Titles That Carry the Argument: Horizontal and Vertical Logic

Presentation track, branch pr-storytelling (Storytelling) · Node PR4.02 · ~10 focused hours · Mastery gate ≥ 85%

Ten slides go to a board meeting. Read the nine titles that follow the cover, in order and nothing else: Agenda. Business Overview. Revenue Performance. Channel Mix. Gross Margin. Cost Structure. Working Capital. Capex and Debt. Risks and Outlook. That is nineteen words, about five seconds at ordinary reading speed, and at the end of it a director knows that a company sold things, spent money and carries some debt. She knew that before she opened the file.

Now read the same ten slides after a rewrite that changed no chart, no table and no number. Revenue grew 21.9% to ₹412 crore. Three channels that were 23.1% of last year's sales delivered 94.6% of this year's growth. Gross margin fell 1.85 points, and a little over two thirds of that fall is channel mix rather than price. Operating cost grew slower than revenue, so the whole 1.70-point fall in EBITDA margin came from gross margin. Working capital absorbed ₹20.7 crore, of which ₹8.5 crore is a longer cash cycle rather than growth. Net debt is 3.03 times EBITDA against a covenant of 3.25, and a 6.8% fall in EBITDA breaches it. The board is asked to defer the second production line by two quarters or approve ₹25 crore of equity.

Same data. Same evening's work by the analyst who built the exhibits. The second stack runs to 114 words and takes about thirty-one seconds to read. Twenty-six seconds of a director's attention bought the entire argument, including the decision she is in the room to take, and it is the cheapest trade available anywhere in professional communication.

That difference is what a title is for, and most working decks throw it away. The title occupies the largest type on the page, sits where every reader's eye starts, and is the only text that survives being skimmed, forwarded, printed at 60% and read on a phone in a car. Filling it with a filing label like "Gross Margin" is the most expensive habit in professional communication, because it spends the one guaranteed impression on a table of contents entry.

Two properties decide whether a set of titles works, and they are independent of each other. Horizontal logic runs across the deck: the titles in order have to make one argument, with no gap the reader must bridge alone and no leap the evidence does not support. Vertical logic runs down a single page: the body has to prove that page's title, exactly, and nothing else. A deck can pass one and fail the other. Titles that read beautifully in sequence while their bodies prove different things is the commoner failure, and it is the harder one to see, because reading the stack feels like checking the work.

The vocabulary comes from Barbara Minto's The Pyramid Principle, which set out the answer-first structure and the two logics that hold a pyramid together. The pyramid itself belongs to the storyline work that precedes this: governing thought, key line, support. What follows takes that structure as given and asks a narrower question. Once you have an argument, how do you put it on the pages so that a reader who reads nothing but the titles gets it, and a reader who reads everything is never misled?

There is one thing this craft cannot do, and stating it early saves disappointment. A true title on a weak argument is still a weak argument, elegantly labelled. The tests below will catch a title that does not follow from its own body and a stack that does not hold together. They will not tell you whether the analysis underneath is right. Nothing on this page substitutes for having done the work.

Learning objectives

You can:

1. Write an action title from a stated fact set, with a subject, a finite verb and the number that makes the claim checkable, and say which of three candidate numbers belongs there.
2. Run the read-only-the-titles test on a deck, write out the argument the stack makes, and state honestly when it makes none.
3. Diagnose a break in horizontal logic as a gap, a leap or a repeat, and name the slide that repairs each.
4. Test vertical logic by asking whether a body proves its own title exactly, and identify bodies that prove less, more, or something else.
5. Repair a title-body mismatch two ways, once by fixing the title and once by fixing the body, and choose between the repairs using the deck's chain rather than the slide alone.
6. Decide the precision and the basis of a number in a title, and show how the same fact yields two defensible percentages on two bases.
7. Cut a slide whose title is not a claim, state what the cut costs, and place the material where it can still be retrieved.
8. Compute the two-line ceiling for a stated type size and slide width, and cut a title to fit without losing its number or its verb.
9. Separate an honest qualifier from a hedge, and rewrite a hedge into a claim with a named condition under which it fails.
10. Apply the machine's title rule, name the three cases where it is wrong, and say what a human overrule costs and what it buys.

Prerequisites & connections

Builds on. The pyramid work that comes immediately before this. You should arrive with a governing thought, a key line whose points don't overlap, and an SCQA opening. You should also be able to say what separates a thought from a topic. And you'll need the takeaway habit from the track's entry point: one sentence, a claim, falsifiable, implying an action. Arithmetic at the level of percentages, shares and a two-factor decomposition is assumed throughout, since almost every title worth writing carries a number.

Feeds forward. PR4.03 takes an ordered set of true titles and asks what turns it into a story with an arc, a memorable exhibit and a costed recommendation. PR5.01 to PR5.03 apply the same title rules inside three archetypes, where a consulting body page, a board page and a research thesis page each impose their own conventions on what a title may claim. PR7.01 runs the whole sequence under time pressure on unseen data, and the read-only-the-titles test is one of its scored gates. The delivery module treats the title stack as the speaker's own map, since a presenter who has to remember an order hasn't written one.

Six neighbours own material that borders this one, and the non-overlaps are deliberate. PR2.03 owns the action title of a single chart, together with the focal point, small multiples and the dashboard layout. The craft of making one exhibit say one thing lives there, and everything below assumes it and works at the level of the sequence. PR4.01 owns the pyramid, SCQA and SCR, the MECE test on a key line, and the executive summary as the top two levels of the pyramid. PR3.02 owns the ghost deck and the dot-dash sketch, which is where headlines are first drafted on paper. Here, the tests run on a stack that already exists. PR0.01 owns the one-sentence takeaway, the audience profile and the medium choice. CN1.03 owns answer-first logic for a consulting case and hands off to the deck craft here. M10.03 owns the written analyst memo and one-pager, where a title becomes a section heading and the rules change.

Core teaching content

1. The title is the slide

Watch how a deck actually gets read and the case for spending real effort on titles makes itself. A reader opens the file and scrolls at roughly a page a second, looking for the part that concerns them. They stop on two or three pages and read those properly. Then they forward the file to somebody else with one line of their own. Every one of those behaviours passes through titles. The scroll reads nothing else. A title decides where they stop. And the forward is summarised from titles. The person receiving it will often read only the summary.

A presented deck is consumed differently and ends in the same place. Your audience reads each slide's title before you've finished your first sentence, then spends the rest of the slide either following or waiting. What they carry out of the room, and what they repeat to a colleague two days later, is close to the title stack plus one number.

So the title is the claim the slide exists to make, and the body is the evidence for that claim. Most decks use it as a label for the contents instead. That difference gives you a test you can run in four seconds on any page. Read the title. If it's a noun phrase, the page is a container for what the author knows, and the reader has to construct the argument themselves. If it's a sentence somebody could contradict, the page is an argument, and the body's only job is to make it believable.

Look at what each version leaves the reader to supply. "Channel Mix" tells a director that channel mix will appear below. Then she has to read the chart, work out which channels moved, decide whether that movement is good or bad, and guess what the author thinks about it. Three of those four steps are the author's job, and the author has handed them over without a word. "Quick commerce tripled to 15.3% of revenue while general trade fell 4.3%" hands back the two facts the author selected, in the order the author chose. The judgement is still open for the director to dispute.

The dispute is the point. A title a reader can't disagree with is a title that has told them nothing.

2. Writing an action title: subject, verb, number

An action title has three parts and the first two are grammar. There is a subject, which is the thing the claim is about. There is a finite verb, which is what the subject did or is. And where the data supports it there is a number, which is what makes the claim checkable against the body underneath.

Take the fall in Kadamba Foods' gross margin. Four candidate titles, in ascending order of usefulness:

The fourth beats the third for one reason. A board that reads the third has to ask why, and a board that has to ask why is a board whose meeting you're about to lose control of. The fourth answers the question the third provokes. That's the vertical-logic test applied one level up.

The verb matters more than authors expect, because the verb is where the claim's strength lives. "Gross margin fell 1.85 points" asserts a fact. "Gross margin fell 1.85 points because quick commerce grew faster than the book" asserts a cause, which needs the decomposition on the page to support it. "Gross margin will fall a further 0.95 points on the current plan" asserts a forecast, which needs the plan and its assumptions. That's three verbs with three different burdens of evidence. A body that carries only the first will fail vertical logic against the second or the third.

Numbers in titles have their own rules, and they get a full section further down. For now, one habit. The number in the title must appear in the body, in the same units, rounded no more coarsely than the title rounds it. Picture a title claiming 68.8% over a chart whose labels only let a reader derive "about two thirds". It isn't wrong, but it's asking for trust it hasn't earned. Put the two component figures on the exhibit and the claim checks itself.

A note on tense and voice. Write in the past tense for what happened and the present for what's true now, and prefer an active subject that a reader could name. "Margin was impacted by mix" hides who did what. "Channel mix took 1.27 points off gross margin" names the actor and gives the size, in six fewer characters.

7. The slide whose title is not a claim

Once every title has to be a claim, some slides turn out not to have one. That discovery is the second gift of the title pass. It's also the harder one to act on, because the slides in question are usually the ones that took longest to build.

Four kinds of slide have no claim and each is handled differently.

Furniture has no claim and no content: the agenda, the "thank you", the divider with a section name. Cut the agenda from any deck people will read on their own. Keep dividers only where a deck really does hold more than one chain, and make the divider's title state the switch.

