Delivering with Confidence: Rehearsal, Q&A, Remote and Hybrid

Presentation track, branch pr-delivery (Delivery) · Node PR6.01 · ~6 focused hours · Mastery gate ≥ 85%

A silent read-through of a twenty-two slide deck, seated at a desk, takes 15 minutes and 13 seconds. The same deck, stood up, spoken out loud, with a stopwatch stopped and restarted on every slide, takes 25 minutes and 17 seconds. The difference is 604 seconds, and the stopwatch run is 1.66 times the desk run. If the slot is 18 minutes, the desk rehearsal says you have nearly three minutes spare and the stopwatch says you are 7 minutes and 17 seconds over. One of those numbers is going to be true in the room, and it is not the comfortable one.

Almost everything that goes wrong in a delivery has a number attached to it before you walk in. The deck is too long for the slot, and the arithmetic of that was available on paper. The Q&A window is 11 minutes and the six questions you will actually be asked need 6 minutes and 15 seconds if you handle them the way you planned, or 11 minutes and 50 seconds if you let two of them run. The camera is 20 centimetres below your eyeline, which at a normal working distance is a 20 degree downward gaze that every person on the call reads as distraction.

None of that is temperament. All of it is arithmetic you can do the day before.

There is a second category, and the boundary deserves to be stated plainly. Whether you are interesting, whether the room trusts you, whether you can hold a hostile director's attention while telling him his favourite project does not pay back: those come from reps. They come from having been wrong in public and survived it, and from caring more about the answer than about how you look while giving it.

No six hours produce that. What six hours can do is remove the failures that have nothing to do with courage, so that the nerve you do have gets spent on the interesting part.

Three working assumptions sit behind everything below, and they are assumptions rather than measured constants. An audience carries about three claims out of a room. They settle within the first ninety seconds on how much attention to lend you. And they judge your competence far more on the questions than on the prepared part. The rehearsal routine exists so that the prepared part fits and stops. The Q&A method exists because the unprepared part is where the assessment happens. The remote and hybrid material exists because a camera 20 centimetres too low and a notification banner at minute 12 will undo both.

One warning about the sequencing. The word "rehearsal" makes people think of memorising a script, and a memorised script is the second-worst way to deliver an analysis. It sounds recited, it collapses the moment somebody interrupts, and it puts every one of your words in competition with the thinking you should be doing about the room. What you memorise is the message and the numbers. What you leave loose is the sentences.

Learning objectives

1. Build a rehearsal plan of four passes, timed with a stopwatch, and state what each pass is for and what it will not tell you.
2. Compute a slide budget from a slot length and a Q&A share, and plan the cut on paper before the deck is ever shown.
3. Write and time a ninety-second opening that reaches its first fact inside two seconds, within a word budget that accounts for the pauses.
4. Classify any question into answer, bridge, park, defer or commit, and deliver each class in the words that class requires.
5. Handle a hostile question by separating the challenge from the challenger, and quantify what mishandling it costs the rest of the window.
6. Answer a question you cannot answer, in a form that leaves your credibility intact and produces a dated follow-up commitment.
7. Set up a remote delivery: camera height, framing, light direction, and the two settings that matter more than any equipment purchase.
8. Run a screen share that never leaks a notification, and state the probability that it would have leaked one if you had not.
9. Design a hybrid meeting that closes most of the per-head airtime gap between the room and the line, and measure the ratio before and after to see whether it worked.
10. Keep a delivery log after every presentation, and read from it the two numbers that show whether you are improving.

Prerequisites & connections

Builds on. Nothing outside the branch. Delivery is the first and only node in pr-delivery. It unlocks without a gate, and you won't need any software beyond whatever tool your deck is already built in. You'll get more out of the mini-project if you already have a deck of your own with an argument in it, which is what PR4.03 produces. If you don't, the project supplies a fallback brief.

Feeds forward. PR7.01, the capstone, requires delivery notes as part of its submission: a ninety-second opening and three anticipated questions with their answers. You'll produce those notes with the Q&A classification and the timed rehearsal. The archetype modules PR5.01, PR5.02 and PR5.03 each end with a deck that has to be presented to somebody, and the timing arithmetic here is what sizes them.

The non-overlaps are deliberate, and each of these modules owns something delivery leans on. PR0.01 owns audience, message and medium, including the distinction between a reading deck and a presented deck. The word budgets below assume that distinction and don't re-argue it. PR2.01, PR2.02 and PR2.03 own chart choice, decluttering and the action title, so when a worked example here builds a chart it applies that grammar without teaching it again. PR3.02 owns the ghost deck and the storyboard. PR4.01 and PR4.02 own the executive storyline and the horizontal and vertical logic of titles, which is what makes a planned cut possible at all: you can only drop a slide safely when the title stack tells you which slides carry the argument. PR5.01 to PR5.03 own the deck archetypes and their templates. M10.03 owns written communication, including the research report and the memo, so nothing here is about what you write down. ES1.03 owns the disclosure obligations that attach to presenting investment views to anybody outside your firm, and a delivery habit never overrides a compliance requirement.

Core teaching content

3. Pace, pause and the silence after a number

Under pressure, your speaking rate rises. Thinking about your rate doesn't help. Installing pauses at fixed places does, because a pause you planned survives adrenaline and a pause you intended to feel doesn't.

Put the pause immediately after a number. You say "nine of the thirty-four sites lose money" and then stop for two seconds. In that silence the audience does the division and gets to roughly a quarter on their own. Listeners hold on to an arithmetic conclusion they reached themselves far better than one you handed them, and it costs you two seconds.

The pause budget is small enough to be reassuring. A twenty-minute talk carrying 14 key numbers, at two seconds a pause, spends 28 seconds pausing. That's 2.3 percent of the talk, and it displaces about 61 words.

Nobody has ever run out of time because of pauses. They run out of time because of slides.

Where else should you stop? After the takeaway sentence, before you start the evidence. Before the ask, because the gear change wants a beat. And always after a question, before you begin answering it. The pause reads as consideration, and it buys you the second you need to decide which class the question belongs to. There are five, and they're next.

Measure your own rate once. Record two minutes of real content and count the words, because there's no other way to know. The planning band used throughout is 110 to 150 words a minute, with 130 as its midpoint. That band moves the opening budget by 60 words across a 90-second window, which is a third of the opening. Once you have your own number, use it for every word budget here and drop the 130.

Volume and articulation get one paragraph. They're simple, and they're rarely the binding constraint. Speak to the back of the room. Land the ends of your sentences so they don't fall away. Take the breath before you need it. The swallowed final clause is the failure that really costs comprehension, because that's usually where the conclusion is.

4. Reading the room, and changing depth without changing the message

When you adjust to a room, you change the depth of the evidence and leave the message alone. The takeaway you wrote is the takeaway you give, whatever happens in front of you. What flexes is how much of the working you show and how much you skip.

