We have never been handed a first-release scope that was too small. Not once. The list always has three products folded into it, and every item on it is entirely reasonable on its own — which is exactly the problem. Nobody adds a feature to an MVP because they think it's frivolous. They add it because it's obviously necessary, and thirty obviously necessary things is a two-year build.
There are three lists inside your list
Read a typical MVP scope closely and it separates into three piles. The first is the thing you're actually testing: a specific bet about a specific person doing a specific job in a way they can't today. The second is the machinery that bet needs to function at all — accounts, one integration, a way to get data in and out. The third is everything that will obviously be needed eventually and has been pulled forward because listing it now is easier than arguing about it later: the admin console, the reporting, the second user type, the settings screen with eleven toggles.
That third pile is routinely more than half the estimate, and none of it teaches you anything. It's the cost of a product you haven't validated yet, paid in advance.
Cut depth, not breadth
The instinctive cut is to remove features, and it often breaks the thing outright. A product with a hole where a step used to be doesn't test anything — a user who can sign up and configure but never complete the job has only told you your onboarding works. The cut that survives goes across depth instead: keep the entire path a person walks, and make every step on it dumber.
In practice that means one user role instead of a permission matrix. A hard-coded plan instead of a billing engine. A weekly CSV instead of a reporting dashboard. A human reviewing submissions by hand instead of the automated approval workflow — the workflow you'll build later, once you've watched someone do it forty times and finally know what the rules are, which right now you don't. Each of those substitutions preserves your ability to run the experiment and removes weeks from the build.
There is a version of this that goes too far, and it's worth naming, because a release nobody would use voluntarily produces no signal either. If the first version asks someone to tolerate something genuinely painful in exchange for very little, what you learn is how patient they are, not whether your product is worth building. The line to hold is that every step can be crude, but no step can be worse than whatever the person is doing today — the spreadsheet, the email thread, the phone call. Beat the status quo on the one job you picked, and be visibly unfinished everywhere else.
The things people cut that they shouldn't
Not everything defers safely, and the same three keep getting cut and cost the most to reinstate. The first is structure in the data model: you can skip building the second user type, but if the schema can't represent one, adding it later is surgery rather than a feature. The second is instrumentation — shipping a product whose entire purpose is to answer a question, with no ability to observe the answer, is the most common self-inflicted wound in this kind of work. The third is anything that has to be true from the first record onward: an audit trail, a tenancy boundary, an encryption decision. Retrofitting those means rewriting history you no longer have.
The rule we use is short. Defer behaviour freely, defer structure carefully, and never defer the ability to see what happened.
Scope is a schedule wearing a costume
The most useful reframe we can offer is that a scope list isn't a description of a product. It's a claim about a date. Every line added to the first release is time added between now and the moment you learn whether any of this was worth doing, and that is the real currency being spent. It's why “we may as well include it, it's small” is such an expensive sentence: the individual item usually is small, and the thirtieth small item still pushes the answer another month out.
So the question to ask about each line isn't whether it belongs in the product — almost everything belongs in the product eventually. It's whether this release can teach you what you need to know without it. If it can, that item isn't being cut. It's being scheduled.