Context has content and no claim: the company overview, the market description, the team page. It isn't worthless, but it's almost never worth the position it occupies. Ask whether the reader can follow the argument without it. If they can, it goes to the appendix, and a line in the body of a later slide points at it.

Orphaned analysis has content, a claim, and no place in this argument. It's real work that answers a question this deck isn't asking. Send it to the appendix with its own action title intact. Then when somebody asks that question in the meeting, the page is one click away and already makes its point.

Two half-claims is a page carrying two weak findings, neither strong enough to title. Sometimes merging them produces one real claim. More often you'll have to admit that both were interesting, neither holds up the argument, and both go.

To cut well, you have to say what the cut costs, out loud, at the time. A cut list with a reason against each line is the difference between an edited deck and a shortened one. It's also what you hand to whoever asks in the meeting why their favourite chart is missing. Write each reason as the question the page answered, since that's the form the objection will arrive in.

8. Title length and the two-line rule

A title that wraps to three lines has stopped being a headline and become a paragraph in large type. It pushes the exhibit down and breaks the page's vertical rhythm. Worse, you can no longer read it in one movement of the eye, and that was the only reason it was in large type in the first place.

You don't have to guess the ceiling. It's arithmetic, and you can compute it for your own template. Take a widescreen slide 13.333 inches wide with half-inch margins, which leaves a text box 12.333 inches across. Set the title at 28 points and take the average character advance in a typical humanist sans as about half the point size. Each character then occupies about 0.1944 inches. That gives roughly 63 characters a line and about 127 characters in two lines. At an average of 6.6 characters a word including its space, two lines hold about 19 words.

Those font metrics are approximate. They depend on the typeface, the tracking and the mix of capitals in your text (checked September 2026 against a widescreen 16:9 default). Check the figure yourself by typing a 127-character string into your own title placeholder. The method survives even when the constant doesn't. Type a string of known length into your actual template, see where it wraps, and you've got your own ceiling in one minute.

Work the sizes through and the rule stops being abstract. The 31-word title in the sixth worked example below runs to 178 characters, which is 2.81 lines. It wraps onto a third line and has to be cut. Its 17-word replacement is 88 characters and 1.39 lines. An 11-word version is 59 characters and reads in one movement of the eye.

Titles get damaged when you cut them to fit, so cut in a fixed order. Remove the throat-clearing first ("Analysis shows that", "It should be noted that", "As can be seen"). Remove the redundant scope next, since a title inside a Kadamba board pack doesn't need the word Kadamba. Remove the second number third, keeping the one that carries the decision. Only then start on the words that carry meaning. If you get that far, ask yourself whether the page is really two pages.

The verb and the number never go. A title cut from 31 words to 18 by deleting its number has been made shorter and useless.

9. Hedging: the honest qualifier and the cowardly one

Qualifiers aren't the enemy. A claim without a qualifier is often false, and a professional who over-claims once is discounted forever afterwards. A working qualifier names a condition somebody could check, and a hedge doesn't.

Three versions of the same Kadamba claim, in ascending order of usefulness:

The third is longer than the second and does far more work, because it hands the board the lever. Renegotiate the platform fee so that the channel's gross margin rises from 30.0% to 34.0%, and the FY27 mix drag falls from 0.95 points to 0.07. So four points of fee are worth 0.88 points of group gross margin. Nobody can act on "continues to monitor". Somebody can act on 0.88 points.

You can tell them apart with four tests, and any one of them is enough on its own.

Ask what would have to be true for the sentence to be wrong. If nothing would, it's a hedge.

Ask whether the qualifier names a quantity, a date, a condition or a source. "Under the current plan mix" qualifies. "Broadly" does not.

Ask who could act on the sentence and what they'd do. A qualifier that removes the action has removed the point of the slide.

Ask whether the hedge is protecting the reader or the author. This is the uncomfortable one, and the answer is "the author" more often than analysts like.

Softening does have a proper use, and that's when the evidence really is thin. Then the professional move is to say so in the same sentence and give the reader the size of your uncertainty: a range instead of a point, a stated assumption, or a sensitivity. A title that reads "on the FY27 plan mix, and the plan assumes quick commerce reaches 22.0% of revenue" shows the reader where the soft ground is.

10. The number in the title: which one, at what precision, on what basis

Most facts offer at least three numbers, and each carries a different claim. Kadamba's gross margin fell from 42.5% to 40.7%. That's a level of 40.7%, an absolute change of 1.85 percentage points, and a relative change of 4.3%. All three are true. They aren't interchangeable.

So which one goes in the title? Lead with the number that maps to the decision, and carry the change as evidence. If the decision is a covenant test, the level matters most and the change is context. If the decision is whether a trend is worth reacting to, the change leads. If the audience manages a ratio against a policy limit, the level leads, always, because the limit is stated as a level.

Percentage points and percent catch more people than any other trap here, so it's worth being blunt. A fall of 1.85 percentage points on a 42.5% base is a relative fall of 4.3%. Write "gross margin fell 4.3%" and a careful reader will reconstruct 42.5 minus 4.3 and get 38.2. That's wrong by two and a half points. Write "fell 1.85 points" and no reconstruction is possible. Use points for changes in ratios and save percent for changes in quantities. Where both appear in one title, label both.

The decision sets the precision, and the spreadsheet doesn't get a say. Round to the last digit at which a reader's action would change. Leverage at 3.03 times against a 3.25 covenant needs two decimals, because 3.0 and 3.3 are the difference between comfortable and in breach. Revenue growth of 21.8935% needs one decimal at most and probably none, since nobody acts differently at 21.9% than at 22%. You can spot an unedited deck by a title carrying four decimals. It announces that the author moved a number without thinking about what it was for.

Basis is the one authors forget, and it's where two careful analysts produce different titles from identical data. Customer-acquisition payback computed on gross profit is a different number from payback computed on revenue. For the same US software business the two are 14.6 months and 11.4 months, a gap of 3.2 months on the same underlying spend. Neither is wrong. But a title that carries one of them without saying which leaves the reader unable to compare it to anything. The fix costs four words in the title or a source line under the exhibit: state the basis, the period and where the figure came from.

Sourcing a number in a title differs from sourcing an exhibit, and the difference matters when a page is forwarded on its own. An exhibit's source line covers everything plotted. The title's number may be derived from that exhibit, and then the components have to be visible on the page. Or it may come from somewhere else entirely, and then the title has borrowed authority the page can't support. Date anything that moves. Somebody else sets a covenant threshold, a policy limit, a platform fee and a tax rate, and every one of them changes. So a title citing one carries the date it was checked, in the footnote, where it doesn't compete with the claim.

11. Exhibit titles and text-slide titles

Both kinds of page take an action title, and the two differ in what the body has to do underneath it.

On an exhibit page, the title states what the chart proves and the chart's own furniture stays quiet. Give the exhibit a short descriptive strap or an axis label saying what's plotted. Your reader needs to know they're looking at gross margin by channel before they can check the claim. Two active headlines competing on one page is a common and confusing defect, because the eye can't tell which one is the claim. Keep the argument in the title and the identification in the strap, and put the source and period in a footnote.

On a text page, the title states the claim and the body is the reasoning that supports it, usually three or four lines of prose or a short structured list. The risk here runs the other way. A text page's body can simply restate the title in more words. A body that adds no evidence hasn't proved anything, so the page should either acquire evidence or become one line on another page.

A mixed page, meaning an exhibit with a written finding beside it, follows the exhibit rule, and the mechanism goes in the written finding. Say the title claims mix caused the fall. Then the exhibit shows the channel margins and weights, and the written note says which two channels moved and by how much. Each element does a different job and none repeats another.

One page takes a different title from every other, and that's the executive summary. Its title is the governing thought of the whole deck, so it's the only title in the file that isn't about its own page. Everything below it on that page is the key line. So the summary page reads as a title stack in miniature, and you can test it the same way. Make the summary's title the deck's argument and its bullets the body pages' titles, in order. Then it can't drift out of step with the deck behind it, because rewriting a body title forces a visible edit on the summary.

Three conventions are worth adopting and writing down once in your own house style, since consistency here is worth more than any particular choice. Keep the period and the currency in the strap and out of the title, so titles stay about the claim. Put the source line in the same place on every page. And use the same word for the same thing throughout a deck. Say gross margin on one page, product margin on the next and contribution on a third, and you've invited the meeting to spend ten minutes on vocabulary.

12. Rewriting the titles without touching the bodies

One exercise produces more insight per hour than any other in this craft, and it works because of its constraint. Take a finished deck. Rewrite every title as a claim. Change nothing else: no chart, no number, no reordering, no cutting.

Three things fall out of it, reliably.

Some titles can't be written, because the page has no claim. You've just found your cut list, and you found it without arguing about whether a chart is pretty.

Some titles turn out to be false, because the body says otherwise. Those are your errors, and they were invisible while the title was a noun phrase, since a noun phrase can't contradict anything.

Some pairs of titles turn out not to connect, because there is a step between them that no page carries. You have just found your missing slides, which are usually the most valuable pages in the finished deck, since a step nobody wrote down is a step nobody has checked.

Don't write the titles from memory. Do the rewrite in one sitting, with the numbers in front of you, and take each title from its body. Then run the read-only-the-titles test on the result. At the end you'll have a three-part output: a rewritten stack, a cut list with reasons, and a short list of pages that need to exist. Only after that's settled do you touch a chart.

The same exercise works when you review somebody else's work, and it's kinder than a redline. Send back a title list with your rewrites and three question marks. It's a more useful review than forty comments on formatting, and it takes twenty minutes.