Watch for three signals. They're easier to read than people expect. Pens down and eyes up means you're at the right level and can proceed. Pens down and eyes away means you've lost them. Win them back with a question, because more evidence won't do it. Ask which of two directions they want to go, and their answer tells you where you are. Then there's cross-talk. When two people at the same end of the table start a side conversation, either they've found something wrong or they're solving a different problem from the one you're presenting.

Your most useful instrument in the room is the interrupting question, because it tells you which layer they're on. A question about your method means you haven't earned the number yet, so drop into the working. A question about the implication means you have earned it. Stop justifying and move up. A question about something you haven't covered means your order is wrong for this room. Jump to that slide. Don't tell them you'll get to it.

Depth has three levels, and for every claim you should know what each one sounds like. At the top is the headline, one sentence with a number. Under that comes the mechanism, three or four sentences on why the number is what it is. At the bottom is the working, the slide behind the slide, with the data and the assumption. Rehearse the headline for every claim, and the mechanism for the three that carry the argument. Keep the working in the appendix, and read it out only on demand.

One habit stops depth-matching turning into drift. Whatever level you drop to, come back up to the claim before you move on. If you don't, a fifteen-minute talk becomes a string of small technical detours with no argument left standing.

8. Time: the cut you plan in advance

Overrunning is the most common delivery failure and the most avoidable, because you can make the cut on paper days earlier, while you're calm. Leave it to the room and you'll be making it while you're not.

Start from the slot and work backwards. Decide the Q&A share first. The questions are where the assessment happens, and a window squeezed to nothing costs you more than the three slides you protected.

Hold the thirty-minute row in your head. It's the commonest slot, and the arithmetic is unforgiving. Twenty presenting minutes at 90 seconds a slide is 13.33 slides, so you build 13 and hold 30 seconds of slack. A 40-slide deck at the same 90 seconds is 60 minutes of talking, 40 minutes over the presenting window. Squeeze those 40 slides into 20 minutes and each one gets 30 seconds. Nobody reads a slide in 30 seconds, and you can't explain it in that time either.

Treat ninety seconds a slide as a working default you can break. A slide carrying one number and one sentence can go in 30 seconds, and a chart that carries the argument deserves two minutes. The default gives you a build budget before you've built anything, which is when the decision is cheap.

You make the cut itself in four steps. Start by marking every slide load-bearing or not. A slide is load-bearing when its title states a claim the argument needs. If the title stack still reads as an argument without it, the slide is support. Then take the rehearsed runtime and subtract the slot. Add back a reserve of about 45 seconds for the interruption you'll get. Finally, remove support slides in descending order of their timed length until you're under target.

Run it on the twenty-two slide deck from the third worked example. You rehearsed 1,517 seconds against an 1,080 second slot, and you want a 45 second reserve, so 482 seconds have to go. Removing the seven longest support slides recovers 494 seconds and leaves 1,023 seconds. That's 57 seconds inside the slot, with every one of the 13 load-bearing slides intact. You've cut 31.8 percent of the slides and 32.6 percent of the runtime, and lost none of the argument.

Don't delete the cut slides. Move them to an appendix, in order, with their titles intact. On your own copy, note which appendix slide answers which anticipated question.

An appendix slide produced within four seconds of a question is worth more than the same slide shown to everybody for 90 seconds.

The other half of time management is what you do when you're already running late in the room. One rule survives contact with a room, and it's that you never speed up. Speeding up costs comprehension on every remaining slide to save a minute. Skip instead, out loud and without apology. Name what you're skipping so nobody thinks you've lost your place. "I am going to jump the sensitivity detail, which is in the pack, and go to the recommendation."

11. Engagement in a hybrid room, without gimmicks

A hybrid meeting has a structural bias that has nothing to do with anybody's manners. People in the room have eye contact, side channels, whiteboards and body language. People on the line have a small rectangle and a queue. Left alone, in-room attendees get several times the airtime per head that remote attendees get, and the remote half stops contributing well before they stop paying.

The fix is mechanical, and it takes about two minutes of meeting time.

Name people. Don't throw the question at the room. "Any thoughts?" is answered by whoever is physically present. "Priya, you have run this in Pune, what breaks first?" is answered by Priya.

Schedule the check-ins, and don't try to feel for the moment. A named check-in every twelve minutes in a 45-minute meeting gives you three, at minutes 12, 24 and 36. At about 40 seconds each they cost 120 seconds, which is 4.4 percent of the meeting, and no other fix here buys as much for as little.

Give the remote side the first question on every topic, because going second in a hybrid room usually means not going at all.

One camera on the room is worse than a camera on a face, and worse still is a camera on the far end of a long table. If the room has one camera, put the presenter in it, and let the in-room discussion come through the microphone.

Run the chat as a queue with an owner. Give somebody in the room the job of reading the chat aloud and holding the order of hands. That person is worth more than any engagement feature, and it means a remote participant doesn't have to interrupt a room they can't see.

Skip the gimmick tier: polls with no consequence, icebreakers that eat the time you needed for the recommendation, and asking everybody to turn cameras on as a matter of principle. You buy attention with relevance. A remote attendee whose question shaped the outcome doesn't need to be told to engage.

12. The deck is a prop, not a teleprompter

If your slides carry your script, three things follow, and all three are bad. Your audience reads faster than you speak, so they finish the slide and stop listening. Your eyes go to the screen and leave the room. And any interruption drops you into a text you now have to find your place in.

A presented slide carries the claim in its title and the evidence in one exhibit. The words you say aren't on it. That division of labour is what the split between a reading deck and a presented deck is for. For delivery, it means the presented version wants far fewer words on the page and far better notes behind it.

Treating the deck as a prop comes down to three habits.

Keep your eyes on the audience. If you need the slide, glance at your own laptop. Don't turn to the wall, because a presenter's back is a presenter who has stopped presenting.

Learn the key that blanks the screen, and use it whenever the conversation becomes a conversation. A dark screen returns every eye in the room to the discussion, and during a decision that's what you want. On a remote call the same move is stopping the share, and the read-out plan in the second worked example schedules two of them.

Make the point first, then advance the slide.

A slide that arrives ahead of its sentence competes with the sentence.

13. The debrief: what to log after every delivery

Delivery improves through a loop, and the loop needs data. Three lines written within an hour of walking out are enough. Don't leave it longer than the hour, because the specifics decay fast.

Line one: planned minutes against actual minutes. Line two: every question you were asked, with a mark against the ones you had anticipated. Line three: the one thing you'd change, stated as a behaviour you can act on. A feeling doesn't count.

Two numbers come out of that log, and they're the two worth tracking. One is overrun against plan, which tells you whether your rehearsal is realistic. The other is the anticipation hit rate, the share of questions asked that you'd written down in advance. That one tells you whether you understand your audience.

Take the eight-delivery log in the sixth worked example. Mean overrun falls from 3.575 minutes over the first four deliveries to 0.650 over the last four, a fall of 2.925 minutes or 81.8 percent. Over the same stretch, the anticipation hit rate rises from 46.4 percent to 86.0 percent, a gain of 39.6 points. The two move together, with a correlation of -0.96 across the eight. The mechanism isn't mysterious. An anticipated question gets a prepared 45-second answer, and an unanticipated one gets an improvised three-minute one.

