking for more, and much rarer. When the answer is yes, the work moves across with its reasoning attached, as described in how to hand off discovery to delivery.
A weekly checklist
- Every idea in play is written as assumptions, not as a feature.
- The assumptions are ranked and we are testing the most dangerous one.
- The test is the cheapest one that could change our mind.
- We have agreed in advance that a negative result will stop us.
- There is a date by which we decide.
- We know what result would count as a no.
- Each question has its four lines written down.
If a box cannot be ticked, that is the next task.
Frequently asked questions
How much time should a discovery question take?
Set the limit before you begin. The guide suggests a few days or a week. Some questions resolve in an afternoon and others take a month, which is why discovery needs its own rhythm, but every one needs a date on which you decide.
What if the test does not give a clear answer?
Decide anyway with what you have. If the honest answer is that you still do not know, write down the one thing you would test next. That is a decision too, and it is better than open-ended research.
Does every idea need discovery?
Different decisions warrant different rigour. A small, reversible change needs far less than a quarter of engineering. The point is to match the effort of the test to the cost of being wrong.
Can we skip discovery and just build?
Yes, if you would ignore the result of any test, and the guide is blunt that skipping is more honest than running a test for show. But building is the most expensive way to test an assumption, so be sure you really are certain.
The full method, with the checklists, is in The Dual-Track Playbook, a free 34-page playbook.