13. The machine's title rule, and where you overrule it

The seven-rule check that grades your slides carries one rule about titles, and it decides on four mechanical tests. The title has to contain a finite verb, so it's a sentence and can't be a label. It has to avoid a list of known filing labels such as Overview, Update, Financials and Performance. Any numeral it carries has to appear in the slide's underlying data at a rounding no coarser than the title's. And it has to fit the two-line ceiling for the deck's title style.

Those four catch most of what goes wrong, cheaply and without an argument. They're also blind in the places you'd expect, and if you defer to the machine there, you'll make your deck worse.

So when should you overrule it? Three overrules are legitimate, and each has to be recorded so that a reviewer can see a decision was taken.

A section divider in a real two-chain deck has no claim and shouldn't have one. The machine flags it. Overrule it, and still make the divider's title name the switch. A topic name won't do.

A derived number in a title will fail the numeral test, because the figure isn't in the data. Derived means the reader computes it from the exhibit and can't read it off. Say the title claims the top three states carry 71.0% of revenue, worked out from six bars, none of them labelled 71.0. Overrule if the components are on the exhibit. Don't if the reader would need a fourth number they haven't got.

A legitimate rounding difference between the title and the exhibit will also trip the test, such as a title saying 3.03 times over a chart labelled 3.0. Overrule it, then ask whether the exhibit should carry the second decimal, since the covenant test lives there.

The machine can't see the part that matters. It can't tell whether the claim is true, whether it's the claim the audience needed, or whether the stack of true claims makes an argument. Those three are the whole of this craft, and they're why the rubric self-score carries 40% of the blended mark. A deck can pass the title rule on all twelve pages and still fail the read-only-the-titles test completely. You need both gates for that reason.

14. Where the assistants help here, and where they are dangerous

An assistant is unusually good at title work, for a narrow reason. Rewriting a noun phrase into a sentence is a language task with a clear target, and the model has seen a very large number of good headlines. Give it a label title and the body's numbers and it'll produce six candidates. One or two will beat what you'd have written unaided.

It's dangerous in one specific way, and that danger matters more for titles than for anything else in presentation work. Ask it for a title and it'll write a plausible one whether or not the body supports it. A plausible false title is the very defect that destroys a reader's trust. It has no access to your working, and no way to check whether 68.8% is the mix share or something you typed. It'll also invent the causal word. Give it a table showing mix and rate effects and it'll happily produce "driven by", "as a result of" and "reflecting". Those are causal claims your decomposition may or may not support.

So split the work this way. Let the assistant attack your titles, and never let it write them. Paste your stack and ask what argument it makes. That's the read-only-the-titles test, run by a reader who really hasn't seen the deck. It's the one thing an assistant does better than a colleague, because it has no memory of what you meant.

{
  "skill": "Running the read-only-the-titles test with a reader who has no context, and stress-testing each title against its own body",
  "use": "After you have rewritten every title yourself, paste the title stack alone (no bodies, no notes) and ask the assistant to state the argument it makes, name the step that is missing between any two titles, and say which titles are labels rather than claims. Then, separately, paste one title with the table beneath it and ask whether the body proves that exact claim, more, less, or something else.",
  "tools": ["Chat assistants (Claude, ChatGPT, Gemini)", "The outline view of your presentation software, to export the title stack without the bodies", "A spreadsheet holding the figures each title cites, for the recomputation step"],
  "prompt": "Here is a stack of slide titles in order, and nothing else. 1) Write the argument these titles make, in four sentences, as if you were briefing someone who will not see the deck. 2) Name any place where the argument jumps, meaning where title N+1 assumes something title N did not establish. 3) List any title that is a label rather than a claim. 4) List any title that makes a causal claim, and say what evidence would be needed to support it. Do not rewrite my titles. Titles: <PASTE>",
  "verify": "Every number in every title must be recomputed from your own source before the deck ships, and the recomputation is the gate, not the assistant's agreement. Three specific checks. First, any percentage-point figure must be checked against the two levels it came from, since a relative change wearing a points label is the commonest error the assistant will not catch. Second, any causal word in a title (driven by, because, as a result of) must be traceable to a decomposition on that page whose components sum to the total, and if the components do not sum, the causal claim is not yet earned. Third, where the assistant reports the argument it read, compare that to the argument you intended and treat any difference as your defect rather than its misreading.",
  "diy": "You must be able to write a full stack of action titles from a fact set with nothing open but the numbers, and you must be able to run the gap, leap and repeat diagnosis by hand on somebody else's deck in under ten minutes. Both are gated here and both are assumed by everything downstream: the storyline crown scores a title stack directly, the archetype modules assume you can title a body page cold, and the capstone runs the whole test under time pressure on data you have not seen. An analyst who can only produce titles with an assistant in the loop cannot defend one in a room, and defending one in a room is the actual job.",
  "market": "IN"
}

Common mistakes & how experts think differently

1. Titling the exhibit instead of the finding. The chart shows gross margin by channel, so the title says "Gross Margin by Channel". It's tempting because it's accurate, and because that's what the chart is called in your own head. An expert titles the reason the chart is in the deck, meaning the claim it supports, and lets the axis labels say what's plotted.

2. Writing the titles last. Titles written after the deck is built describe what got built. Titles written first choose what to build, and they routinely delete two or three planned pages before anyone spends an hour on them. Experts write the stack on paper before opening the software, then treat every later change to a title as a signal that the argument moved.

3. Confusing points with percent. A 1.85-point fall on a 42.5% base is a 4.3% relative fall, and people who know the difference perfectly well still swap the two in titles. The habit that fixes it is to write the unit before the number, never after. Then each "1.85 points" or "4.3%" on the page is something you decided.

4. Under-claiming to feel safe. The analyst who computed a two-factor decomposition and titled the page "Gross Margin Analysis" has done the hard part and thrown away the credit. Experts state the strongest claim their evidence supports and then add the condition under which it fails, which is both braver and safer than the vague version.

5. Hedging in the title and claiming in the meeting. The written page says mix "may be a factor" and the presenter says mix is the whole story. Whichever one is right, the deck is now a record that contradicts its author. Experts write what they will say, so the file and the room agree.

6. Letting a title drift when its chart is rebuilt. A chart gets rebuilt on Thursday night, the numbers move slightly, and the title still carries Tuesday's figure. In real decks this causes more title-body contradictions than anything else. You catch it with one pass through the deck, reading every title against its own body, done last, with nothing else being edited.

7. Building the appendix as a graveyard. Cut pages get dragged behind a divider and forgotten, keeping their label titles. An expert's appendix carries action titles and an index, because people go looking in the appendix during a meeting, under time pressure. Nobody can find a page called "Additional Analysis".

8. Treating the machine's title score as the goal. A stack of twelve mechanically valid sentences that don't follow from one another scores well and communicates nothing. Experts use the machine check as a floor that takes thirty seconds. They spend the time it saves on the two tests it can't run: whether the claims are true, and whether they make an argument.

9. Writing a title longer than the reader's eye. A three-line title has stopped being a headline. Experts cut in a fixed order: throat-clearing, then redundant scope, then the second number. If the title still won't fit in two lines after that, they take it as evidence the page is two pages.

10. Letting a recommendation appear without a bridge. The findings are sound, the recommendation is sensible, and no page connects them, so the reader is asked to take the last step on trust. Experts catch this by reading only the titles. While the bodies are there to distract you, the leap is invisible.

Worked examples

Every figure below comes from one computation, and the Verify block at the end of the section restates each at full precision. Both companies are synthetic composites, and their statements are internally consistent: shares sum to 100, decompositions sum to their totals, and the margin bridges tie.

Worked example 1: A ten-slide board pack, retitled (India, ₹)

The setting. Kadamba Foods Ltd (synthetic) makes packaged snacks and sells through five channels. FY26 ended 31 March 2026. Revenue was ₹412.0 crore against ₹338.0 crore in FY25. Channel revenue moved as follows, in ₹ crore: general trade 186.0 to 178.0, modern trade 74.0 to 86.0, quick commerce 21.0 to 63.0, e-commerce marketplaces 38.0 to 55.0, exports 19.0 to 30.0. Channel gross margins, after trade spend and platform fees, were 44.5%, 41.5%, 31.0%, 38.5% and 47.5% in FY25 and 44.0%, 41.0%, 30.0%, 38.0% and 47.0% in FY26. Operating costs were employee ₹36.4 crore to ₹42.6 crore, advertising and promotion ₹31.2 crore to ₹38.9 crore, distribution ₹24.9 crore to ₹31.4 crore, other overhead ₹22.1 crore to ₹26.2 crore. Receivables, inventory and payables were ₹47.9, ₹44.6 and ₹38.1 crore in FY25 and ₹63.4, ₹58.9 and ₹47.2 crore in FY26. Net debt went from ₹52.0 crore to ₹86.0 crore against a lender's covenant of 3.25 times EBITDA. A second production line at Nashik costs ₹34.0 crore in total, of which ₹22.0 crore falls in FY27.

The message the analyst is carrying. Growth is real and is coming from channels that pay less. The company is close to a covenant and is about to spend. The board has to decide how the second line is funded, today.

The wrong version. Ten slides with the titles in the left column below. The exhibits are competent. The analyst spent two evenings on the margin waterfall and the working-capital bridge. Both are correct, and both are titled in a way that hides what they found.