After about six entries the log has a second use. Sort the questions by how often they recur, and it tells you what to build into the deck. In that log, "how did you get that number" is 25.0 percent of all questions asked, and the top three types are 60.7 percent between them.

A recurring question is a missing slide, or more often a missing sentence on a slide you already have.

14. Where an assistant helps, and where it will hurt you

An assistant is useful here in two narrow ways. It can generate the questions you haven't thought of, and it can analyse a recording you've already made. The dangerous uses are anything that writes your words for you, because you can't defend a delivery in somebody else's sentences once the questioning starts.

{
  "skill": "Anticipating the questions a specific audience will actually ask, before the room does",
  "use": "Paste your title stack and your key numbers, name the audience by role and by what they are deciding, and ask for the fifteen questions that audience is most likely to ask, ranked, each classified as answer, bridge, park, defer or commit, with the specific number or slide each one needs. Use the output as a candidate list to triage, never as an answer script.",
  "tools": ["Claude", "ChatGPT", "Gemini", "any general assistant with a long context window"],
  "prompt": "Here is the title stack of a 13-slide deck and the eight numbers it rests on. The audience is a four-person board of an Indian D2C personal-care company deciding whether to approve a Rs 4.2 crore marketing overspend. For each director role (chair, founder, investor director, independent director), list the five questions they are most likely to ask, ranked by likelihood. For each question give: the class (answer / bridge / park / defer / commit), the exact number or slide it needs, and a 40-word draft response. Flag any question I cannot answer from the material given.",
  "verify": "Delete every question the material cannot support, then check each surviving draft response against your own source: the number must appear in your workbook at the precision you quote it, and the slide reference must exist. Time each answer out loud with a stopwatch; anything over 90 seconds is rewritten or reclassified. Any question the assistant flags as unanswerable becomes a commit with a date, never an estimate. At least two questions must come from you rather than the tool, because the assistant only knows what you pasted and cannot know that the investor director sat on the board of the competitor for six years.",
  "diy": "You must be able to produce the top five questions unaided, because the classification decision happens in one second of live silence and there is no tool in the room. The gate tests this directly: the mastery items ask you to classify and time answers with no list in front of you, and the mini-project scores your anticipation hit rate against the questions you actually got.",
  "market": "IN"
}

The second application is very good, and almost nobody uses it. Record the rehearsal and transcribe it. Then measure the transcript instead of trusting your memory of how it went.

{
  "skill": "Turning a rehearsal recording into delivery metrics you can act on",
  "use": "Transcribe your own rehearsal audio, then ask for words per minute by segment, the location and length of every pause, the words before your first fact, filler-word counts by type, and the three longest sentences. Compare against your own targets rather than a generic ideal.",
  "tools": ["a local or hosted speech-to-text model", "Claude", "ChatGPT"],
  "prompt": "This is a transcript of a rehearsal with timestamps. Report: (1) words per minute overall and for each 60-second window; (2) every pause longer than 1.5 seconds, with its timestamp and the sentence before it; (3) the number of words spoken before the first numeric fact; (4) counts of 'um', 'sort of', 'kind of', 'basically', 'obviously'; (5) the three longest sentences by word count. Report numbers only, with no advice.",
  "verify": "Check the transcript against the audio for the numbers you spoke, because speech-to-text mangles figures and currency units more than words; a transcript that says 'four point two crore' where you said 'four point two lakh' will send you after the wrong problem. Recompute the words-per-minute figure by hand on one 60-second window before trusting the rest. The pause list is only useful next to your own plan: a pause the tool finds where you planned one is a success, not a defect.",
  "diy": "You need to be able to time a talk with a stopwatch and a slide list, because that is what you will have the night before a board meeting. The rehearsal arithmetic in the practice set is done by hand for exactly this reason.",
  "market": "US"
}

Common mistakes & how experts think differently

1. Timing the talk by reading it silently. The desk read on the worked deck came in at 913 seconds and the stopwatch pass at 1,517, so the silent number understated reality by 39.8 percent. It's tempting because it's quick, and because it gives the answer you want. An experienced presenter never quotes a runtime that wasn't measured standing up and out loud. Any timing done seated and silent tells them about completeness, and they don't trust it for length.

1. Cutting in the room instead of on paper. Realise at minute 14 that you have 11 slides left, and you'll reach for one of the two worst responses available. You'll either speed up, which costs comprehension everywhere, or skip at random, which usually drops a slide the argument needs because panic doesn't read title stacks. The cut is a calm, offline decision made from a timed slide list, and the reserve of about 45 seconds is part of it.

1. Treating Q&A as the part after the presentation. It's the part where the assessment happens, and the only part where the audience sets the agenda. Presenters who prepare for it write the fifteen most likely questions and classify each one. Then they rehearse the five hardest out loud. In the eight-delivery log, the anticipation hit rate rose from 46.4 to 86.0 percent purely from writing the list, and the overrun fell 81.8 percent alongside it.

1. Answering the challenger instead of the challenge. Match a hostile tone and you've turned a technical disagreement into a personal one in front of an audience. They'll remember the temperature long after the content. The expert move is a two-second pause, a neutral restatement, the substance in three parts, and a full stop. Concede whatever is right. Agreeing costs nothing and buys the room.

1. Guessing a number instead of committing to send it. A guessed figure enters the minutes and becomes yours. The structure that costs nothing: state that you don't have it, give the adjacent number you do have, and commit to what you'll send, to whom, by when. An unbounded "I will look into it" is worse than the admission it's trying to soften.

1. Buying a camera before fixing camera height and light. A laptop on the desk is a 20.0 degree downward gaze, and a 14 centimetre riser removes 68.8 percent of it for nothing. A window behind you is a back-to-front light ratio of 3.59, and turning the desk inverts it to 0.28. Both fixes are free, and better hardware improves neither. Money does help with audio, which drives perceived competence more than resolution does.

1. Sharing a screen without silencing it. On an ordinary 47-message day, a 45-minute share has a 98.8 percent chance of showing at least one message banner, and an 81.6 percent chance of three or more. Presenters who've been burned once share only a single window. They run do-not-disturb at both the operating system and the application level, and they take a screenshot of their own share before the call.

1. Letting a hybrid room run itself. In-room attendees will take several times the airtime per head that remote attendees do, and no amount of goodwill changes it. The counter is mechanical: named questions, scheduled check-ins every twelve minutes, the remote side first on every topic, and a chat owner in the room. Three check-ins in a 45-minute meeting cost 120 seconds, or 4.4 percent of the time.

1. Reading the slides. If a slide is dense enough to read from, the audience will finish it before you do, and it's also the slide that traps you when somebody interrupts. A presented deck carries a claim in the title and one exhibit. The sentences belong in your notes and in your head.

