Product sense and execution

A user asks for a filter. What do you ask next?

What decision the filter would change, and how often. The request names a design. It may not name the problem.

Interviews ask for customization, and almost nobody uses it. What do you trust?

The behavior, then go find out why the interviews said otherwise. Shipping the requested control ignores the log.

An impact estimate rests on one customer's claim. What is missing?

Evidence from more than that account, and a measure you can check after launch. One promise is an anecdote.

What is wrong with counting anyone who opens the app as active?

It rises when you send more notifications, and it says nothing about the value. Define active as the action that is the product.

A product is used once a year. Which retention fits?

Whether they come back for the next season, not a 30-day return. The window has to match the job.

The biggest unknown is whether neighbors will order together. What do you build?

The smallest test that answers that, not the full marketplace. A full build before the answer spends the year on a guess.

What belongs in a spec, and what does not?

The user, the outcome, the constraints, and the out-of-scope list. The implementation plan belongs to the people who will write it.

Harvest starts in two weeks and the tool is not ready. What do you cut?

Everything except the job they must do this season. The rest can ship after, when missing it does not cost the year.

Signups fell in one country and the product did not change. What do you check?

A local cause: a law, a competitor, a store listing, a payment method. A product change will not fix a drop that started outside.

A vendor ships a new model behind your feature. What do you do first?

Evaluate it on your task and your data before you turn it on. Their benchmark is not your user's outcome.

When is a success metric chosen?

Before the build, with the threshold that would count as a win. A metric picked after launch can always be made to look fine.

What is a guardrail?

A measure that must not get worse while you chase the goal, such as support contacts or an error rate.

What is a counter-metric for?

To show the cost of the win. A faster signup that doubles fraud is not a win until you look at both.

Why name the audience?

An average can hide a feature that helps one segment and harms the one you depend on. The spec says who it is for.

What is the job to be done?

The progress the user is hiring the product for, in their words. A feature list is not a job.

What do users do when you do not exist?

The workaround is the real competitor. A feature that is worse than the spreadsheet will not be used.

How do you cut scope without cutting the point?

Remove a slice the user can still finish the job without. A cut that leaves the job impossible is a delay, not a cut.

What is a minimum test of a product idea?

The cheapest thing that will change one named decision. You write that decision down first, and you stop building once the result is in.

What is a launch criterion?

The measure, the audience, and the operational bar, such as support being ready. A date alone is a calendar entry.

What is a product rollback?

You can turn the change off, and you know what 'worse' looks like before you turn it on.

When do you experiment instead of just shipping?

When the cost of being wrong is high and you can get a real comparison. A tiny copy change on a quiet page may not be worth a test.

What is a user interview for, once the logs exist?

To learn why a behavior already in the logs happened. How many people do that behavior is a count from the logs, or from a test sized to measure it.

What can the logs not tell you?

Why. A drop says something changed. A conversation, a ticket, or a session replay is how you find the cause to test.

What is a north-star metric?

The one number that stands for the value you create, which teams can move. Revenue alone can rise while the product gets worse.

What is an input metric?

A behavior the team can affect this quarter that you believe moves the north star. It is what the roadmap is aimed at.

What is a vanity metric?

A number that tends to go up and does not change a decision, such as a raw page-view count with no action attached.

Why look at a cohort?

Users who started in the same period, so a later group's behavior is not mixed with people who have had a year to learn the product.

What do you check in a funnel?

Where people drop, and whether that step is the one you meant to improve. Optimizing a later step does nothing if nobody reaches it.

How do you order two good ideas?

By the outcome they move, the evidence, and the cost. 'The customer shouted' is one input, not the order.

What do you owe a stakeholder who disagrees?

The goal you are optimizing and the evidence. A roadmap that is a list of favors has no goal.

How do you say no?

Name the outcome already committed and the work you would stop to say yes. The answer is that dropped work, or a quarter when the new request starts.

What is a product risk you write down?

The assumption that, if wrong, makes the project pointless, and how you will find out cheaply.

What is the riskiest assumption?

The one the plan needs and you have the least evidence for. You test that before you build the parts you are sure about.

Where does a growth idea get a review?

Before the experiment, if it hides a cost, traps a user, or uses data they would not expect. A metric lift is not the only bar.

What do you do with a spike in tickets after launch?

Treat it as a guardrail firing. Read the tickets, decide whether to roll back, and do not wait for the weekly metric.

A change is 'just copy.' What can it still break?

The meaning of a commitment, a legal line, or a step in the flow. Someone who knows the constraint reviews it.

When do you invest in a platform instead of a feature?

When several features you already intend will need the same capability, and building it once is cheaper than building it inside each.

A date is fixed by a season. What is flexible?

The scope. Pretending both are fixed produces a half-tested launch on the morning of the season.

What is a beta for?

A named audience, a question, and an end. A beta with no question is a slow launch to a confused group.

When do you remove a feature?

When the evidence says it is unused or harmful, and you have a path for the few who depend on it. Leaving it costs a test and a support case forever.

How do you use a competitor's feature?

As a hypothesis about the job, which you check with your users. Copying the screen copies their context, which you may not have.

What do you test about a price?

Whether the people you want will pay it, and what they stop doing. A price picked from a competitor is a guess with a logo.

What is a roadmap?

The bets in order and the outcome each bet is for. A request joins that list when it serves one of those outcomes, with the evidence named.

Why separate now, next, and later?

Now is committed. Next is shaped. Later is a direction. Treating later as a promise makes every new fact a broken promise.

What do you record when you kill an idea?

Why, and what would reopen it. Otherwise the idea returns every quarter as if it were new.

What is a launch review checking?

The measure, the rollback switch, support staffing, and who will see the change. It passes when those four are true, before anyone walks through the screen.

What do you schedule with the launch?

The day you will read the metric and decide to keep, change, or remove. A launch with no read is a guess you never grade.

What is alignment for?

Ask each person, separately, for the user, the metric, and the non-goal. Alignment is those answers matching. A shared slide can hide the mismatch.

What is an outcome versus an output?

An outcome is the change in user behavior you wanted. An output is the feature you shipped. The roadmap is done when the outcome moved.