The stack read alone. Nineteen words in the nine body titles, five seconds of reading, and no argument. Two structural defects show up the moment the labels are replaced by claims, and neither is visible while the labels are in place.

The first is a gap between slides 5 and 6. The pack asserts that channel mix moved and then asserts that gross margin fell, and never connects them. A director can't tell from the pack whether the margin fell because of mix, because of price, or because of input costs, and those three imply completely different responses. The bridge exists as arithmetic, and nobody has given it a page.

The second is a gap at the end. The pack finishes on a list of risks and never says what the board is being asked to approve. End a pack on risks and you've handed the meeting its agenda. That agenda is anxiety, with no decision in it.

The arithmetic behind the rewrite, in order. Revenue growth is 412.0 / 338.0 − 1 = 21.8935%. Quick commerce, e-commerce and exports were 21.0 + 38.0 + 19.0 = ₹78.0 crore of FY25 revenue, which is 78.0 / 338.0 = 23.0769%. They grew to ₹148.0 crore, an increase of ₹70.0 crore against the company's total increase of ₹74.0 crore, so they delivered 70.0 / 74.0 = 94.5946% of the growth. Quick commerce went from 6.2130% to 15.2913% of revenue, a multiple of 63.0 / 21.0 = 3.0000, and general trade fell 178.0 / 186.0 − 1 = −4.3011%.

Gross profit is 143.6450 crore in FY25 and 167.4800 crore in FY26, which is 42.4985% and 40.6505% of revenue, a fall of 1.8480 points. Now apply FY26 channel weights to FY25 channel margins, and the blend is 41.2269%, so the mix effect is 41.2269 − 42.4985 = −1.2716 points and the rate effect is 40.6505 − 41.2269 = −0.5765 points. The two sum to −1.8480, and that's your check. Mix is 68.8071% of the fall.

Total operating cost went from ₹114.6 crore to ₹139.1 crore, a rise of 21.3787%, against revenue's 21.8935%. As a share of revenue that is 33.9053% falling to 33.7621%, an improvement of 0.1432 points. EBITDA is 29.0450 crore and 28.3800 crore, so margin fell from 8.5932% to 6.8883%, a fall of 1.7048 points. The bridge is −1.8480 + 0.1432 = −1.7048, which ties with nothing left over. Mix alone accounts for 1.2716 / 1.7048 = 74.5862% of the EBITDA margin fall.

Working capital went from ₹54.4 crore to ₹75.1 crore, an absorption of ₹20.7 crore. Hold the FY25 cash cycle and apply it to FY26 volumes and net working capital would have been 66.5647 crore, so growth explains 12.1647 crore and the longer cycle explains 8.5353 crore, which sum to 20.7000. The cycle lengthened from 63.9334 days to 73.6323 days, or 9.6989 days.

Last comes the covenant. Leverage is 86.0 / 28.3800 = 3.0303 times against a covenant of 3.25. The EBITDA floor that covenant implies is 86.0 / 3.25 = 26.4615 crore, so headroom is 1.9185 crore, which is 6.7599% of FY26 EBITDA. Fund the ₹22.0 crore of FY27 capital spending with debt and net debt becomes ₹108.0 crore, which at flat EBITDA is 3.8055 times. Raise ₹25.0 crore of equity instead and net debt is ₹83.0 crore, or 2.9246 times.

The corrected version. Two slides cut, two added, ten pages out, no chart rebuilt.

Slides 5 and 10 are new and are the two most valuable pages in the pack, because each closes one of the two gaps. Slide 5 is the analyst's own two evenings of work, finally titled with what it found. Slide 10 is the decision, which had been in the analyst's head all along and never reached the paper.

Here's the cut list, with the cost of each cut stated. The agenda goes, because a pack circulated 48 hours ahead gets read, and nobody walks the board through it. The cost is that the chair loses a prompt, and a covering email replaces that at no charge. The business overview goes, because this board approved the strategy it describes. The cost is that a newly appointed director has no orientation page. Keep the overview in the appendix and point at it in the covering email, and that's covered.

The check that makes this defensible. Every claim in the stack is verified by a second route. Channel shares sum to 100.000000% in both years. Mix and rate effects sum to the total margin change to ten decimal places. Add the gross margin change to the operating-cost ratio change and you get the EBITDA margin change. Growth and cycle effects sum to the working-capital absorption with nothing left over. When a title stack's arithmetic reconciles in four separate places, a director can interrogate it for twenty minutes without finding a loose end. That's what the pack is for.

What would break it. Slide 3 claims general trade fell for the first time. That's a fact about this company's history, and the FY26 numbers can't establish it. It needs the prior years, and if they aren't in the pack, the words come out. Slide 6 makes a claim about the aggregate, and it survives only because somebody checked the aggregate. Two of the four cost lines grew at 24.7% and 26.1%, well ahead of revenue. Write the title from those two and you'd have shipped a false statement.

The checklist to apply.

1. Extract the titles into a list with the bodies hidden, and read them cold.
2. Write down the argument the list makes, in your own words, before looking at any slide.
3. Mark every place where one title assumes something the previous title did not establish.
4. Mark every title that is a noun phrase, and decide whether the page has a claim at all.
5. Rewrite each title from its own body, with the number that makes it checkable.
6. Cut the pages that still have no claim, and write the cost of each cut next to it.
7. Add the pages the gaps demand, and check that each new page's arithmetic ties by a second route.
8. Re-read the stack. If it does not end on a decision, the pack is not finished.

Worked example 2: The body that proves a different claim (US, $)

The setting. Northsound Software Inc (synthetic) sells a B2B subscription product in two segments. Annual recurring revenue at the start of FY25 was $70.0m, split $42.0m enterprise and $28.0m SMB. At the start of FY26 it was $88.2m, split $56.0m enterprise and $32.2m SMB. Gross revenue retention was 94.0% for enterprise and 86.5% for SMB in FY25, and 96.5% and 74.0% in FY26. Net revenue retention was 108.0% in FY25 and 112.0% in FY26. The company is preparing a board pack to support a FY27 plan.

The message on the page. The title on slide 4 reads: Net revenue retention improved to 112% on the back of stronger renewals. Beneath it sits a two-year table of gross retention, expansion and net retention, split by segment.

The diagnosis. Cover the title and read the table. Blended gross retention is (42.0 × 0.940 + 28.0 × 0.865) / 70.0 = 91.0000% in FY25 and (56.0 × 0.965 + 32.2 × 0.740) / 88.2 = 88.2857% in FY26. Renewals fell 2.7143 points. Expansion is the residual against net retention, so 108.0 − 91.0000 = 17.0000% in FY25 and 112.0 − 88.2857 = 23.7143% in FY26, a rise of 6.7143 points. The two components sum to the +4.0000 point move in net retention. That's the check.

The title claims renewals drove the improvement. The body proves that renewals subtracted from it. Expansion contributed 6.7143 / 4.0000 = 167.8571% of the change and retention contributed −67.8571%, and the two contributions sum to 100%.

And it's worse than the headline suggests. A careful reader will find this part in the meeting if the author doesn't find it first. Hold FY25's segment weights and apply FY26's segment rates: 0.6000 × 0.965 + 0.4000 × 0.740 = 87.5000%. So the rate effect is 87.5000 − 91.0000 = −3.5000 points and the mix effect is 88.2857 − 87.5000 = +0.7857 points, summing to −2.7143. The enterprise share of the base rose from 60.0000% to 63.4921%, and that favourable shift masked eight tenths of a point of underlying deterioration. SMB gross retention fell 12.5 points and SMB is still 36.5% of the base.

Repair one: fix the title. Net revenue retention rose 4.0 points to 112%, entirely from expansion, while renewals fell 2.7 points. That's sixteen words and 102 characters, or 1.6082 lines on the template worked through in the title-length section. It fits inside the two-line ceiling, and it now describes the table under it. The body needs no change beyond adding the expansion row, if it isn't already visible.

Repair two: fix the body. Keep a claim about stronger renewals and give it evidence that exists. Enterprise gross retention did improve, from 94.0% to 96.5%. Retitle to: Enterprise renewals improved 2.5 points to 96.5%, and replace the blended table with the segment table. That page is true.

Which repair is right, and why the slide alone cannot tell you. Look at the chain. This deck's later pages compute the FY27 plan from company-level net retention, because the plan's retained-and-expanded line is the opening base multiplied by net retention. Narrow the page to enterprise and the chain loses the company number it needs two slides later. It also drops, without a word, a 12.5-point fall in a segment that's 36.5% of the base. The board would then meet the SMB problem for the first time in the risk section, the worst possible place to introduce it.

So repair one is right here. Repair two would be right in a different deck, one whose governing thought was a segment strategy decision, and there the company-level number would be the distraction. The slide doesn't carry enough information to choose, and that holds generally: vertical logic tells you the page is broken, and horizontal logic tells you how to fix it.

What would break it. The expansion figure is a residual, computed as net retention less gross retention, so it inherits every definitional choice in those two. If the company defines net retention on a cohort basis and gross retention on a period basis, the subtraction isn't valid and the 23.7% is meaningless. The title carries the number, so the title owes the reader the definition, which belongs in the source line under the exhibit.

The checklist to apply.

1. Cover the title. Write what the body proves, in one sentence.
2. Compare that sentence to the title, and classify: proves it, proves less, proves more, proves something else.
3. For a causal title, check that the components on the page sum to the total they claim to explain.
4. Check the aggregate whenever the title makes a claim about a whole.
5. Decompose into rate and mix whenever a weighted average moved, since the mix can hide the rate.
6. Write both repairs before choosing between them.
7. Choose using the chain: which version leaves the next slide with what it needs.
8. Put the definition of any residual figure in the source line.