1. Skipping the debrief. Without a log, every delivery is the first delivery. Three lines within the hour are enough to produce the two tracking numbers that show whether the loop is closing. Write planned against actual, questions asked against questions anticipated, and one behaviour to change.

Worked examples

Six deliveries, three Indian and three American, each one carrying the arithmetic that decides it. Every figure in them is worked out in full, and each example ends with the checklist you'd actually run.

Worked example 1: Six board questions, classified and timed (India, ₹)

Nirvaan Naturals (synthetic) is a direct-to-consumer personal-care brand. The quarterly board pack gives you a 25-minute agenda item: 14 minutes of presenting and a Q&A window of 11 minutes, which is 660 seconds. Gross margin has fallen from 61.80 percent to 58.00 percent. That's a fall of 3.80 percentage points, or 380 basis points, made up of 210 basis points of input cost, 120 of discounting and 50 of channel mix. Marketing spend ran ₹4.2 crore over budget. Acquisition cost runs ₹1,240 in tier 1 cities against ₹1,000 in tier 2, which is 24.0 percent higher. You've written down the six questions you expect.

Solution. Step 1, classify each question and plan its handling time. Classification is the whole exercise. Put a question in the wrong class and it eats somebody else's minutes.

Step 2, total the plan. Planned handling is 375 seconds, or 6 minutes and 15 seconds, against a 660 second window. That leaves 285 seconds of slack, 4 minutes and 45 seconds. At a mean planned handling time of 62.5 seconds, the slack buys 4 more questions, so the plan supports 10 questions in a window sized for 6.

Step 3, price the two errors. On the day, Q3 ran 165 seconds instead of 30, and Q4 ran 240 instead of 40. Delivered handling was 710 seconds against a 660 second window, an overrun of 50 seconds. Q3 and Q4 cost 135 and 200 seconds, so 335 seconds together. That's 5 minutes and 35 seconds, and 47.2 percent of all the Q&A time spent.

Step 4, see what the errors actually were. Q3 got a full answer when it needed a defer. The city-tier acquisition data existed, the temptation to show it was strong, and showing it took the room through four tiers nobody had asked about. Q4 got a full answer when it needed a park, and that turned a strategic question into an unstructured fifteen-question debate inside a Q&A window.

The wrong version. Handled as delivered, the window overran by 50 seconds, and two directors who had questions never got to ask them. The chair closed the item. You made the commitment on Q6 in a rush and without a date. It didn't appear in the minutes.

The corrected version. Q3 in the words the class requires: "I have that by city tier, and the short version is that tier 1 runs 24.0 percent above tier 2. Rather than take the room through four tiers, I will send the table this afternoon." That's 35 words, 16.2 seconds of speech, or 18.2 seconds with a pause after the number, against 30 seconds planned. Q4: "Whether we should be in quick commerce at all is the right question and it is bigger than this item. Can I bring it to the next meeting with the channel economics?" Thirty-two words, 14.8 seconds, 16.8 with the pause, against 40 planned, and the chair's assent is on the record. Now the window closes at 375 seconds with 285 seconds spare, and four more questions get asked.

The chart behind Q1. The margin question deserves one exhibit, and choosing it is where most decks go wrong. The version that fails puts gross margin as a percentage on a left axis and revenue in ₹ crore on a right axis, across eight quarters. That invites the room to read a relationship the chart can't support, and you end up arguing about the axes when the meeting should be about the margin. The version that works is a single sorted bar of the three drivers, 210, 120 and 50 basis points, summing to the 380 the title states: "Input cost is 210 of the 380 basis points, 55.3 percent of the fall." It's one message in one form, sorted by value, with a title that states the claim.

Read it. The 47.2 percent figure is the point of this example. Nearly half the Q&A time went to two questions that should have taken 70 seconds between them, and the directors who never spoke paid for it. The number doesn't mean Q3 and Q4 were bad questions. Q4 in particular is the most important question in the room. Parking it is how you give it the hour it deserves, instead of the four minutes it got.

The checklist. Write the fifteen most likely questions. Classify each one. Write the first sentence of each answer. Time the five hardest out loud. Check that planned handling is under half the window. Confirm that no planned answer exceeds 120 seconds. Have a date and a recipient ready for every commit.

Worked example 2: A remote client read-out, planned minute by minute (US, $)

You have a 50-minute video call with the executive team of a US hospital group. The finding concerns operating-room throughput across four service lines. You're presenting alone to five people in three locations, on a platform you've used twice.

Solution. Step 1, block the call before building anything.

The blocks sum to 50 minutes on the nose. Screen-shared time is 26 minutes, 52.0 percent of the call, and faces-on time is 24 minutes, 48.0 percent. Questions take 16 minutes, or 32.0 percent.

Step 2, size the deck from the shared minutes. Twenty-six shared minutes at 90 seconds a slide supports 17 slides. At 60 seconds a slide it would support 26, and that's the trap. A deck sized at 60 seconds a slide has no room for the sentence that explains the slide.

Step 3, place the two stop-share moments. The first is a click. The second is a refusal to click, and it's the harder of the two. At minute 24 the share stops, and the recommendation runs to minute 29 with faces on. At minute 29 the question block opens, and the deck stays down for all sixteen minutes. Most presenters lose it here, by bringing the deck back up behind the questions. Across those two blocks, twenty-one of the call's minutes are faces-on, and that's where the client decides. Deliver a recommendation over a slide and the client reads it instead of hearing it.

Step 4, plan the failure. Assume the connection drops at minute 12, and recovering takes 3 minutes while you dial in on a phone and re-share. That leaves 35 wall-clock minutes against 38 minutes of unspent plan, a deficit of 3 minutes. You've decided the cut in advance. It's finding 3, worth 5 minutes, which leaves 33 minutes of plan and 2 minutes of slack. Recovery costs 6.0 percent of the call, and the plan absorbs it without touching the recommendation or the question block.

Step 5, build the memorable chart. Contribution margin per case, four service lines: orthopaedics $4,820, cardiology $3,960, general surgery $1,740, bariatrics $910. They total $11,430. Orthopaedics is 42.2 percent of that total and 5.30 times bariatrics, and the top two lines are 76.8 percent between them.

The wrong version. A bar chart of contribution margin per case, sorted by value, titled "Contribution Margin by Service Line". It's the right chart form with the wrong measure, and the title states the topic and leaves out the finding. Read alone, it says cardiology is the second most valuable line. But cases aren't what the client is short of.

The corrected version. In an operating-room business, the constraint is theatre time. Cases consume 1.9, 4.2, 1.5 and 2.6 hours respectively, so contribution margin per theatre hour is $2,536.84, $942.86, $1,160.00 and $350.00. Sort by value and the ranking changes. Cardiology falls from second to third, behind general surgery, because it consumes 2.8 times the theatre hours per case that general surgery does. The action title states the finding: "Per theatre hour, general surgery beats cardiology by 23 percent, which reverses the case-level ranking."

