Key takeaways
- →Treat a 45-minute slot as three blocks of roughly twelve minutes, each ending with something the audience does rather than watches.
- →The first contribution has to land before minute five - after that the room has already decided it is an audience and will not switch.
- →Open the question queue before you start speaking, because the best questions occur to people mid-talk and are forgotten by minute forty.
- →Only formats that work from a seat scale past 200 people: polls, word clouds, upvoted questions, quick ranking and a turn to your neighbour.
- →Test the join flow from the back row on venue wifi, and carry a no-network fallback you can run with raised hands.
- →Put the poll results and the unanswered questions into a recap link on your closing slide, or the session ends when the room stands up.
Conference sessions beat the lecture format when you stop treating a 45-minute slot as one continuous talk and start treating it as three blocks of about twelve minutes, each ending with something the audience actually does. The doing does not have to be elaborate - a two-option poll, a one-word word cloud, sixty seconds turned towards the person next to you - but it has to be real, and it has to be scheduled before you walk into the room rather than improvised when you sense the energy dropping.
The reason is not that lecturing is a moral failing. It is that a conference audience is on hour four of sitting down, has a phone in its hand, and carries no accountability whatsoever for what you say. Attention in that setting is not earned once at the start and then kept; it decays through the slot, and the only reliable way to reset it is to change what the room is doing, not what the slide says.
This piece covers the shape of a 45-minute interactive session, five formats that still function at 400 seats, the room and tech checks that stop the whole thing collapsing at minute three, how to run Q&A that produces questions instead of speeches, and what to do when a large room quietly declines to answer you.
Why conference sessions default to the lecture
The lecture is not chosen because speakers love the sound of their own voice. It is chosen because it is the only format that cannot visibly fail. A talk nobody enjoys still finishes on time and looks like a talk; a poll nobody answers happens in front of 400 people and takes about eleven seconds to become unbearable. Given unfamiliar AV, a deck approved three weeks ago and no relationship with the room, most speakers pick the invisible failure over the visible one.
The cost lands on the audience rather than the speaker, which is exactly why the pattern survives. The evidence on active learning is unusually consistent: sessions that make people do something with the material outperform straight presentation on what people can still use afterwards. A conference is the hardest possible case for that, because there is no assessment, no follow-up and four other talks the same day competing for the same shelf in someone's memory.
Organiser incentives do not help. Sessions are booked on title and speaker reputation, scheduled back to back with ten-minute changeovers, and evaluated afterwards with a form asking whether the content was relevant. Nothing in that pipeline rewards you for making the room work. If you want a session that beats the lecture format, you will usually design it yourself and mention it to the organisers afterwards.
The shape of a 45-minute slot that is not a lecture
A 45-minute slot has room for roughly three ideas and three contributions. Push past that and you are rushing the audience through work they cannot finish; drop below it and you are padding. The shape below assumes a standard conference slot with a changeover either side, so it is built to finish at minute 44 and leave the stage clear.
0:00 - 0:03 Open with a question
Join code already on the holding slide. One two-option poll about the problem, not about the audience. Read the split out loud before you introduce yourself.
0:03 - 0:15 Block one: the setup
One idea, the evidence for it, and the thing most rooms get wrong. Twelve minutes is the longest stretch of uninterrupted talking the slot can carry.
0:15 - 0:18 First reset
A one-word word cloud or sixty seconds turned to a neighbour. Cheap, parallel, and it converts the room from receiving to producing without anyone standing up.
0:18 - 0:30 Block two: the hard part
The material people came for, delivered while attention is highest post-reset. Put your most contestable claim here so the queue fills up with real questions.
0:30 - 0:36 Apply it
The audience ranks, estimates or picks something using what you just gave them. Six minutes including the reveal and your thirty seconds of commentary.
0:36 - 0:44 Q&A from the queue
Take the top upvoted questions in order. Read each one verbatim, answer in ninety seconds, move on. No microphone runners, no roaming.
