Once people use your tool they'll tell you things. Building everything they ask for produces an incoherent product; ignoring them produces one nobody wants. The skill is in the translation.

Separate the problem from the solution

Users are excellent at describing what's wrong and unreliable at prescribing what to build. "I need a bulk export button" might really mean "I check this every morning and it takes ten minutes".

Always ask what they were trying to do:

dig-into-feedback.txt
A user asked for: <the request>

Give me five questions that would uncover what they're actually
trying to achieve, without leading them.

Then: what are three different solutions to the underlying need,
including one that isn't the feature they asked for? What would each
cost me to build?

The best outcome is often something simpler than what was requested.

Weight by behaviour, not volume

One person who uses your tool daily and hits a wall matters more than five who tried it once and had opinions. Ask "how often do you use this?" before deciding how much weight to give a request.

Also notice who's asking. A request from someone in your target audience is different from one from someone who was never going to be a user.

Count, don't react

One request is an anecdote. Five is a pattern. Keep a list, add a tally each time something comes up, and act on the ones that recur. This stops the most recent conversation dominating your roadmap.

Say no well

You'll decline most requests. Do it clearly and without hedging: what you're not building, and briefly why. "This doesn't fit where I'm taking it" is a complete and respectful answer.

Vague deferral — "great idea, I'll look at it" — is worse than a no. It leaves people expecting something that isn't coming.

The feedback that matters most

Watch someone use it without helping them. Every hesitation is a design problem you can't get from a survey, and it's the one form of feedback nobody has to articulate for you.

"I stopped using it because…" is the most valuable sentence you'll receive. Ask lapsed users directly — most of them will tell you.