When do you know a feature is actually worth building?

I think many tech product builders, including myself, can easily fall into tunnel vision, whereby I am overly convinced that the feature we’re working on is clearly justified. With vibe-coding tools, the cost of building new features has also dropped dramatically. But lower development costs didn't seem to translate to product growth in my organization. Many newly launched features still receive little adoption, or have a hard time "syncing" with the rest of the product.

So I’m curious: how do you decide when a feature is actually worth building? Do you wait for repeated user requests, test a lightweight version first, trust your product instinct, or rely on behavioral data before committing?

Also curious: which feature seemed unnecessary at first but later became essential? and which “obvious” feature ended up barely being used?

28 views

I have been asking myself this exact question a lot lately.

Building a feature is the easy part now. With the AI coding tools the cost of shipping one dropped to almost nothing. The hard part moved somewhere else: getting anyone to see it and actually use it. You ship the thing and it just sits there.

And what I keep seeing is that people use one or two features 99% of the time. The rest is stuff we were sure they needed. On our side (support tooling) almost every ticket is the same small handful of questions, over and over, and the "smart" features nobody asked for get almost no traffic.

So the signal I trust most now is not requests and not instinct, it is: does this feature touch the one or two things people already do every day? If yes, build it. If it needs the user to pick up a new habit, I am very careful, because a feature that asks people to change their behaviour usually loses.

I do not have a clean before-and-after example for you, I am still watching for it. What I can say is the misses were almost always the features that assumed a new behaviour, not the ones that made an existing one faster.

Still figuring this out, so asking you back: when a feature gets low adoption, how do you tell if it is a bad feature or just badly shown? Those two need the opposite fix and I get them wrong a lot.

When do you know a feature is actually worth building?

Replies