0:44 - 0:45 Close and hand over
One sentence on what to do on Monday, and the recap link with the poll results and the unanswered questions. Then stop talking.
Two rules the shape depends on
The first contribution has to happen before minute five. A room that has watched you talk for fifteen minutes has already worked out that this is a listening session, and it will not switch modes on request - you will ask for a poll at minute twenty and get a response rate of about a fifth. Whatever else you cut, do not cut the opening ask.
The question queue has to be open before you start speaking. Good questions occur to people during the talk, at the exact moment something does not fit what they already believe, and they are gone by minute forty. Collecting from minute zero also removes the worst part of conference Q&A: the hunt for a microphone while 400 people watch someone decide whether they really want to ask.
Session Flo runs the poll, the word cloud and the question queue from a single room code, which matters more at a conference than anywhere else. You get one join, on the holding slide, while people are still finding seats. A second join step at minute eighteen loses the back third of the room and you never get them back.
Five interaction formats that survive a big room
Not every interaction scales. Anything requiring people to move, form groups or queue for a microphone falls apart somewhere between 100 and 200 seats, and usually the reason is the furniture rather than the audience. Fixed theatre rows kill anything that needs a circle. The formats below are the ones that work from a seat, on a phone, with no instructions beyond one sentence.
| Format | Works at | Time cost | Fails when |
|---|---|---|---|
| Two-option live poll | 50-2000 seats | 60-90 sec | The two options are not genuinely contested |
| One-word word cloud | 50-2000 seats | 90 sec | The question is so narrow everyone types the same word |
| Turn to your neighbour | 30-500 seats | 2-3 min | Rows are fixed and nobody can physically turn |
| Live ranking or dot vote | 40-800 seats | 3-4 min | More than seven options are on screen at once |
| Upvoted question queue | 50-3000 seats | Runs throughout | You open it at minute 40 instead of minute zero |
| Scored quiz on the material | 30-500 seats | 4-6 min | Questions test recall of your slides, not judgement |
| Numeric estimate then reveal | 30-1000 seats | 2 min | The real answer turns out not to be surprising |
Which of the five to actually pick
The numeric estimate is the most under-used and the cheapest to run. Ask the room to guess a number you are about to reveal - how long the migration took, what proportion of tickets came from one feature, how many of their peers do the thing you are about to criticise. Aggregate guesses from a large room are often closer than any individual guess, which gives you a second thing to talk about when the distribution comes up on screen.
The scored quiz is the one to be careful with. Retrieval practice genuinely helps people keep what they have just learned, but a quiz that tests whether people remember your slide titles teaches nothing and feels like a test. Write questions about judgement instead: give a short scenario and three plausible responses, and make the wrong answers the ones a reasonable person would pick before hearing your talk.
The turn-to-your-neighbour is the only format on the list that gets better as the room gets more tired, and the only one that fails purely for architectural reasons. Check the seating the day before. In fixed rows with no gaps it becomes an instruction people cannot obey, and asking a room to do something impossible costs you more credibility than skipping the interaction entirely.
Numbers to design your slot against
These are design constraints rather than research findings - they are the numbers that make the shape above fit into the time you have been given. The twelve-minute block is the one people argue with, and the argument is usually 'my material needs longer'. It rarely does; what it needs is a different split, because a thirty-minute stretch of talking gets remembered as one undifferentiated lump regardless of how good the middle was.
The seven-option limit is about the back of the room, not about cognition. From row thirty, a screen with eleven ranked options is unreadable, and people vote for the two they can make out at the top. If you genuinely have twelve candidate items, run two rounds of five to seven rather than one round of twelve, or collect the long list as text and put the shortlist to a vote.
The first three minutes decide everything else
Open with a question, not with your biography. The audience read the abstract and picked this room over three others; a slide of client logos spends the highest-attention two minutes of your slot on the least useful content in your deck. Introduce yourself in one sentence after the first poll result is on screen, when you have earned the right to.