Read it. The two charts use the same four numbers and support opposite decisions. So the denominator is a delivery question as well as an analysis question. Say the denominator out loud when you show it, and pause. The second chart doesn't say cardiology should shrink. It says an hour of theatre time given to cardiology earns less than an hour given to general surgery. The strategic, clinical and referral consequences of acting on that are outside the exhibit.

The checklist. Block the call to the minute before building. Decide the Q&A share first. Size the deck from shared minutes only. Place the stop-share moments where decisions happen. Pre-plan the cut for a connection failure. State the denominator on every per-unit chart. Give every chart an action title that could be contradicted.

Worked example 3: The rehearsal, and the cut made on paper (India, ₹)

Meghdoot Foods (synthetic) is a Pune frozen-foods business presenting a margin review to its lender's credit committee. The slot is 18 minutes, or 1,080 seconds, and the deck is 22 slides. You've marked 13 of them load-bearing, meaning their titles carry claims the argument needs.

Solution. Step 1, time it properly. Four passes: a silent desk read at 913 seconds, a standing out-loud pass at 1,442, a stopwatch pass with notes closed at 1,517, and a post-cut pass at 1,068. The stopwatch pass is the planning number. It runs 1.66 times the silent read, which understated the true runtime by 604 seconds, or 39.8 percent.

Step 2, size the problem. 1,517 seconds against 1,080 is an overrun of 437 seconds, 7 minutes and 17 seconds, which is 28.8 percent of the rehearsed runtime. Mean time per slide is 68.95 seconds, so the slot supports about 15.7 slides at the pace you actually speak.

Step 3, set the target with a reserve. Hold back 45 seconds for the interruption you'll get. The target is 1,035 seconds, so 482 seconds have to go.

Step 4, cut support slides only, longest first.

That's seven slides and 494 seconds recovered. Runtime falls to 1,023 seconds, 17 minutes and 3 seconds, which is 57 seconds inside the slot. You've cut 31.8 percent of the slides and 32.6 percent of the runtime, and all 13 load-bearing slides survive. The new mean is 68.2 seconds a slide, which is almost unchanged. You removed slides and didn't speed up.

Step 5, rehearse the cut version cold. Pass four opens cold on slide 1, with a colleague primed to interrupt. Runtime is 1,023 seconds of slides plus a 45-second planted question, so 1,068 seconds, leaving 12 seconds of margin against the slot. All four passes cost 4,940 seconds, 82 minutes and 20 seconds, which is 4.63 times the length of the delivered talk.

The wrong version. Walk in with 22 slides and manage it live. At minute 14 you notice eleven slides remain. You speed up, and the last six slides get 30 seconds each. The credit committee's final impression of your margin review is a blur. The two slides you needed them to remember arrived after they'd stopped listening.

The corrected version. Fifteen slides, timed at 17:03, with the other seven moved to an appendix in order and each tagged on your own copy with the question it answers. When a committee member asks about competitor pricing, the scan is on screen in four seconds, and it's worth more there than it was as slide 12.

Read it. The cut is 32.6 percent of the runtime and none of the argument. That's only possible because you wrote the title stack first, and it told you which slides carried the argument. The arithmetic can't tell you whether the 13 surviving slides are the right 13. It only guarantees that the cut didn't damage the argument you already had.

The checklist. Time it standing, out loud, with a stopwatch. Use that number and no other. Mark load-bearing slides by whether their titles carry the argument. Subtract the slot, add a 45-second reserve, and cut the longest support slides first. Move the cut slides to an appendix and index them by question. Rehearse the cut version cold, with an interruption.

Worked example 4: The first ninety seconds, three drafts (US, $)

Cascade Distribution (synthetic) runs 34 sites in the United States. Nine of them lose money at the contribution line, which is 26.5 percent of the network. Those nine cost $11.4 million a year, $1,266,667 each, or $24,359 a week per site. You've got 90 seconds to open a review with the operating committee. At 130 words a minute with five two-second pauses, your budget is 173 words.

Solution. Step 1, write all three drafts and count them.

Step 2, notice what the table does not separate. All three fit inside 90 seconds, and all three are under the 173-word budget. So runtime doesn't separate them. A presenter checking only the clock would call all three acceptable.

Step 3, measure the thing that matters. Draft A spends 44.3 seconds before the audience learns anything, and draft C spends 1.8. Those are 42.5 seconds of attention spent on throat-clearing, and they're spent at the very moment the room is deciding how much of itself to lend you.

Step 4, write C in full and time it with the pauses. This is our own draft, written for the example, with the five planned pauses marked:

Nine of our thirty-four sites lose money before we pay a single dollar of head-office cost. [two-second pause] That is 26.5 percent of the network, and it is not the nine sites you would guess from looking at the revenue table. [two-second pause] Together they cost us 11.4 million dollars a year at the contribution line, which works out at 1.27 million dollars a site, or 24,359 dollars every single week for each of the nine. [two-second pause] Seven of those nine share one cause, and the cause is route density rather than labour rates, which is the opposite of what last year's own review concluded. [two-second pause] So I am going to take you through the density arithmetic on those seven sites, and then ask you to approve a route consolidation at four of them, starting in October. [two-second pause] I will take fourteen minutes, and the rest of the pack is background that I will not present.

At 149 words plus five two-second pauses, that runs 78.8 seconds and leaves 11.2 seconds under the ninety-second ceiling. The fact is complete at word 4, the mechanism arrives at word 73, the ask at word 101 and the time commitment at word 132.

The wrong version. Draft A, which is what most people write, because gratitude and context feel like manners. "Thanks everyone for making time, I know this got moved twice. Before I start I should say the Q3 data is still being confirmed by finance, so treat the last column as indicative. I have put an agenda on the second slide. There is a lot here and I have tried to keep it tight." That's ninety-six words before a single fact, and they include a caveat that discounts everything the audience is about to hear.

The corrected version. Draft C, in the order above. The caveat about Q3 doesn't disappear. It moves to the slide where the affected number appears, and that's where someone can actually act on it.

Read it. The 42.5-second gap measures attention spent at the highest price of the meeting, so it beats any argument about style. The numbers don't say opening C is well written. They say it's correctly ordered. That's a different and more teachable property, and ordering is what survives being nervous.

The checklist. Write the opening out in full. Count the words against your own rate and your own pause plan. Mark the word at which the first fact arrives, and get it under ten. Delete every apology. Keep the self-introduction to one sentence. State the ask and the duration before you start the evidence.

Worked example 5: A hybrid room, audited (India, ₹)

Tamarind Retail (synthetic) runs a monthly operations review from its Mumbai office. Four people sit in the room and seven dial in from city teams, so the meeting has 11 heads, of which 63.6 percent are remote. The meeting runs 45 minutes. You time who speaks.

Solution. Step 1, measure the baseline. Participant airtime, excluding the presenter, is 22.5 minutes: 18.4 minutes from the four in the room and 4.1 minutes from the seven on the line. In-room attendees take 81.8 percent of participant airtime while holding 36.4 percent of the heads.