Worked example 3: Which number goes in the title (US, $)

The setting. Northsound's FY25 sales and marketing spend was $15.4m and it produced $12.6m of new-logo annual recurring revenue at a 76.0% gross margin. FY26 spend was $17.5m producing $18.4m of new-logo ARR at a 78.0% gross margin. The company's own policy paper sets an 18-month ceiling on customer-acquisition payback. The slide is an exhibit page in the FY27 plan deck.

The wrong version. Title: Sales efficiency improved 24%. Body: a two-bar chart of payback in months, labelled 19.3 and 14.6, with no axis title and no note on how payback was computed.

Six words, three separate defects. The number is a relative change on a ratio, which nobody can reconstruct. It carries no level, so a reader can't test it against the 18-month policy. And it doesn't state the basis, which is the defect that makes the other two unfixable.

The arithmetic. Payback in months on a gross-margin basis is spend divided by new ARR times gross margin, times twelve. FY25 is 15.4 / (12.6 × 0.760) × 12 = 19.2982 months and FY26 is 17.5 / (18.4 × 0.780) × 12 = 14.6321 months. The absolute change is −4.6661 months and the relative change is −24.1791%.

Now compute the same thing on a revenue basis, which is how roughly half of working analysts do it: 15.4 / 12.6 × 12 = 14.6667 months and 17.5 / 18.4 × 12 = 11.4130 months, an absolute change of −3.2536 months and a relative change of −22.1838%. Both pairs are correct. They differ by 3.2191 months in FY26 and by two points of relative improvement, and nothing on the slide says which one is on the chart.

The corrected version. Title: CAC payback fell 4.7 months to 14.6, on a gross-margin basis, against an 18-month policy limit. That's sixteen words and 94 characters, 1.4820 lines on the same template, so it's inside the ceiling. The strap under it identifies the exhibit as payback in months, FY25 and FY26. The footnote gives the formula, the spend definition and the source system, with the date the figures were pulled.

Look at what each element of that title does. The level, 14.6, comes first because the decision this exhibit serves is a policy test, and the policy is written as a level. Next comes the absolute change, 4.7 months, as the evidence. It's stated in months because a reader can subtract months from the two bars and check it, which nobody can do with 24%. The basis takes four words, and it's what makes the number comparable to anything else. Last, the policy limit is the standard your reader measures against. Stating it turns a fact into a verdict: 18.0000 − 14.6321 = 3.3679 months of headroom.

What would break it. Two things, both ordinary. If new-logo ARR is defined to include expansion within the first year, the denominator grows and the payback shortens. So the definition goes in the footnote. And if gross margin moves again in FY27, the gross-margin-basis figure moves without any change in sales efficiency. A series of these titles across four quarters must hold the basis fixed or say when it changed.

The checklist to apply.

1. List the three candidate numbers: the level, the absolute change, the relative change.
2. Name the decision the exhibit serves, and pick the number that maps to it.
3. Use points for a change in a ratio and percent for a change in a quantity, and label both.
4. Round to the last digit at which a reader's action would change.
5. State the basis in the title when two defensible bases exist.
6. Put the formula, the period and the source in the footnote, with a date.
7. Check the title's number appears on the exhibit at no coarser a rounding.
8. If a comparison standard exists, put it in the title, because it converts the fact into a verdict.

Worked example 4: The risk slide, hedged and unhedged (India, ₹)

The setting. Kadamba's FY27 plan carries channel revenue of general trade ₹170.0 crore, modern trade ₹96.0 crore, quick commerce ₹105.0 crore, e-commerce ₹68.0 crore and exports ₹38.0 crore, a total of ₹477.0 crore and growth of 15.7767%. The plan holds FY26 channel gross margins. Net debt is ₹86.0 crore, FY26 EBITDA ₹28.38 crore, and the covenant is 3.25 times.

The wrong version. A slide titled Risks and Outlook, with six bullets: input cost inflation, quick-commerce platform fee pressure, general trade decline, competitive intensity, key personnel, regulatory change. None of the six carries a number, and the final bullet reads that management continues to monitor these risks closely.

Nobody can act on that slide, and the reason has nothing to do with style. A risk without a size can't be ranked. A list you can't rank can't be mitigated in any order, so the board leaves with an impression of danger and no instruction.

The arithmetic that turns three of them into claims. The covenant risk is a direct calculation: EBITDA must stay above 86.0 / 3.25 = 26.4615 crore, so headroom is 28.3800 − 26.4615 = 1.9185 crore, or 6.7599% of FY26 EBITDA.

The platform-fee risk is a mix calculation. On the FY27 plan mix at FY26 channel rates, gross margin is 39.6981%, which is 0.9524 points below FY26, entirely from mix since the rates are held. Renegotiate the quick-commerce fee so that the channel's gross margin rises from 30.0% to 34.0% and the plan margin becomes 40.5786%, a drag of only 0.0719 points. Four points of platform fee are therefore worth 0.8805 points of group gross margin on the plan mix.

The general-trade risk is a downside case. The plan assumes general trade falls 4.4944%. Suppose it falls 10.0% instead, to ₹160.2 crore. FY27 revenue becomes ₹467.2 crore, ₹9.8 crore below plan, and gross profit falls from ₹189.36 crore to ₹185.048 crore, a shortfall of 4.3120 crore. Against covenant headroom of ₹1.9185 crore, that shortfall is 2.2476 times the available cushion. One channel missing its plan by five and a half points is more than twice the size of the cushion that exists today. Be careful what you let the title say, though. The arithmetic supports a ratio. It doesn't support a breach. The FY27 covenant test runs on FY27 EBITDA and on net debt at the test date, and the plan states neither. A page claiming the covenant is breached would be claiming more than it has shown.

The corrected version. Three slides instead of one, each with a title that is a claim, and the three unquantified risks kept and labelled as such.

The remaining three risks stay on the third slide, under a line that says they haven't been quantified and why. Competitive intensity, key personnel and regulatory change are real. None of them has a number that would survive a director's question, so they're listed as watch items with an owner against each. Saying that two of six risks can't yet be sized is a stronger position than pretending all six are equal. It also tells the board what next quarter's work is.

The hedge, rewritten. Management continues to monitor platform fee pressure becomes: at the current 30.0% channel margin the FY27 mix drag is 0.95 points, and a four-point fee reduction removes almost all of it. The first version protects the author. The second gives the commercial team a target and the board a number to hold them to. It can also be wrong, and that's what makes it worth reading.

What would break it. Every one of the three quantified risks holds the other variables fixed, and real downside cases arrive together. If general trade falls 10% and the platform fee doesn't move, the two effects compound. A slide showing them independently will understate the joint case. Say so on the page, and show the combined case if the plan carries one.

The checklist to apply.

1. Ask of each risk: what is the number, and against what standard.
2. Compute the constraint first, since it is what every other risk is measured against.
3. Express each risk as a change in the quantity the constraint is written in.
4. Rank by size, not by how uncomfortable each one is to discuss.
5. Keep the risks you cannot size, and say plainly that they are not sized.
6. Name an owner and a date for each unquantified item, so the list is work rather than decoration.
7. Rewrite every hedge into a claim with a stated condition under which it fails.
8. State whether the cases are independent, and show the joint case where it matters.

Worked example 5: The leap between two true titles (US, $)

The setting. Northsound's FY27 plan deck runs five pages before the ask. Ending ARR was 88.2000 at the end of FY25, computed as 70.0 × 1.080 + 12.6, and 117.1840 at the end of FY26, computed as 88.2 × 1.120 + 18.4. FY26 growth is 32.8617%. FY26 payback on a gross-margin basis is 14.6321 months at a 78.0% gross margin. The FY27 target is $150.0m of ARR and the ask is $24m of sales and marketing spend.

The wrong version. Five titles, every one of them true.

1. ARR grew 32.9% to $117.2m, up from 26.0% growth in FY25
2. Net revenue retention reached 112%, four points better than FY25
3. CAC payback fell 4.7 months to 14.6, on a gross-margin basis
4. The FY27 plan targets $150m of ARR
5. The ask is $24m of FY27 sales and marketing spend

Read those five in order. You get three facts about the past, then a target, then a number. Nothing on any page says what $150m requires, and nothing says why the requirement costs $24m. A board approving this is approving a number it's been given no way to test. And the three pages of good news before it make the approval feel like a formality. That's the effect that gets analysts into trouble later.

The arithmetic the missing page carries. The plan's opening base is $117.1840m. If net revenue retention holds at 112.0%, retained and expanded ARR is 117.1840 × 1.120 = 131.2461, so new-logo ARR must be 150.0 − 131.2461 = 18.7539, which is 1.9235% above FY26's $18.4m. At FY26's payback, the spend that buys it is 18.7539 × 0.780 × 14.6321 / 12 = 17.8366, so $17.8m.

If net revenue retention slips to 106.0%, which is what SMB gross retention at 74.0% implies if that segment doesn't stabilise, retained and expanded ARR is 124.2150 and new-logo ARR must be 25.7850, or 40.1357% above FY26. The spend that buys that is 25.7850 × 0.780 × 14.6321 / 12 = 24.5237, so $24.5m.

The gap between the two cases is 6.6871 million dollars of spend, and the ask of $24m is sized for the second one. The deck never says so.