Get the join code up before the room fills. It belongs on the holding slide the previous speaker leaves behind, alongside a five-word instruction: 'join now, first question in two minutes'. People who join while wandering to their seat cost you nothing; people asked to join at minute two while you wait in silence cost you the momentum you opened with.
Then say the response count out loud as it climbs. 'Sixty... a hundred and forty... two hundred and ten, that will do.' Counting converts a room full of private typing into a shared event, and it tells the undecided that participating is what is happening right now. Finish by changing something visible in response to the result - skip a section, reorder two blocks, say which answer surprised you. The room needs one piece of evidence that answering changes what you do next, and it needs it in the first three minutes.
What to check before you walk on stage
Venue wifi is the usual point of failure, and it fails in a specific way: fine at the front, degraded at the back, catastrophic when 400 devices join in the same ninety seconds. Test from the worst seat in the room during a busy slot, not from the stage at eight in the morning. If the network cannot take it, run the first interaction on hands and keep the digital activity for the smaller, later ask.
The last item on the list is the one that changes the session most. Someone watching the question queue while you speak can flag the question that is being asked in six different ways, and can tell you at minute thirty-two which theme the room actually wants. Doing that yourself, mid-sentence, on a laptop turned away from the audience, is not realistic.
- ✓Join code and QR tested from the back row on venue wifi, not from the stage
- ✓Results screen font size checked from the last row of seats
- ✓Question queue opened before the previous session ends
- ✓AV team briefed on who switches the screen when results go up
- ✓Fallback ready for no network: a show of hands or a count of fingers
- ✓Poll questions re-read on a phone screen, where the audience will see them
- ✓Seating checked for whether people can physically turn to a neighbour
- ○A host or co-speaker briefed to watch the queue while you talk
Run Q&A as a queue, not a microphone hunt
Traditional conference Q&A selects for confidence and volume, which is why the same three question-shaped statements turn up in every session. A written, upvoted queue changes who gets heard: the quiet specialist in row twenty submits at minute nine, forty people upvote it, and it goes first. Run it in this order.
Open the queue before you speak
Same room code, same join. Say once, at the start, that questions can go in at any point and that the room will vote on which ones you take. That single sentence is what makes people type at minute nine instead of waiting for permission at minute forty.
Let the room upvote instead of curating yourself
Picking questions yourself puts you in the position of visibly ignoring people, and you will unconsciously pick the flattering ones. Upvoting hands the choice to the audience and gives you cover for skipping the question about your pricing that four hundred people declined to support.
Read each question out verbatim
Reading it as written, without paraphrasing it into a question you would rather answer, is the whole trust mechanism. The person who wrote it hears their own words on the main screen, and the rest of the room learns that submissions are used rather than filtered.
Cap answers at ninety seconds and say the cap out loud
'I am giving each of these about ninety seconds so we get through five.' The cap protects the queue from your own tendency to give a complete answer to the first question and a rushed one to everything after it. Five short answers beat two thorough ones almost every time.
Take one uncomfortable question deliberately
If the third-highest question is the sharp one about what did not work, take it and answer it plainly. Skipping it is more visible than answering it, because the whole room can see the queue. Answering the hard one is also the fastest way to get better questions in the remaining minutes.
Close with a promise rather than an apology
Do not end on 'sorry, we are out of time'. End on 'there are eleven questions left, and the answers go into the recap with the poll results'. Then actually send it. The queue is a record of exactly what your audience wanted to know, which is worth more than the feedback form.
Panels are the hardest format to make interactive
A panel has four people who each want airtime, a moderator who wrote questions in advance, and an audience with no way in until the last eight minutes. The default outcome is four sequential mini-lectures and a Q&A that reaches two people. Interactivity has to be designed into the structure, not bolted onto the end.
