Two-Phase Activities: Collect, Then Vote

Two-phase activities are activities where the group first submits ideas openly, then votes on what the group produced. Phase one is generation and nothing is judged; phase two is selection and nothing new is added. Keeping those two jobs apart is what makes the format work, and collapsing them into one open discussion is what makes most brainstorms disappointing.

The pattern shows up everywhere once you notice it. Collect risks, then vote on which to plan for. Collect retro observations, then vote on which to fix this sprint. Collect questions for the exec, then vote on which get answered. In every case the shape is identical: an open text or word cloud collection, a short curation pause, then a poll or ranking built from the answers the room just gave you.

This guide covers the full run: what each phase is for, what to do in the gap between them, how to size the ballot, the timings that hold a room, and the four ways the format fails. Everything here works for a team of eight around a table and for three hundred people on a webinar, with the differences called out where they matter.

What a two-phase activity actually is

The defining rule is that the ballot does not exist until the group has spoken. You do not walk in with five options you wrote on the train. You walk in with a question, collect whatever comes back, and build the vote from that. That is what makes people take the second half seriously: they are choosing between their own words, not ratifying yours.

It also means you cannot fully prepare the second half, which is why facilitators avoid the format. You have to do live editing in front of an audience. The good news is that this is a learnable, mechanical skill, and the room is far more forgiving of a facilitator merging duplicates on screen than of a facilitator asking a question nobody can answer.

The closest formal ancestor is the nominal group technique, where people generate silently, contributions are pooled without attribution, and only then does the group rank them. The two-phase live activity is the same logic compressed into six or seven minutes and run on the devices people already have in their hands.

The four moves in order

Every failure of this format is a failure of one of those four moves, usually the third. Facilitators who are comfortable collecting and comfortable voting still lose the room in the two minutes between, because they go quiet and start editing while thirty people watch a screen with no narration.

Phase one: collect without contaminating the pool

The goal of phase one is volume and independence. You want as many genuinely separate ideas as the room holds, which means nobody should hear anyone else's idea before their own is submitted. In an open discussion this is impossible: the first two speakers set the frame and everyone else works inside it, which is the production-blocking problem that makes verbal brainstorms weaker than silent ones of the same size.

So run phase one silently and, if the tool allows it, hide the incoming feed while people type. If you cannot hide it, put it up but tell people plainly: type your own before you read anyone else's. That instruction works better than it should.

Cap the submissions per person. Three is the number that works for most prompts. Uncapped collection in a group of forty produces two hundred entries, of which you will read maybe sixty, and the people whose entries you skipped will notice. A cap of three with forty people gives you a manageable hundred-and-twenty ceiling and forces each person to pick their strongest.

Word the prompt so the answers are comparable

Answers you cannot compare cannot be voted on. 'What do you think about the roadmap?' returns a mixture of complaints, questions, praise and predictions, and there is no honest way to build one ballot from that mixture. 'What is the biggest risk to hitting the March date?' returns one kind of thing.

Ask for a noun phrase, not a sentence, when the output is going to become a ballot option. 'Name the blocker in five words or fewer' is a real instruction and people follow it. It also saves you most of the merge work, because short entries make duplicates obvious at a glance.

The merge step is the whole job

When submissions close, you have raw material, not a ballot. Somewhere between a fifth and a third of entries will be duplicates in different words, a handful will be off-topic, and one or two will be jokes. Your job is to turn that into eight to twelve options in under four minutes, in public.

Narrate while you do it. 'I have got four versions of the same handover point, I am merging those into one and calling it handover between design and build - shout if that loses your meaning.' The narration is not politeness, it is the mechanism that keeps the shortlist legitimate. A shortlist that appears silently looks like the facilitator's opinion, and people vote on it accordingly.

Two rules make the merge defensible. First, never delete a minority view because only one person said it - a single entry that names something no one else spotted is often the most valuable line on the board. Second, when you are unsure whether two entries mean the same thing, keep them separate and let the vote decide. Over-merging is much harder to undo than under-merging.

Merge checklist for the two minutes between phases

Phase two: pick the voting mechanism to match the decision

Not every second phase is dot voting. The mechanism should follow from what you will do with the result, and picking the wrong one produces a tidy chart that nobody can act on.

If you are choosing one thing to do next, single-choice voting is honest and fast. If you are choosing a small set to carry forward, multi-select with a budget of three is better, because it lets people support a second preference without abandoning their first. If you need a defensible order rather than a winner, ranking is worth the extra thirty seconds it costs. And if the question is about strength of feeling rather than choice, a rating scale on each shortlisted item tells you more than a tally does.

Hide the tally while people are voting

In phase one you kept the pool clean by stopping people reading each other's submissions. Phase two needs the same protection for the same reason. A visible running tally tells people what the room already prefers, and later voters drift towards it - the bandwagon effect in its purest form. Early votes also anchor the range people consider reasonable at all.

So hide results while the vote is open and reveal in one move when it closes. The reveal is a better moment anyway: a bar chart that appears fully formed gets a reaction, one that creeps up for sixty seconds gets a shrug.

There is one exception. When the decision has effectively been made and the vote exists to build visible commitment behind it, showing the tally as it grows is the point. Be honest with yourself about which of those two situations you are in.

Timings that hold a room

Four failure modes and the recovery

When one phase is enough

The format costs you eight to twelve minutes and a chunk of facilitator attention. Do not spend that when a single activity does the job. If you already know the options and you only need a decision, just run the poll. If you only need to see what is on people's minds and no choice follows, just run the word cloud and talk about it.

Two-phase earns its keep when the options are genuinely unknown to you, when ownership of the shortlist matters, or when you suspect the obvious answer is not the real one. It also earns its keep in rooms where a senior voice would otherwise set the agenda in the first ninety seconds, because silent collection puts that person's idea on the board as one line among forty rather than as the frame everyone else responds to.

Running it without a stack of tools

The practical obstacle is that most setups make you collect in one tool and vote in another, which means a copy-paste job in front of the room and a gap where the energy drains out. Session Flo is built around this specifically: an open text or word cloud activity feeds a poll built from the submitted answers, so the ballot is created from the room's words without leaving the session or asking anyone to join a second link.

Whatever you use, rehearse the handover once. Run the collection with two colleagues the day before, practise merging live, and time yourself. Four minutes of editing feels like a very long time when people are watching, and knowing that in advance is most of the battle.

Finally, close the loop. Put the shortlist and the tally in the session recap with a line against each of the top three saying who owns it. A two-phase activity that produces a winner nobody follows up teaches the room that both phases were decorative, and next time you will collect half as many entries.