The corrected version. One page inserted between the third and fourth titles, with this title: The $24m ask is sized for net revenue retention falling to 106%; at 112% the plan needs $17.8m. The body is a two-column bridge from $117.2m to $150.0m under each retention case, with the new-logo requirement and the implied spend in each column, and the payback assumption stated once.

That page does three things at once. It turns the ask from a number into a consequence of an assumption. It makes the ask depend on the SMB retention problem, so the earlier segment page now has a job. And it gives the board a lever. A director who believes retention holds at 112% can approve $17.8m and ask for the rest at the half-year. That's a better meeting than a yes-or-no on $24m.

What would break it. The bridge assumes FY26's payback holds through FY27, and payback usually deteriorates as a company works through the easiest part of its market. If payback returns to FY25's 19.3 months, the 106% case needs 25.7850 × 0.780 × 19.2982 / 12 = 32.3443 million, which is a third more than the ask. A plan page that holds payback constant without saying so is making a forecast in a footnote. The better version states the assumption in the title's own body and shows one sensitivity.

The checklist to apply.

1. Read the titles in order and mark the first place where you would ask why.
2. Ask whether the next title follows from the previous one or from something unstated.
3. Where a recommendation follows findings, look for the page that sizes the recommendation.
4. Compute the bridge from where you are to where the plan says you will be.
5. Compute it under two assumptions, not one.
6. State which assumption the ask is sized for.
7. Check that every earlier page in the deck is now doing work, and cut any that is not.
8. Name the assumption that would most change the answer, and show one sensitivity on it.

Worked example 6: Exhibit titles, the two-line cut, and overruling the machine (India, ₹)

The setting. The bridge page from the first worked example, Kadamba's mix-versus-rate decomposition, has to be built. The exhibit is a two-bar waterfall from 42.4985% to 40.6505% with a mix bar of −1.2716 points and a rate bar of −0.5765 points. The deck's title style is 28 point on a 13.333-inch widescreen slide with half-inch margins, which the title-length arithmetic earlier puts at about 63 characters a line and 127 characters in two lines.

The wrong version. The analyst's first title runs: Analysis of the FY26 gross margin performance by channel shows that the decline in the margin was largely attributable to the change in the mix of channels rather than to pricing. That's 31 words and 178 characters. At 63.4 characters a line it's 2.8064 lines, so it wraps onto a third line, pushes the waterfall down and breaks the page's grid.

It also fails in three quieter ways. It opens with throat-clearing (Analysis of ... shows that), which costs 22 characters and says nothing. It repeats the scope, since the reader knows from the strap that they're looking at FY26 gross margin. And it hedges with largely attributable where the decomposition gives the figure itself.

The cut, in the fixed order. Remove the throat-clearing first. Remove the redundant scope second, since the strap under the title identifies the exhibit. Remove the second number third, if two are competing. Only then cut words that carry meaning.

The 17-word version reads: Channel mix took 1.27 of the 1.85 points of gross margin lost in FY26; pricing took 0.58. It keeps the verb, both numbers and the year. The 11-word version reads: Channel mix took 1.27 of the 1.85 points; pricing took 0.58. It drops the year and the word gross. That's a fair trade on an exhibit page whose strap carries both, and a bad one on a page that'll be forwarded on its own.

The exhibit's own furniture. The title claims the cause. The strap identifies the exhibit: Group gross margin bridge, FY25 to FY26, percentage points. The footnote gives the method and the source: mix effect computed at FY26 channel weights on FY25 channel margins; rate effect as the residual; channel margins after trade spend and platform fees; management accounts, FY26. Each of the three does a different job, and no two of them repeat each other. A common defect is a strap that restates the title in smaller type. It spends a line of the page saying nothing new.

Where the machine is wrong, and what you do about it. Run the seven-rule check on this page and the title rule flags it. The title carries 1.27, 1.85 and 0.58. The underlying data table holds the channel revenues and margins and none of those three figures, so the numeral test fails. The title's numbers are derived, and they aren't printed anywhere in the data.

Overruling is correct here, for a specific reason a reviewer must be able to see. The two component figures do appear, as the labelled bars of the waterfall, and they sum to the total change. So a reader can check every number in the title against the exhibit without leaving the page. Record the overrule in one line: derived from labelled bars, components sum to the total. That line separates a considered overrule from an author ignoring a check they found inconvenient.

Two other overrules belong to this deck. The divider before the Nashik decision has no claim and shouldn't have one, since it announces a new chain. And slide 8's title says 3.03 times where the exhibit's axis is labelled to one decimal. The machine can't judge a rounding mismatch like that. Overrule the flag and add the second decimal to the exhibit, because the covenant test is decided in the second decimal.

What would break it. The character-per-line constant depends on the typeface and the tracking. A deck set in a wider face will wrap earlier than 63 characters, and the 17-word title may itself run to three lines. The method still works. Type a 127-character string into the actual title placeholder and see where it breaks.

The checklist to apply.

1. Write the claim first, at whatever length it comes out.
2. Count the characters and divide by your template's measured characters per line.
3. Cut throat-clearing, then redundant scope, then the second number.
4. Stop before cutting the verb or the number that makes the claim checkable.
5. Give the exhibit a strap that identifies rather than argues.
6. Put method, source and date in the footnote, once.
7. Run the machine check and read every flag.
8. Overrule only with a written reason, and only where a reader can still verify the title from the page.

Verify

Every figure printed above, at full precision, grouped by the computation that produced it. Where a decomposition appears, its residual is shown, and every residual is zero.

Kadamba Foods, revenue and channels. FY25 revenue 338.0000; FY26 revenue 412.0000; growth 21.8935%; increase 74.0000. Channel FY26 shares: general trade 43.2039%, modern trade 20.8738%, quick commerce 15.2913%, e-commerce 13.3495%, exports 7.2816%; sum 100.000000%. FY25 shares: 55.0296%, 21.8935%, 6.2130%, 11.2426%, 5.6213%; sum 100.000000%. Channel growth: −4.3011%, +16.2162%, +200.0000%, +44.7368%, +57.8947%. Three growth channels: 78.0000 to 148.0000, an increase of 70.0000, which is 94.5946% of the total increase and was 23.0769% of FY25 revenue.

Kadamba Foods, margin. Gross profit 143.6450 and 167.4800; gross margin 42.4985% and 40.6505%; change −1.8480 points; relative change −4.3485%. FY26 weights at FY25 rates 41.2269%. Mix −1.2716 points; rate −0.5765 points; sum −1.8480352732. Mix share 68.8071%; rate share 31.1929%.

Kadamba Foods, costs and EBITDA. Operating cost 114.6000 to 139.1000, growth 21.3787%. Component growth: employee +17.0330%, advertising +24.6795%, distribution +26.1044%, other +18.5520%. Opex ratio 33.9053% to 33.7621%, change −0.1432 points. EBITDA 29.0450 and 28.3800, change −2.2896% and −0.6650 crore. EBITDA margin 8.5932% and 6.8883%, change −1.7048 points. Bridge −1.7048457517 points. Mix share of the EBITDA margin fall 74.5862%.

Kadamba Foods, working capital. COGS 194.3550 and 244.5200. Receivable days 51.7263 and 56.1675; inventory days 83.7591 and 87.9212; payable days 71.5521 and 70.4564. Cash conversion cycle 63.9334 and 73.6323 days, lengthening 9.6989 days. Net working capital 54.4000 and 75.1000, absorption 20.7000. At the FY25 cycle on FY26 volumes 66.5647. Growth effect +12.1647; cycle effect +8.5353; sum 20.7000000000. Cycle share 41.2333%.

Kadamba Foods, debt and the covenant. Net debt 52.00 and 86.00. Leverage 1.7903 and 3.0303 times. EBITDA floor at 3.25 times 26.4615; headroom 1.9185 crore and 6.7599%. Net debt ceiling 92.2350; net debt headroom 6.2350. FY27 debt-funded net debt 108.00 and leverage 3.8055 times; with ₹25 crore of equity, 83.00 and 2.9246 times.

Kadamba Foods, the FY27 plan. Revenue 477.0000, growth 15.7767%. Shares 35.6394%, 20.1258%, 22.0126%, 14.2558%, 7.9665%; sum 100.000000%. General trade plan growth −4.4944%. Gross margin at FY26 rates 39.6981%, a drag of −0.9524 points. With quick commerce at 34.0%, 40.5786% and a drag of −0.0719 points; the four-point fee move is worth 0.8805 points. Downside general trade 160.2000; revenue 467.2000, shortfall 9.8000; gross profit 185.0480, shortfall 4.3120; margin 39.6079%, a further −0.0902 points; shortfall as a multiple of covenant headroom 2.2476.

Northsound Software, ARR and retention. FY25 opening 70.0000, retained and expanded 75.6000, new logos 12.6000, closing 88.2000, growth 26.0000%. FY26 opening 88.2000, retained and expanded 98.7840, new logos 18.4000, closing 117.1840, growth 32.8617%. Blended gross retention 91.0000% and 88.2857%, change −2.7143 points. Expansion 17.0000% and 23.7143%, change +6.7143 points. Net retention change +4.0000 points; expansion contribution 167.8571%, retention contribution −67.8571%. Enterprise share 60.0000% and 63.4921%, shift +3.4921 points. FY25 weights at FY26 rates 87.5000%; rate −3.5000 points; mix +0.7857 points; sum −2.7142857143. Enterprise gross retention +2.5000 points; SMB −12.5000 points; SMB share of the FY26 base 36.5079%.