Step 2, put it per head, which is the fair comparison. 18.4 minutes across four people is 4.60 minutes each. 4.1 minutes across seven is 0.586 minutes each, about 35 seconds. The ratio is 7.85 to 1. Two thirds of the meeting's people are getting an eighth of the airtime per head.

Step 3, apply the four mechanical fixes. Named questions instead of open ones. Scheduled check-ins at minutes 12, 24 and 36, at about 40 seconds each, costing 120 seconds or 4.4 percent of the meeting. Remote side first on every topic. A chat owner in the room reading the queue aloud.

Step 4, measure again. Participant airtime is still 22.5 minutes, and the presenter still speaks for 22.5 minutes, so the fixes took nothing from the presentation. The split is now 12.9 in the room and 9.6 on the line. That's 3.225 minutes per in-room head and 1.371 per remote head, a ratio of 2.35 to 1. Remote airtime rose 5.5 minutes, 134.1 percent more than before, and the per-head imbalance fell by a factor of 3.34.

The wrong version. The chair opens the floor with "any thoughts on that?" and takes the first hand. Every time, the first hand is in the room. The remote team stops trying at about minute 20, and the meeting reaches a decision that four people shaped and seven will implement.

The corrected version. "Priya, you have run this in Pune, what breaks first?" then two more named remote questions before anybody in the room speaks. The chat owner reads the queue in order. Three scheduled check-ins cost two minutes of a 45-minute meeting.

Read it. A ratio of 2.35 isn't parity, and it isn't meant to be, because people in the room will always have more channels available to them. The audit shows that a factor of 7.85 was a design failure and a factor of 2.35 is a normal meeting. The numbers can't show whether the remote contributions were better used. That's a judgement the minutes have to carry.

The checklist. Count heads in and out of the room before you start. Name the person you're asking. Put a check-in every twelve minutes into the agenda itself, because intentions don't count. Give the remote side the first question on each topic. Appoint a chat owner in the room. Measure airtime per head at least once a quarter.

Worked example 6: Eight deliveries, logged (US, $)

An analyst logs every delivery for two quarters: planned minutes against actual, questions asked against questions anticipated in advance.

Solution. Step 1, the log.

Step 2, split the log in half. Mean overrun falls from 3.575 minutes across deliveries 1 to 4 to 0.650 across 5 to 8, a fall of 2.925 minutes or 81.8 percent. As a share of the planned slot, the overrun falls from 18.3 percent to 3.5 percent. Delivery 8 is the first that came in under plan.

Step 3, look at the other column. The anticipation hit rate rises from a mean of 46.4 percent to 86.0 percent, a gain of 39.6 points. Across all eight deliveries, 37 of 56 questions were anticipated, a hit rate of 66.1 percent. That's 7 questions a delivery on average, with a range of 6 to 8.

Step 4, connect the two. The correlation between overrun and hit rate across the eight deliveries is -0.96. There's nothing statistical about the mechanism. An anticipated question has a prepared 45-second answer, and an unanticipated one gets improvised at three minutes.

Step 5, sort the questions and build the exhibit. Across the 56 questions, seven types recur.

The wrong version. A pie chart of seven question types. That's seven slices, four of them under 13 percent, with a legend the reader has to match to colours. There's no way to compare the third and fourth without measuring angles. Beyond three parts, a pie stops being a chart and becomes a puzzle.

The corrected version. A horizontal bar chart, one bar per type, sorted by value, with the counts labelled and a neutral single colour except for the top bar. The title states the finding: "One question in four is 'how did you get that number', so the method belongs on the slide." It's sorted by value because the order isn't intrinsic. It uses one colour because the bar length already carries the comparison. And it has an action title because the chart exists to change what you build next.

Read it. The two tracking numbers are overrun and hit rate, and the second causes the first. The log can't tell you whether the analysis got better, because a weak argument delivered on time is still a weak argument. It tells you the delivery loop is closing, and that narrower claim is the one it's entitled to make.

The checklist. Log within an hour, while the questions are still exact. Record planned against actual to one decimal. Mark each question anticipated or not before you rationalise. Write one behaviour to change, never a feeling. Sort the recurring questions by frequency after six entries. Turn any question above 20 percent into a slide or a sentence.

Verify

Every figure printed above, at full precision, straight from the underlying calculation.

Practice set

Twelve problems, alternating between Indian and American settings. Work each one before reading its solution. The arithmetic is short enough that peeking costs you the whole benefit. You'll need a calculator. Numeric answers within 2 percent count as correct, except where a problem asks for a whole number of slides or questions. There, the floor is the answer. The first three carry more scaffolding than the rest.

P1 (guided). Size a deck from a slot (US). You have a 40-minute slot and you want 30 percent of it for questions. You speak at a pace that has averaged 95 seconds a slide across your last three decks. How many slides do you build, and what would a 31-slide deck force?

Solution. Presenting time is 40 × 0.70 = 28.0000 minutes and questions get 40 × 0.30 = 12.0000 minutes. At 95 seconds a slide, 28 × 60 / 95 = 17.6842 slides, so you build 17 and hold the remainder as slack. A 31-slide deck forces 28 × 60 / 31 = 54.1935 seconds a slide, which is 43 percent below your own measured pace, so either 14 slides get cut in the room or the question window disappears. Notice which number came from measurement. The 95 seconds is yours. It isn't a rule, and using somebody else's default here would move the answer by several slides.

P2 (guided). Budget an opening (India). You are opening a 75-second slot with four planned pauses of 2.5 seconds. Your measured rate is 135 words a minute. What is the word budget, and what does a 190-word draft actually cost?

Solution. Pauses take 4 × 2.5 = 10.0 seconds, leaving 65 seconds of speech. The budget is 65 / 60 × 135 = 146.2500 words. A 190-word draft is over by 190 − 146.25 = 43.7500 words, which is 29.9 percent too long, and with the pauses it runs 190 / 135 × 60 + 10 = 94.4444 seconds against a 75-second slot. The fix is 44 words of cutting, and the words to cut are the ones before the first fact.

P3 (guided). Plan a Q&A window (India). You expect seven questions and have planned handling times of 100, 45, 25, 35, 85, 40 and 50 seconds, classified as answer, bridge, defer, park, answer, commit and bridge. The window is 8 minutes. Is the plan safe, and how much room is left?

Solution. Planned handling totals 100+45+25+35+85+40+50 = 380 seconds, 6 minutes and 20 seconds, against a window of 8 × 60 = 480 seconds. The slack is 100 seconds. The mean handling time is 380/7 = 54.2857 seconds, so the slack buys 100/54.2857 = 1.8421 further questions, which is 1 whole one. Full answers take (100+85)/380 = 48.68 percent of the planned time, which is a healthy mix. If that share went above about 70 percent, you'd be planning a lecture. The plan is safe with one spare question, and no single answer is anywhere near the 120-second ceiling.

