How Many Votes Each Person Gets in a Retro

The number people reach for is three, and three is usually fine. It stops being fine in two places: a team of four with nine cards, and a team of thirty with sixty cards. Both of those produce a vote tally you can't act on, and you find out at the worst moment, when you're standing in front of the board trying to decide what to talk about with twenty minutes left.

What the tally should look like

The only job of voting is to produce separation. You want the top item to be clearly ahead, and a visible drop somewhere around position three. If your top five items are on six, six, five, five and five votes, voting didn't tell you anything and you're about to pick the topic yourself while pretending the board picked it.

The rough rule I use: one vote per four cards, minimum two, maximum five. Twelve cards after grouping, three votes. Twenty-four cards, five votes. Eight cards, two votes. More votes than that and everyone votes for everything they mildly agree with, which flattens the distribution and tells you only that the team is agreeable.

Count the cards after grouping, not before

This is where most of the damage happens. If four people wrote a card about the staging environment and you vote before merging them, the staging problem splits across four cards and loses to one card that one person felt strongly about. Group first, then set the vote count based on what's left. If grouping cuts forty cards to eighteen, your vote allocation changes from five to four, and that's a real difference in the shape of the result.

Stacking, and voting for your own card

Allow stacking, capped at two on any single item. Somebody who has been chewing on one problem for three sprints should be able to say so with the full weight of what they're holding. Letting them put all three votes on one card lets a single person set the agenda, which is exactly what voting is meant to prevent.

Let people vote for their own cards. The bias is small, and policing it costs more in awkwardness than it saves in accuracy.

Hide the counts until everyone has finished

Live vote counts produce bandwagoning, reliably and fast. The first two people to vote are usually the two most senior people in the room, and the tally they create becomes the tally everyone else ratifies. I've watched a staff engineer put two votes on a card about CI flakiness and seen seven of the next nine votes land on the same card. Hide the counts, reveal them when the last person is done.

Small teams

Below about six people, voting is theatre. Twelve votes across nine cards gives you three-way ties and no information. Go round the table and have each person name their one must-discuss item, write those down, and work through them in the order people said them. It takes ninety seconds and the result is better.

Ties, and the facilitator's override

For a genuine tie at the cut line, run one vote each between only the tied items. Ten seconds, done.

And when something with two votes is clearly the thing nobody wants to name, put it on the list anyway. Say out loud that you're doing it and why. A team will accept a facilitator making a judgement call. What they won't accept is a facilitator pretending the board made it.