Dot voting

Goal

Dot voting turns the individual preferences of a group into a numbered ranking of the options. It sits at the end of a divergent phase, when a workshop has produced more ideas than it can handle. Everyone has the same number of votes and places them at the same time as everyone else.

The deliverable is a list of options, each carrying a vote count, copied into the product backlog or into the workshop minutes. That figure records how many of the people in the room that day placed one of their votes on the item. It does not measure the business value of the option. Each participant placed their own dots, which makes the result hard to dispute.

Dot voting belongs to the prioritisation family, whose members each leave the decision to a different authority. Grouping files items into predefined bins. Ranking imposes a total order from first to last. Time boxing and budgeting let the calendar or the envelope decide. Negotiation has the stakeholders argue until they agree. Dot voting and bullseye are the two instruments of collective judgement, and neither of them goes through a debate. Bullseye distributes items across concentric zones of limited capacity, which forces the trade-offs through the space available, whereas dot voting adds up votes.

Usage

When to use it

• Coming out of a divergent phase: bringing twenty ideas down to three.
• A large group and a short slot: getting a signal from everyone without making everyone speak.
• A reversible or low-stakes decision: the cost of the exercise stays proportionate to what is being settled.
• An agenda to set: letting the group choose which subjects it takes first.

When not to use it

• A decision that commits a budget or a regulatory responsibility: it belongs to whoever answers for it, prepared by negotiation.
• A total order expected across the whole list: the count leaves ties and items with no votes, so take ranking.
• A resource ceiling that settles the matter on its own: start from the envelope with time boxing and budgeting.

Description

The common mechanism

All the variants follow the same four steps. A list of options already exists, as sticky notes on the wall, as pinned-up sketches or as lines on a flip chart: dot voting chooses among ideas that already exist. The count can only be read if those options are of the same kind and the same grain. The facilitator announces the allowance, that is the number of votes per person and what may be done with them. Participants place their dots at the same time, without commenting. The facilitator counts the dots in front of the group and reads out the ranking.

Simultaneity is what drives the technique. Speaking in turn gives disproportionate weight to whoever goes first, then to whoever pushes hardest; placing a dot at the same moment as everyone else removes that dependence on the order of speaking.

Dot voting

The wall after the vote

One option ahead, a mid-tier cluster, blank cards

Three sets of rules under one name

Gamestorming, by Dave Gray, Sunni Brown and James Macanufo, gives the simplest and most widespread form: around five votes per person and the explicit right to stack several votes on one item. The number five is an order of magnitude there, which the facilitator adjusts. The dots go up in one pass, in the open, on the wall. The book provides no tie-break round.

Multivoting, from the American Society for Quality, comes out of the nominal group technique: silent individual idea generation, a round of the room with no discussion, then a vote. The voter distributes distinct weights across five items, five for their favourite down to one for the fifth, which rules out concentrating the whole allowance on a single option. A second form has each person keep the top third of the surviving list, round after round, until the list is short enough to discuss. Ballots there are most often written, so more discreet than a wall of dots, which is faithful to what the nominal group technique exists for: protecting the result from dominant personalities.

Sprint, by Jake Knapp, John Zeratsky and Braden Kowitz, layers three votes into the Wednesday of the design week. For the heat map, everyone receives twenty to thirty dots, placed in silence on every appealing detail of the solutions on display, including two or three stacked on the most exciting ones: this vote is about fragments of a solution. The straw poll gives one vote each, with a reason said out loud afterwards. The supervote closes: the designated Decider holds three votes marked with their initial and their placement outranks the group's. Citing Sprint as a reference for dot voting without naming that last layer would give a false picture: the process there is not one person one vote.

The Nielsen Norman Group adds a fourth benchmark, on the number of votes alone: around a quarter of the number of options. On a list of twenty items that gives five votes, which lines up with Gamestorming; on a list of eight items it gives two, where five votes would let everyone vote for more than half the list. With too few votes, the count returns a majority of one-vote ties; with too many, every option receives one and nothing separates.

Dot voting

One voter, three sets of rules

Where one voter's dots may land

Running the session

1. Clean the list before handing out the dots
   Merge duplicates and wordings that say the same thing.
2. Announce the full rule before the first dot goes up
   Number of votes, stacking allowed or not, silence or not, duration. A rule stated afterwards looks like an arrangement of the result.
3. Time-box the placing
   Three to five minutes is enough for twenty or so options.
4. Count out loud and write the total on each card
   The ranking becomes a fact the room has witnessed.
5. Record the result and what it commits
   The vote count, the date, who was in the group and the decision taken afterwards on that basis.

What distorts the count

The most documented bias is the one the Nielsen Norman Group calls group think. Someone who hesitates reads the dots already placed before choosing, which widens the lead of the options already in front. The article proposes two counters. The first has the people with the least hierarchical seniority vote first and keeps the managers for the end. The second asks everyone to write their choice on a slip before going up to place a dot, so that the private commitment comes before the public act. Sprint answers the same risk with the silence rule during the heat map, for a reason it states: a well-argued case distorts the opinion of whoever hears it, and knowing where an idea comes from distorts the judgement passed on it.

Persuaded voting is the spoken version of the same thing. A stakeholder makes the case for their option while the others are voting, and the dots follow their authority. Discussion belongs before the vote or after the count, with the same speaking time for each option.

Split voting plays out upstream, in the list itself. Three cards expressing the same idea in three wordings collect three votes each, where the merged idea would have taken nine and come out on top. That is the reason for cleaning the list first. Downstream it shows up as a tight pack at the head of the count, four items between five and six votes, where the gap says nothing beyond the luck of how the allowances fell. That is the situation the tie-break round handles.