P4. Cut a deck to fit (US). Ten slides, rehearsed at 40, 95, 118, 76, 132, 88, 61, 104, 57 and 49 seconds. Slides 3, 5, 7 and 9 are support; the rest are load-bearing. The slot is 11 minutes and you want a 45-second reserve. Make the cut.

Solution. Rehearsed runtime is 820 seconds, 13 minutes and 40 seconds, against a 660-second slot, so you're 160 seconds over. With a 45-second reserve the target is 615 seconds, so 205 seconds have to go. Cut support slides in descending order of length. Slide 5 goes first at 132 seconds, which leaves 73 still to find, then slide 3 at 118. That recovers 250 seconds in total. Runtime falls to 570 seconds, 9 minutes and 30 seconds, which is 90 seconds inside the slot and twice the reserve. Eight slides remain, and every load-bearing slide survives. You've overshot the target by 45 seconds, and that's fine. It's the price of cutting whole slides instead of trimming each one, and the spare seconds go to the questions.

P5. Price the notification risk (India). Your working day carries 62 messages across 9 hours. You are about to share your screen for 25 minutes with do-not-disturb off. What is the expected number of banners and the probability that at least one appears?

Solution. The arrival rate is 62/9 = 6.8889 an hour. Over 25 minutes the expected count is 6.8889 × 25/60 = 2.8704. Treating arrivals as independent, the probability of at least one is 1 − e^(−2.8704) = 0.9433, or 94.3 percent, and the probability of a clean share is 5.7 percent. Four seconds of do-not-disturb moves the expected count to zero. The independence assumption is the weak point. Messages arrive in bursts, which makes the probability of at least one slightly lower and the probability of three or more considerably higher.

P6. Fix a camera (US). Seated, your eyes are 30 centimetres above the desk, the laptop camera is 10 centimetres above it, and you sit 48 centimetres from the screen. Compute the gaze angle, the angle after a 12-centimetre riser, and the riser height that removes the problem entirely.

Solution. The vertical drop is 30 − 10 = 20.0 centimetres, so the angle is arctan(20/48) = 22.6199 degrees. A 12-centimetre riser leaves a drop of 8 centimetres and an angle of arctan(8/48) = 9.4623 degrees, removing 22.6199 − 9.4623 = 13.1575 degrees. A 20-centimetre riser puts the camera right at eye level, at zero degrees. The fix is geometric. No camera at any price changes the angle you're looking from.

P7. Audit a hybrid meeting (India). Five people in the room, nine dialled in. Over the meeting, in-room participants speak for 21.5 minutes and remote participants for 6.3. Compute the per-head figures and state the finding in one sentence.

Solution. Per in-room head, 21.5/5 = 4.3000 minutes. Per remote head, 6.3/9 = 0.7000 minutes. The ratio is 4.3/0.7 = 6.1429 to one. Remote participants hold 9/14 = 64.29 percent of the heads and take 6.3/27.8 = 22.66 percent of participant airtime, an airtime share 41.62 percentage points below their headcount share. The finding: two thirds of the people in the meeting are getting a sixth of the airtime per head, so the decision this meeting reaches was shaped by five of the fourteen people who will have to live with it.

P8. Check a read-out plan (US). A 45-minute client call is blocked as: open 2, context 5, finding 1 six, finding 2 six, finding 3 six, recommendation 4, questions 12, close 4. Does it fit, what share is questions, and what do you cut if the client joins ten minutes late?

Solution. The blocks total 45 minutes on the dot, so it fits with zero slack. That's the first finding in itself, because a plan with no slack has already spent the buffer you'll need. Questions are 12/45 = 26.67 percent of the call. If the client joins ten minutes late, 35 minutes of clock remain against 45 minutes of plan, a deficit of 10. One cut won't close it. Dropping finding 3, worth six minutes, brings the plan to 39, which is still 4 minutes long. Drop the five-minute context block as well and it comes to 34 against 35, so 1 minute of slack. The 12-minute question block would close the whole deficit in one move. It's also the block you never touch, because it's the part of the call where the client decides.

P9. Read a delivery log (India). Four deliveries: planned 30 actual 34.2 with 9 questions of which 3 were anticipated; planned 30 actual 32.5 with 8 questions of which 5 were anticipated; planned 25 actual 26.1 with 7 questions of which 5 were anticipated; planned 25 actual 24.6 with 9 questions of which 8 were anticipated. What are the two tracking numbers doing?

Solution. Overruns are 4.2, 2.5, 1.1 and −0.4 minutes, a mean of 1.85. The first two average 3.35 minutes and the last two average 0.35, a fall of 3.0 minutes. Hit rates are 33.3, 62.5, 71.4 and 88.9 percent, and across all four, 21/33 = 63.64 percent of questions were anticipated. Both numbers are moving the right way, and they're moving together. That's the pattern to expect, because preparing for questions shortens the answers, and shorter answers is where the overrun goes. Four points are far too few to claim a trend. After four deliveries, all you can fairly say is that nothing yet contradicts the hypothesis.

P10. Budget the pauses on a long talk (US). A 35-minute talk carries 22 key numbers, each followed by a two-second pause. Your rate is 125 words a minute. What do the pauses cost?

Solution. Pausing costs 22 × 2 = 44 seconds, which is 44/2100 = 2.10 percent of the talk. In words, 44/60 × 125 = 91.6667 words displaced, and in slides at 90 seconds each, 44/90 = 0.4889 of one slide. Half a slide is the entire price of letting an audience absorb 22 numbers. Nobody has ever been asked to leave a room for pausing, and the arithmetic shows why. The pause budget is smaller than the rounding error in your slide count.

P11. Recover from a dropped connection (India). A 60-minute call. Your connection drops at minute 34 and takes 4 minutes to recover. Your plan has 26 minutes of content left. What is the deficit, and what does one pre-planned five-minute cut leave you?

Solution. Wall-clock time left after recovery is 60 − 34 − 4 = 22 minutes against 26 minutes of unspent plan, a deficit of 4 minutes. Dropping one pre-planned five-minute block leaves 21 minutes of plan against 22 of clock, so 1 minute of slack. The decision that matters is which block. You choose the cut before the call, and it's never the recommendation, the questions or the next steps, because those three are why the client is on the line.

P12. Choose the exhibit under time pressure (US). You have nine site-level series to show across twelve months, ranging from −4.2 to +11.8 points. A colleague has put all nine on one line chart. State the problem numerically and give the alternative.

Solution. Nine series on one set of axes gives the reader 9 × 8 / 2 = 36 pairs to track, and every crossing is a place where two lines have to be told apart by colour alone. A three-by-three grid of small multiples gives 9 panels, 1 series each and 0 crossings within a panel, on a shared scale spanning 11.8 − (−4.2) = 16.0 points so the panels stay comparable. It follows a standing chart rule: small multiples beat a crowded single chart, and they beat a dual axis in every case. What you lose is the direct visual comparison of any two chosen sites. You buy it back by adding a faint grey line of the network median behind every panel.

Applied mini-project

Deliverable. A recorded five-minute delivery of your own analysis, plus a delivery plan, a question sheet, a self-scored rubric and a three-line debrief. Audio alone is enough. Video adds nothing to the assessment and a great deal to the reluctance.

