Skip to content
Swingback, home

BlogFeature requests

Does feature voting work? What votes do and don't tell you

Feature voting measures reach, not need. What votes tell you, what they hide, and how to weigh them by who is voting before you decide what to build.

By Elie Baier · · 4 min read

Feature voting works for one thing: it tells you how many people share a problem. It doesn't tell you how much the problem costs them, who they are, or what to build, and it never hears the customers who don't vote. Votes are a good measure of reach and a poor ranking on their own. Read them together with who is voting, what they wrote and where your product is going, and they earn their place on your board.

What votes tell you#

  • Reach. How many people bothered to say "me too". All else equal, a request with eighty votes touches more of your users than one with three.
  • Their words. The request and its comments say how people describe the problem, which is how they'll look for the answer in your app and in your docs.
  • Momentum. Votes that arrive this month mean more than votes from two years ago. A request that suddenly gathers votes often follows a change: a new kind of customer, a competitor's launch, something you broke.
  • Who to ask, and who to tell. Every voter is someone to talk to before you build it, and someone to tell when it ships.

What votes don't tell you#

  • How much it matters. A vote costs one click. "Nice to have" and "we'll leave without it" weigh the same.
  • Who is asking. A visitor from last week and your largest customer each count as one.
  • The problem behind the solution. People vote for the fix someone proposed. Behind "add a CSV export" there may be a weekly report, a migration and a spreadsheet kept by hand: three problems, one of which an export solves.
  • What nobody asked for. Most of your users never vote. Neither do the customers who left, nor the people you haven't reached yet. Your strategy isn't on the board either.
  • A fair race. The requests at the top are the ones people see first, so they collect more votes for being at the top. Old requests pile up votes simply by being old.
  • One count per need. Duplicates split the votes for one need across several requests, each of which looks smaller than the need is. See duplicate feature requests.

None of this makes voting useless. It makes a vote count a place to start reading, not a ranking to build in order.

Make votes worth reading#

Ask for the problem#

A request that says what the person is trying to do, how they do it today and how often it gets in their way is worth far more than its votes. People who vote for it vote for the problem, which is what you can compare. The feature request template asks those questions, on a board as in a GitHub issue form.

Merge duplicates#

Votes only add up on one request. Catch copies while people type, and merge what slips through, so that the count reflects the need.

Weigh votes by who is voting#

Give each company a value, such as its monthly revenue, and add up the values of the companies whose people voted for a request. A request with twelve votes from your three largest customers may matter more than one with sixty votes from free accounts, and a list sorted by that sum shows it.

A request with few votes and valuable companies behind them: its customer value is the sum of what those companies are worth to you.

Count each company once per request. Five people from the same customer are one customer's need, voiced five times.

Look at what's rising#

Sort by recent activity as well as by total votes. The total favors what has been there longest; recent votes and comments show what people need now.

Read the comments, then talk to voters#

Votes tell you where to look. The comments, and a short call with two or three voters, tell you what to build. Often the answer is smaller than the request.

Say how you decide#

Tell your customers that votes are an input, not an election: you weigh them with what customers are worth, how much it costs to build, and where the product is going. Then a request with many votes that you decline doesn't read as a broken promise. When the answer is no, say it with a reason: how to say no to a feature request has replies to copy.

Votes in Swingback#

Voting in Swingback is built to be read this way:

  • One vote per person. Voting again takes the vote back. As someone types a new request, similar requests show up, so they can vote for one of those instead.
  • Customer value on every plan. A company's value comes from your server, in the identity token that signs your users in to the widget, from the REST API on the Pro plan, or from the dashboard. A request's customer value adds up the companies of its voters, each company once. Under Posts, sort by Customer value to see what your most valuable customers want first, or filter by a company to see what its people voted for. The portal never shows it.
  • Trending by default. The portal sorts by recent activity unless people choose Top, so a new request with momentum isn't buried under old ones.
  • Voters hear what happens. Voting follows the request, so every status change, and the changelog entry that ships it, reaches the people who voted. See closing the feedback loop.

Rank posts by customer value in the docs has the details, and Signing in your users shows the token's companies claim.

Guests, where your portal takes them, vote too. Their votes count toward reach and nothing toward customer value, since nobody knows which company they belong to: one more reason to sign your users in.