Northsound Software, payback and the plan. Payback on a gross-margin basis 19.2982 and 14.6321 months, change −4.6661 months and −24.1791%. On a revenue basis 14.6667 and 11.4130 months, change −3.2536 months and −22.1838%. Gap between the bases in FY26, 3.2191 months. Headroom against the 18-month limit 3.3679 months. FY27 target 150.00. At 112.0% net retention: retained 131.2461, new logos required 18.7539, growth on FY26 new logos 1.9235%, spend 17.8366. At 106.0%: retained 124.2150, new logos required 25.7850, growth 40.1357%, spend 24.5237. Difference in spend 6.6871. Growth to the target 28.0038%.

Title mechanics. Text box 12.3330 inches; character advance at 28 point 0.194444 inches; 63.4269 characters a line; 126.8537 characters in two lines. First-draft title 31 words, 178 characters, 2.8064 lines. Cut version 17 words, 88 characters, 1.3874 lines. Compressed version 11 words, 59 characters, 0.9302 lines. The retention repair title 16 words, 102 characters, 1.6082 lines; the payback title 16 words, 94 characters, 1.4820 lines. Character counts exclude a terminating full stop, which a slide title does not carry. Label stack 19 words, 5.1818 seconds at 220 words a minute; claim stack 114 words, 31.0909 seconds; difference 25.9091 seconds. The ten rewritten titles of the first worked example run 153 words. Font metrics are approximate and typeface-dependent (checked September 2026 against a 16:9 widescreen default; confirm against your own template).

Practice set

Work each problem before you read the solution. You'll need a calculator. Numeric answers within 2% of the stated figure count as correct unless a problem says otherwise. Where a problem asks for a title, the test is whether a reader could disagree with it and check it against the data given. The datasets are the two synthetic companies from the worked examples, so every figure ties to the Verify block.

P1 (guided). Decompose and title (India). Kadamba's gross margin fell from 42.4985% to 40.6505%. Applying FY26 channel weights to FY25 channel margins gives 41.2269%. Split the fall into mix and rate, show that they sum to the total, and write the action title for the page.

Solution. Mix is 41.2269 − 42.4985 = −1.2716 points and rate is 40.6505 − 41.2269 = −0.5765 points. They sum to −1.8480, which is the whole change, and that sum is the check the page must show. Mix is 1.2716 / 1.8480 = 68.8071% of the fall. A title that carries it: Channel mix took 1.27 of the 1.85 points of gross margin lost in FY26; pricing took 0.58. It has a subject, a verb and two numbers a reader can check against the labelled bars. A title reading Gross Margin Bridge names the exhibit and throws away the finding. One reading Margin fell on adverse mix asserts the cause without the size, so nobody can tell whether it matters.

P2 (guided). Say what the body proves (US). Northsound's slide is titled: Net revenue retention improved to 112% on the back of stronger renewals. The table shows blended gross retention of 91.0000% in FY25 and 88.2857% in FY26, and net retention of 108.0% and 112.0%. State what the body proves, give the two component changes, and repair the title.

Solution. Gross retention fell 88.2857 − 91.0000 = −2.7143 points. Expansion is the residual: 108.0 − 91.0000 = 17.0000% and 112.0 − 88.2857 = 23.7143%, a rise of +6.7143 points. The two sum to the +4.0000 point rise in net retention. The body proves that expansion rose and renewals fell, which is the opposite of the title's causal claim. Repair: Net revenue retention rose 4.0 points to 112%, entirely from expansion, while renewals fell 2.7 points. Expansion contributed 6.7143 / 4.0000 = 167.8571% of the change and retention −67.8571%, and those two contributions sum to 100%. That's the arithmetic check that the causal claim is now the right way round.

P3 (guided). Size the constraint and title it (India). Kadamba's net debt is ₹86.0 crore, FY26 EBITDA is ₹28.38 crore and the covenant is 3.25 times. Give leverage, the EBITDA floor, the headroom in rupees and in percent, and write the risk title.

Solution. Leverage is 86.0 / 28.3800 = 3.0303 times. The floor is 86.0 / 3.25 = 26.4615 crore, so headroom is 28.3800 − 26.4615 = 1.9185 crore, which is 1.9185 / 28.3800 = 6.7599% of EBITDA. Title: A 6.8% fall in EBITDA breaches the 3.25 times covenant, leaving ₹1.92 crore of headroom. Two properties make it work. It states the constraint as a change in EBITDA, which management controls. Nobody manages the ratio directly. And it's falsifiable, since a director with a different EBITDA forecast can dispute it immediately. That's the conversation the slide exists to start.

P4. Points against percent (US). Northsound's net revenue retention went from 108.0% to 112.0%. Give the absolute change and the relative change, and say which belongs in a title and why.

Solution. The absolute change is 112.0 − 108.0 = 4.0000 percentage points. The relative change is 112.0 / 108.0 − 1 = 3.7037%. The absolute change belongs in the title, for two reasons. Retention is itself a ratio, so a percent change on it produces a number the reader can't reconstruct from the two levels shown on the chart. And the plan arithmetic downstream multiplies the opening base by the retention level, so the level and its point change are the quantities the rest of the deck consumes. Writing that retention improved 3.7% invites a reader to compute 108.0 plus 3.7 and reach 111.7, which is wrong, and to compute 112 minus 3.7 and reach 108.3, which is also wrong.

P5. Separate growth from slippage (India). Kadamba's net working capital went from ₹54.4 crore to ₹75.1 crore. At the FY25 cash cycle applied to FY26 volumes it would have been ₹66.5647 crore. Split the absorption, verify the split, and write the title.

Solution. Growth explains 66.5647 − 54.4000 = 12.1647 crore and the longer cycle explains 75.1000 − 66.5647 = 8.5353 crore. They sum to 20.7000, the full absorption, which is the check. The cycle share is 8.5353 / 20.7000 = 41.2333%. Title: Working capital absorbed ₹20.7 crore, of which ₹8.5 crore is a 9.7-day longer cash cycle rather than growth. The page exists for that distinction. Growth-driven absorption is the cost of a bigger business, and slippage is a collections and inventory problem with an owner. A title saying only that working capital rose leaves the board unable to tell which conversation to have.

P6. Choose the number and state the basis (US). Northsound's customer-acquisition payback is 19.2982 months falling to 14.6321 on a gross-margin basis, and 14.6667 falling to 11.4130 on a revenue basis. The policy limit is 18 months. Give the two absolute changes, the FY26 gap between the bases, and write the title.

Solution. On the gross-margin basis the change is 14.6321 − 19.2982 = −4.6661 months; on the revenue basis it is 11.4130 − 14.6667 = −3.2536 months. The FY26 gap between the two bases is 14.6321 − 11.4130 = 3.2191 months. Title: CAC payback fell 4.7 months to 14.6, on a gross-margin basis, against an 18-month policy limit. The level leads because the decision is a policy test and the policy is written as a level. Next comes the change, as evidence in months that a reader can subtract off the two bars. And the basis is stated because two defensible bases differ by 3.2 months on the same spend. Headroom against the limit is 18.0000 − 14.6321 = 3.3679 months.

P7. Rewrite a hedge (India). Kadamba's FY27 plan mix at FY26 channel rates gives a gross margin of 39.6981%, against FY26's 40.6505%. If the quick-commerce channel margin rises from 30.0% to 34.0%, the plan margin becomes 40.5786%. The current slide says that management continues to monitor platform fee pressure. Compute the two drags, value the fee move, and rewrite the sentence as a claim with a condition.

Solution. The drag on the plan mix is 39.6981 − 40.6505 = −0.9524 points, and with the improved channel margin it is 40.5786 − 40.6505 = −0.0719 points. Four points of platform fee are therefore worth 40.5786 − 39.6981 = 0.8805 points of group gross margin on the FY27 mix. Rewrite: quick commerce at a 30.0% channel margin costs 0.95 points of group gross margin on the FY27 plan mix, and a four-point fee reduction recovers 0.88 of them. The original sentence can't be wrong under any outcome, so it carries no information and no instruction. The rewrite can be wrong in three named ways, and that's what makes it worth a director's attention. It also hands the commercial team a target expressed in the unit the board cares about.

P8. Find the missing page (US). Five titles run: ARR grew 32.9% to $117.2m; net revenue retention reached 112%; CAC payback fell to 14.6 months; the FY27 plan targets $150m; the ask is $24m of sales and marketing. The gross margin is 78.0% and payback is 14.6321 months. Name the missing page and compute what it must show under net retention of 112.0% and of 106.0%.

Solution. The missing page bridges the ARR base to the target and prices the gap. At 112.0%, retained and expanded ARR is 117.1840 × 1.120 = 131.2461, so new-logo ARR of 18.7539 is needed, which is 1.9235% above FY26, and the spend is 18.7539 × 0.780 × 14.6321 / 12 = 17.8366. At 106.0%, retained and expanded is 124.2150, new-logo ARR of 25.7850 is needed, which is 40.1357% above FY26, and the spend is 24.5237. The difference is 6.6871 million. Title for the missing page: The $24m ask is sized for net revenue retention falling to 106%; at 112% the plan needs $17.8m. Without it the board approves a number whose assumption has never been stated, and the deck's own retention page turns into decoration without anyone noticing.

P9. Cut a title to fit (India). A template runs 28-point titles on a 13.333-inch slide with half-inch margins, giving 63.4269 characters a line. A draft title runs 178 characters. Give the line count, cut it to fit, and give the line count of the cut version, which runs 88 characters.