The move that works is to let the audience set the agenda rather than the moderator. Collect questions and topics from the room in the first three minutes, put the top three on screen, and run the panel through those in order. It costs the moderator their prepared script and gains a panel that is visibly about the room's problems. It also gives you an honest way to cut a panellist off: the next topic is on screen and everyone can see it.
Use forced-choice polls between topics to break the pattern of speech-after-speech. 'Before they answer: would you ship it or delay it? Pick one.' Put the split on screen, then ask a panellist who disagrees with the majority to go first. Panels flatten into agreement very quickly, and a visible disagreement between the room and the stage is the most reliable way to keep the next ten minutes honest.
Workshop slots and breakouts at a conference
Top Tips
- Design for the people who wandered in by mistake. A conference breakout has no pre-work, no shared context and no obligation, so the first activity must be doable by someone who read only the session title.
- Group by table, never by counting off. Asking sixty strangers to form groups of six burns four minutes and produces one group of eleven and two groups of two. Whoever is sitting near you is your group, and eight to twelve people per table is the practical ceiling for one conversation.
- Give every group a named output and a countdown on the main screen. 'One sentence, on the board, in seven minutes' produces output; 'discuss this among yourselves' produces a conversation that stops when the noise level drops.
- Collect from tables digitally rather than by touring the room with a marker pen. One person per table types the group's answer into the same activity, everything lands on one screen, and you skip the ten minutes of report-backs where each table repeats the previous one.
- Halve the content you planned. A 60-minute breakout with real group work carries about twenty minutes of your material. The most common failure is a workshop slot that is a talk with a rushed exercise glued to the end.
- Plan for the room being wrong. Conference breakout rooms are booked by capacity, not layout, so assume you may get theatre rows when you asked for tables and keep a version of the activity that works with people staying in their seats.
When a large room declines to answer
Ninety seconds in, thirty responses from a room of 300, and the counter has stalled. This is common, it is recoverable, and it is never fixed by repeating the question more brightly. Name it first: 'thirty out of three hundred, so either my question is bad or the wifi is'. Saying the number aloud removes the awkwardness faster than pretending nothing is wrong, and it buys you licence to change the ask.
Then shrink the request. Move from open text to two options, or from 'what is your biggest constraint' to 'is your constraint money or people - pick one'. A binary choice collects from everyone who had decided not to type a sentence in front of strangers. If the network is genuinely the problem, switch to hands immediately and keep going: 'hands up if it is money'. A room that has answered once with its hands will answer the next digital question at a much higher rate.
If two attempts have failed, stop asking and start doing. Move into your strongest content, get the room engaged with the material itself, and reintroduce a small ask fifteen minutes later once you have their attention on merit. Refusal early in a slot is usually about the previous session or the walk between rooms rather than about you, and it evaporates once people have a reason to care.
The same slot, before and after
Nothing in the right-hand column requires more preparation than the left - it requires different preparation. The three blocks are the same content, split. The poll is one slide. The queue runs itself. The genuinely new work is deciding, in advance, what the audience will do at minute fifteen and minute thirty, and writing those two asks properly.
How a session is remembered is disproportionately shaped by its peak and its ending, which is an argument for spending your preparation time on the applied section in the middle and the last ninety seconds, rather than on smoothing the transitions in block one. Most speakers rehearse the opening obsessively and improvise the close. Swap that round.
✗ A standard 45-minute conference talk
- •Thirty-eight slides, one voice, no break in the pattern
- •Two minutes of biography and client logos at the start
- •Half the room on email by minute fifteen
- •'Any questions?' at minute 43, answered by two people
- •One microphone runner for four hundred seats
- •Nothing to take away but a photo of the summary slide
✓ The same 45 minutes, redesigned
- •Three blocks of about twelve minutes, one idea each
- •A two-option poll at minute two, read aloud from the stage
- •A word cloud reset after the densest section
- •Questions collected from minute zero and ranked by the room
- •Six minutes where the audience ranks something themselves
- •A recap link carrying the poll results and unanswered questions
Give people something to leave with
A conference session ends when the room stands up, unless you give it somewhere to continue. The cheapest version is a link on your final slide that carries three things: the poll and ranking results the room produced, written answers to the questions you did not reach, and one page of what you actually said. It takes minutes to assemble if the activities were digital, and it turns a slot into something people forward to a colleague on Tuesday.