Use a deck you already have. The storyline work produces one. If you don't have one, take any analysis you've done and give yourself this brief: five minutes, one decision-maker, one recommendation with a number attached.

Artefact 1: the delivery plan

One page. The slot in minutes, the question share you've chosen, the resulting slide budget at your own measured seconds per slide, and a block plan with clock times. If the delivery is remote, mark each block shared or unshared and name the two stop-share moments.

Test 1. The blocks add up to the slot with no remainder, and the slide count is consistent with the presenting minutes and your own measured pace. Show that as arithmetic. An assertion doesn't pass.

Artefact 2: the opening, written and timed

The first ninety seconds in full, with the pause marks in place, a word count, and the word position at which the first fact arrives.

Test 2. The word count is inside your own budget at your own measured rate, and the first fact arrives within the first ten words. Both are checkable from the page.

Artefact 3: the question sheet

Fifteen questions you expect, each classified as answer, bridge, park, defer or commit, each with a first sentence written out and a planned handling time. At least two must be questions you would rather not be asked.

Test 3. Planned handling time totals less than half your question window, and no planned answer exceeds 120 seconds. Any commit carries a named recipient and a date.

Artefact 4: the recording

Five minutes, stood up, cold, one take. Don't re-record to get a clean run. A first take with a stumble in it is the real artefact, and the stumble is data.

Test 4. The recording is a single take of at least four and a half minutes, and the timing you logged afterwards matches its actual length to within 15 seconds.

Artefact 5: the debrief

Three lines: planned against actual, questions anticipated against questions asked, and one behaviour to change stated as an action. The questions can come from a colleague who listened, or you can check your own list against a rehearsal partner's questions.

Test 5. The behaviour line names something observable. "Be more confident" fails; "stop after each number and count two" passes.

Rubric. Twenty points, passing at 16 with no zero on any row.

Any artefact whose test fails scores zero on its row, however good the delivery sounded. A five-minute recording with a beautiful opening and no question sheet is only a rehearsal, and the rubric is built to say so.

Budget about 45 minutes on the plan and the question sheet, 20 minutes on the opening, 30 minutes of rehearsal across three passes, 5 minutes recording, and 15 minutes on the debrief. Most of the two hours goes on the parts nobody enjoys, and that's the finding this project exists to produce.

Reading & resources

Core (do these).

1. Barbara Minto, The Pyramid Principle (1987, revised editions since). Paid, intermediate. Read the sections on ordering an argument for a listener instead of a reader. The delivery consequence of a pyramid is that you can cut from the bottom without breaking the top.
2. Cole Nussbaumer Knaflic, Storytelling with Data (Wiley, 2015). Paid, beginner. The chapters on focus and on the exhibit that carries a message are the ones a presenter uses, and her 2019 companion volume of graded practice exercises is where the reps are.
3. Chris Anderson, TED Talks: The Official TED Guide to Public Speaking (Houghton Mifflin Harcourt, 2016). Paid, beginner. The most useful chapters are on the through-line and on what to do with nerves, and the book admits that its format isn't a business meeting.
4. Jerry Weissman, Presenting to Win (FT Press, 2003, updated editions since). Paid, intermediate. The extended treatment of question handling, including the topspin idea of returning to your own message after answering, is the closest published match to the five-class method used here.
5. Matt Abrahams, Think Faster, Talk Smarter (Simon Element, 2023). Paid, beginner. Specifically for the unprepared moment: the Q&A, the corridor, the question you did not expect.

Going deeper.

1. Nancy Duarte, Resonate (Wiley, 2010). Paid, intermediate. On structuring a talk as a movement between what is and what could be. Read it for the shape of an argument delivered aloud. For visual advice, slide:ology (2008) is better.
2. Garr Reynolds, Presentation Zen (New Riders, 2008, later editions). Paid, beginner. The case for restraint on a presented slide, which is the position the deck-as-a-prop section takes.
3. Edward Tufte, The Visual Display of Quantitative Information (Graphics Press, 1983, 2nd edition 2001). Paid, advanced. For why decluttering is a data argument rather than a taste argument.
4. Stephen Few, Show Me the Numbers (Analytics Press, 2004, 2nd edition 2012). Paid, intermediate. The practical rules for chart form and table design that the exhibits here follow.
5. Alberto Cairo, How Charts Lie (W. W. Norton, 2019). Paid, beginner. Useful before you present a chart to a sceptical room, because the questions in it are the questions the sceptic will ask.

Standards, tools and free material.

1. Your own meeting platform's documentation on presenter view, screen sharing and participant layout. Free. Feature names and defaults change between releases, so read your own version rather than a tutorial written for a different one (confirm against current documentation; checked September 2026).
2. Toastmasters International club meetings. Free to attend, low cost to join. The value is the reps and the immediate feedback rather than the curriculum, and a working analyst can get most of the benefit in a dozen sessions.
3. Amy Cuddy, Presence (Little, Brown, 2015). Paid, beginner, and read with a caveat. The book's best-known claim, that adopting an expansive posture changes hormone levels and behaviour, did not survive replication: Ranehill and colleagues (2015) found no hormonal or behavioural effect in a larger sample, and Simmons and Simonsohn (2017) reported that the evidential value of the original literature was weak. The self-report effect on felt confidence has held up better than the physiological one (confirm the current state of this literature; checked September 2026).

Teach it back & journal

Feynman prompt. Explain to somebody who has never presented anything why a deck that rehearses at 25 minutes can't be delivered in 18 by talking faster, and what you do instead. Your explanation has to include three things: the arithmetic that shows the overrun, the distinction between a slide that carries the argument and a slide that supports it, and what happens to the slides you remove. Then have them ask you the one question you'd least like to be asked about your own most recent piece of work. Answer it in under ninety seconds, using the structure for a question you can't answer. If they can repeat your cut rule back to you afterwards, and if your answer to their question named a deliverable, a recipient and a date, you've got it.

A harder version, if you have somebody patient. Give them your slide list with the timings and let them make the cut. Where their cut differs from yours, the disagreement is almost always about which slides are load-bearing. That argument is more useful than the timing exercise, because it's really an argument about what your title stack claims.

Journal prompt. Write about the last time a presentation of yours went badly. Be specific about the mechanism, and leave the feeling out of it. Which of the failures was it? Too long for the slot, or an opening that spent its first forty seconds on you. A question that turned into a second presentation, or a number you guessed. A notification at the wrong moment. A room where the people who mattered were on a phone line and never spoke. Write down the number you'd have needed to know in advance to prevent it, and where you'd have found that number. Then write the three lines of the debrief you didn't write at the time.

Come back to this entry after the mini-project. The gap between what you now think went wrong and what you thought at the time is what the six hours bought you. It's usually larger than people expect. What feels like a failure of nerve is almost always a failure of arithmetic that happened three days earlier, at a desk, with a stopwatch that was never started.