Solution. 178 / 63.4269 = 2.8064 lines, so it wraps onto a third line and has to be cut. The two-line ceiling is 2 × 63.4269 = 126.8537 characters. The cut version at 88 characters is 88 / 63.4269 = 1.3874 lines, which fits with room to spare. Cut in the fixed order: throat-clearing first (Analysis of ... shows that), redundant scope second (the strap already says FY26 gross margin by channel), the competing second number third, and only then words that carry meaning. The verb and the number survive, because a title cut to fit by deleting its number has been made shorter and useless.

P10. Overrule or fix (US). The machine's title rule flags three pages. Page 1: a divider titled The FY27 plan, no verb. Page 2: a title claiming a 167.9% expansion contribution, a figure absent from the underlying table. Page 3: a title saying payback is 14.6 months over a chart whose bars are labelled 15 and 19. Decide overrule or fix for each, with the reason, and compute the figure page 2 needs.

Solution. Page 1 is an overrule. A divider in a deck that really does hold two chains has no claim and shouldn't have one. The flag is the machine applying a rule outside its domain. The divider's title should still name the switch, though, and a bare topic won't do. Page 2 is an overrule only if the components are on the page. The figure is 6.7143 / 4.0000 = 167.8571%, derived from the expansion change and the net retention change. If both of those appear as labelled elements, the reader can check it without leaving the page and the overrule is sound. If they don't, the fix is to add them. Page 3 needs a fix, and overruling it would be wrong. The mismatch is real. The title carries one decimal and the exhibit carries none, so a reader can't verify 14.6 from bars labelled 15 and 19. With the policy limit at 18 months, the decimal is where the verdict lives. Every overrule gets one written line of reason, so a reviewer can see you took a decision and didn't just ignore a check.

Applied mini-project

Take a supplied twelve-slide deck, rewrite every title, run the read-only-the-titles test, and submit the before and after title lists with a cut list and reasons. Budget about 100 minutes. You don't rebuild any chart or change any number, and that constraint is what makes the exercise work.

The deck. Corbin Outdoor Inc (synthetic) sells outdoor equipment in the United States, direct to consumers and through wholesale. Its FY26 category review runs twelve slides with these titles, in order: Corbin Outdoor FY26 Category Review; Agenda; Our Categories; Revenue Performance; Category Detail; Gross Margin; Marketing Investment; DTC and Wholesale; Returns; Inventory; Operating Results; Next Steps.

The fact sheet. Every number the deck's exhibits contain. Revenue is net of returns.

Marketing spend was $24.5m in FY25 and $31.0m in FY26. Other operating expense was $46.8m and $52.4m. Inventory was $33.5m and $41.0m. The returns rate was 11.8% and 14.2%. Direct-to-consumer was 49.0% of revenue in FY25 and 58.0% in FY26. All figures in $m.

Step 1: compute before you write (about 25 minutes). Produce the numbers the titles will carry. Revenue growth. Category growth and FY26 shares, which must sum to 100. Gross profit and gross margin for both years. The mix-versus-rate decomposition of the margin change, with the two effects summing to the total. Total operating expense, its growth and its ratio to revenue in both years. Operating income and operating margin, and the bridge from the gross margin change to the operating margin change. Inventory days on cost of goods sold. Every title you write afterwards has to come from this sheet.

Step 2: rewrite every title (about 30 minutes). One claim per page, with a subject, a finite verb and the number that makes it checkable. Write each title from the body underneath it, and ignore what you remember the deck saying. Keep each inside your template's two-line ceiling, and measure that ceiling. Don't guess it.

Step 3: run the read-only-the-titles test (about 15 minutes). Hide the bodies, read the twelve titles in order, and write the argument they make in four sentences. Then mark every gap, leap and repeat. There's at least one gap in this deck, and finding it is most of the mark.

Step 4: produce the cut list (about 15 minutes). For every page that still has no claim, write the reason it goes and the question it answered, and say where the material lands: cut, appendix, or merged into another page. A cut without a stated cost isn't a decision.

Step 5: name the pages that must be added (about 15 minutes). Each gap you found in step 3 becomes a page. Write its title and state the exhibit it needs, using only the numbers on the fact sheet. If a required number isn't on the sheet, don't invent it. Say plainly that it's missing, and state what you'd ask for.

What a strong submission finds. A strong answer finds two things an adequate one misses. First, gross margin rose 1.0668 points and almost all of it is mix, since the FY26 weights at FY25 rates give 51.0652% against FY25's 50.0745%. Pricing did very little. Second, the gap. The deck never connects the margin gain to the marketing spend, and the operating margin moved 0.0264 points. So the gross margin gain of 1.0668 points was spent by an operating-expense ratio that rose 1.0404 points. A title stack that reports a margin win and a marketing investment without connecting them leaves the board with two facts and no argument.

Deliverables. Three artefacts. The before and after title lists, side by side, in one table. The cut list with a reason and a destination against each line. The added-page list with titles and required exhibits. Plus the computation sheet from step 1, since every number in every title has to be traceable to it.

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

Any row whose evidence is absent scores zero regardless of how well the submission reads. A rewritten stack with an arithmetic error in a title is worse than the label titles it replaced, because a label can't be wrong and a claim can.

Reading & resources

Core (do these).

• Barbara Minto, The Pyramid Principle: the source of horizontal and vertical logic and of the answer-first habit. Read the sections on the two logics and on the introduction structures. The material on ordering is what turns a title stack from a list into a sequence. [Paid] [Intermediate]
• Gene Zelazny, Say It With Charts: written inside a consulting firm and still the clearest statement of the idea that the message chooses the form, with an appendix on chart titles that is directly this craft. [Paid] [Beginner]
• Cole Nussbaumer Knaflic, Storytelling with Data: the most useful single book on the surrounding craft, and its treatment of what she calls the takeaway title is the closest published version of the action title taught here. [Paid] [Beginner]
• Jean-luc Doumont, Trees, Maps, and Theorems: short, dense and opinionated on the idea that a slide's headline should state the message. Read it for the compression. Almost every page is a rule you can apply the same afternoon. [Paid] [Advanced]

Going deeper.

• Stephen Few, Show Me the Numbers: the reference on table and graph design, and the place to go for the reasons behind the rules. Its chapter on titles, labels and legends is the one to read against the exhibit-versus-text distinction. [Paid] [Intermediate]
• Alberto Cairo, How Charts Lie: the best short treatment of how a true chart supports a false headline, which is the title-body contradiction from the outside in. [Paid] [Beginner]
• Nancy Duarte, DataStory: written for the audience-facing side of the same problem, and useful for the recommendation slide that a title stack has to end on. [Paid] [Beginner]
• Edward Tufte, The Visual Display of Quantitative Information: read it for the standard of care. The field has argued with some of its prescriptions since. [Paid] [Intermediate]
• Garr Reynolds, Presentation Zen: the case for restraint on a presented slide, which is the constraint the two-line rule expresses in arithmetic. [Paid] [Beginner]

Free, and worth the hour.

• The US Securities and Exchange Commission, A Plain English Handbook: free from the SEC's own site, written for disclosure documents, and the cheapest available improvement to anyone whose sentences are long. Its treatment of the informative heading transfers directly to titles.
• The US federal plain-language guidelines at plainlanguage.gov: free, and the section on writing headings that answer the reader's question is the government version of the same rule.
• Your own last three decks: extract the title stacks into a text file and read them cold. Three stacks take twelve minutes and will tell you more about your own habits than any book on this list.

Do this, not just read. Take the deck you're most proud of. Hide the bodies, read the titles to a colleague who has not seen it, and ask them to tell you the argument. Write down what they say before you correct them. The gap between what they said and what you meant is your entire backlog for titles and storylines.

Teach it back & journal

Feynman prompt. Explain to a colleague who builds decks, and has never thought about titles, why a stack of nineteen words can be worthless while a stack of a hundred and fourteen carries an entire board meeting. Don't use the words design, polish or storytelling. Use the arithmetic. Nine label titles read in about five seconds and leave a director knowing what she knew before she opened the file. The rewritten claims read in about thirty-one seconds and hand her the growth, the margin cause, the working-capital split, the covenant test and the decision. Twenty-six seconds of attention bought the argument. Then make the harder point. The rewrite found three things the labels were hiding: a missing page between the mix slide and the margin slide, a missing page at the end where the ask should have been, and a cost title that was about to ship a false claim. Total operating cost grew 21.4% against revenue's 21.9% while two of its four lines grew above 24%. If your colleague can tell you, unprompted, why a false title is worse than a label title, you've taught the idea itself and not only the arithmetic.

Journal prompt. Open the last deck you built and the last one somebody sent you. Extract both title stacks into a text file, hide the bodies, and read each one cold. For each stack, write the argument it makes in four sentences, or write that it makes none. Then count the gaps, the leaps and the repeats. Now go one level further into your own deck. Pick the three pages you spent the most time on and ask, page by page, whether the title says what the body found, or whether it names the exhibit. If it names the exhibit, write what the finding actually was, and notice how it feels to put your own conclusion in the largest type on the page. That discomfort is worth sitting with. It's the real obstacle here, and it isn't a skill problem. Under-claiming is safer in the moment and expensive over a career. An analyst whose pages never assert anything is never wrong, and is never the person whose view gets asked for either. Finish by writing one sentence about the pattern you found across both decks. Then write one habit small enough to survive a bad week: extract the titles into a text file before you touch a single chart, and refuse to build until the list makes an argument you'd defend out loud.