Make the close a single instruction rather than a summary. 'Pick one thing you ranked in the top three and run it as an experiment next sprint' is actionable; 'thank you, I hope that was useful' is not. Say it, hold it on screen for the last thirty seconds, and stop - the final image people carry out should be the instruction, not a slide of your contact details.
Then ask the organisers for the one thing that improves your next session more than any feedback form: the queue. The questions your audience wrote, in their words, ranked by how many people wanted them answered, are the most honest description you will ever get of the gap between the talk you gave and the talk the room needed. Read them before you write the next version.
Frequently asked questions
How interactive should a 45-minute conference session be?
Three audience contributions is the right target for 45 minutes: one in the first three minutes, one around the halfway reset, and one applied activity before Q&A. That leaves roughly thirty minutes of actual content, which is more than most speakers deliver clearly in forty-five anyway. Going beyond three starts to feel like a workshop that keeps interrupting itself, and each interaction costs you between ninety seconds and six minutes once you include the instruction, the responses and your commentary on the result.
What if the conference wifi cannot handle everyone joining at once?
Assume it might not and design a fallback you can run without embarrassment. Hands, a count of fingers one to five, or standing up for one option and staying seated for the other all work at any size and take no setup. Test the join flow from the back row during a busy slot rather than from the stage first thing in the morning, and if it is marginal, put the digital activity later in the slot when fewer devices are joining simultaneously.
Do interactive formats work for keynotes as well as breakout sessions?
They work better in keynotes, because the room is larger and the alternative is worse. A thousand-seat auditorium cannot do group work, but it can answer a two-option poll in ninety seconds and it can fill a question queue. The constraint is the format rather than the size: pick things that work from a seat with a phone, keep the on-screen results legible from the back, and be more disciplined about timing than you would be with sixty people.
How do I collect questions without the queue filling up with noise?
Let the room upvote and take the questions in that order. Voting is what filters the queue, and it does it more fairly than you would - people reliably promote the specific, uncomfortable questions and ignore the self-promotional ones. Show the queue on screen during the interaction breaks so people can see what has already been asked, which cuts duplicates dramatically, and answer the top items verbatim rather than paraphrasing them into softer questions.
Should conference session polls be anonymous?
Yes, for essentially everything at a conference. People are surrounded by strangers, competitors and sometimes their own manager, so named responses suppress honesty on anything about how their organisation actually operates. Anonymous polling gets you the real distribution, which is the only reason to run the poll. The one exception is a prize-driven quiz, where you need names to award something - and even then, only the leaderboard needs to be named.
How do I convince organisers to let me run an interactive session?
You usually do not need to ask. Everything described here happens inside your own slot, using your own slides and the audience's phones, so it requires nothing from the AV team beyond a screen you already have. Two things are worth flagging in advance: that you would like the join code on the holding slide before your slot, and that you will finish at minute 44 rather than running into the changeover. Organisers care about the second one far more than the first.
Keep reading
- participation techniques built for big rooms — More formats that work when the audience cannot move from their seats.
- running a session for 100 or more people — The logistics side of scale: joining, timing and reading results from a stage.
- Q&A that avoids the awkward silence — Going deeper on queues, upvoting and handling the hostile question.
- interactive presentations that are more than slides — How to rebuild a deck around what the audience does rather than what you say.
- run your conference session with Session Flo — Polls, word clouds, ranking and an upvoted question queue from one room code.