That leaves the most expensive confusion, the one that takes the count for the decision. Dot voting produces a group preference, on a date, in a given composition. Whoever answers for the product goes on answering for it, and they sometimes keep an option that came third for reasons the room did not have. Sprint handles this by giving the Decider a supervote that carries, which at least announces the rule before the vote. A group that is made to vote in the belief that the count will settle the matter, then sees its result set aside, will not vote in good faith again.

AI considerations

A language model given the twenty wordings that came out of the brainstorming proposes the duplicate merges and picks out options of uneven grain, a very broad idea sitting next to a detail improvement. Fixing the list is worth more than anything a model would do with the result, since split voting and uneven grain distort the count from the outset. The group validates the proposed merges, a few seconds per pair.

After the count there is nothing to compute: a tally is an addition. The temptation is to ask for an analysis of the results, and a model will produce one, with trends and an emerging consensus that forty-five dots cannot support. The useful output is limited to formatting, the vote table copied into the minutes and the selected items entered into the product backlog with their score and the date.

A model cannot vote. Simulating voters to stand in for absent stakeholders destroys the only product the technique has. A dot voting result holds as a commitment by people who were in the room, who placed their own dot and who will have to live with the choice. A synthetic voter has no stake in the decision and no account to give for it. A workshop option list often contains unannounced projects and acknowledged weaknesses of the product, so handing it to an outside service is a confidentiality decision.

Examples

A retail bank in French-speaking Switzerland is working on its online account-opening flow. A workshop of nine people, from product, client relations, compliance, user experience and IT, produced eighteen improvement proposals. Everyone receives five dots, stacking is allowed, three minutes to vote.

The count starts with a check sum: the total of the votes must equal the number of voters multiplied by the allowance, forty-five in this workshop. A gap signals dots that were never placed or that fell off, and it can be repaired while the room is still there.

Two options contest third place with six votes each. A second round limited to those two cards, one vote per person, separates them six to three. This tie-break is a common facilitation practice, borrowed from the repeated round of ASQ multivoting.

That leaves the ten options that collected no votes at all. They are read out one by one before being set aside, because a card pinned in a corner of the wall has not been rejected: it went unnoticed. Gamestorming flags this point. Two of them are kept in the product backlog: they answer a compliance requirement the vote has no business arbitrating.

Visualisations

A tally is made of rows and columns, so it is written in HTML. The table can be sorted, copied into the product backlog and recounted, none of which a photograph of the wall allows. The bar in front of each number makes the gap visible without reading the figures, and the two tied rows show the only situation that calls for a second round.

The vote itself is a matter of positions and densities, so it is drawn: a wall of cards, dots piled up in uneven quantity, one card well ahead, a mid-tier cluster, cards left blank, a stack of three dots from a single voter on one option and two tied cards joined by a bracket pointing to a second round. That view carries the input, the act and the output of the technique in one image.

A third drawing carries what the two tables do not show. The vote table gives totals, the rules table says whether stacking is allowed; the drawing shows where one voter's dots may land: five identical dots that may all go on the same card, five distinct weights that must be spread over five different cards or a single vote.

Cost

Tools

Sticky dots remain the reference tool on site, with different colours when votes are to be told apart by role, for instance compliance votes from those of the delivery team. A marker does the job too: a cross is worth a dot. The physical wall has one advantage: the result is readable by the whole room as soon as the last dot goes up. It carries the matching defect, that same readability during the vote.

Online whiteboards, Miro, Mural, FigJam or Microsoft Whiteboard, carry a voting function that treats this defect. The running count stays hidden until the end of the round, the allowance is enforced by the tool without anyone having to watch how many dots have gone up and the result exports without manual transcription. For a distributed workshop or for a group where hierarchy weighs, that is the best reason to prefer the tool to the wall. In-session polling tools, Mentimeter or Slido, go further towards anonymity, at the cost of the overview: the whole wall is no longer in front of the voter while they choose.

Downstream, the spreadsheet takes the totals and keeps the trace of the successive rounds. A product backlog accepts the score as an attribute of an item.

Sources

• Dave Gray, Sunni Brown and James Macanufo, Gamestorming: A Playbook for Innovators, Rulebreakers, and Changemakers, O'Reilly Media, 2010, chapter 4 "Core Games", section "Dot Voting": the five-vote rule as an order of magnitude, stacking explicitly allowed on one item, the simultaneous placing of the dots, the check on options left with no votes and the lengthening of the count in large groups. No tie-break round is provided there.
• American Society for Quality, Multivoting and Nominal Group Technique, Quality Resources: the five-choice ballot weighted 5 to 1, the narrowing variant where each round keeps the top third of the surviving list, the written ballot and the repeated round as a tie-break mechanism. The lineage goes back to André Delbecq, Andrew Van de Ven and David Gustafson, Group Techniques for Program Planning: A Guide to Nominal Group and Delphi Processes, Scott Foresman, 1975.
• Jake Knapp, John Zeratsky and Braden Kowitz, Sprint: How to Solve Big Problems and Test New Ideas in Just Five Days, Simon & Schuster, 2016, the Wednesday "Decide" part: the heat map with twenty or thirty dots placed in silence on fragments of solution, the straw poll at one vote per person and the supervote of three votes reserved for the Decider, which carries the decision. Cited also for the reasoning behind the silence rule.
• Sarah Gibbons, Dot Voting: A Simple Decision-Making and Prioritizing Technique in UX, Nielsen Norman Group, 7 July 2019: the allowance of around a quarter of the number of options and the three biases it names, persuaded voting, split voting and group think, with their counters.