Reference · Slide craft

Slide Craft — building one slide

Load when one slide — action titles & the flip test, lead-with-message bullets, one-message-per-slide, frameworks, 2x2, callouts, exhibit & layout choice, gestalt/grid.

Part of the Consulting Deck skill · loaded on demand from SKILL.md

Methods for constructing and fixing a single slide once the storyline is set. The unit of a consulting deck is one message, one exhibit, one action title.

Action titles (action-title-writer)

Convert a topic-label headline into an action title: one full sentence stating the slide's single takeaway. Read in sequence, the titles form the argument's spine; the body only proves the title.

  1. Find the one message — what the reader must take away in five seconds. Two competing ideas → split the slide.
  2. State a claim, not a topic. "Customer churn" (topic) → "Churn concentrates in the first 90 days, where onboarding fails" (claim).
  3. Full sentence with a verb carrying the so-what — drives, erodes, concentrates, doubles, shifts, fails to. A title with no verb is a label.
  4. Quantify when the body supports it — carry the headline number; never invent one not on the slide.
  5. One line, ~8–14 words / under ~90 chars. Cut hedges and filler ("various", "key", "we believe", "in order to") and throat-clearing openers.
  6. Title–body agreement: the body proves exactly what the title claims and nothing else. If the chart shows correlation and the title claims causation, soften the title or add evidence.

Action-title audit / flip test (action-title-audit)

A review pass over the title layer alone on a near-final deck.

  1. Extract the title strip — every title in order, divorced from bodies.
  2. Assertion not label: flag any title with no verb / no point of view.
  3. Title–body agreement: flag overclaim (chart doesn't support it), underclaim (chart shows more), mismatch (title and exhibit are about different things).
  4. Horizontal logic: the strip must start from the question, follow one thread, land on the recommendation, and match the storyline. Flag non-sequiturs, backward steps, missing links, misplaced slides.
  5. Level consistency: dividers frame, content slides assert.
  6. Output the strip with a per-title verdict (pass / rewrite / move / cut) + suggested rewrites, and the flip-test verdict: does the deck tell its whole story on titles alone?

Lead-with-message bullets (lead-with-message-bullets)

  1. Extract the point of each bullet (often buried at the end after setup) and move it to the front: "After reviewing the data… we concluded coverage is too thin" → "Coverage is too thin: data shows wide variance."
  2. Front-load — the first 3–4 words carry the point; caveats follow.
  3. Parallel structure — every bullet in a group starts the same grammatical way (all verbs, or all noun phrases).
  4. MECE — merge duplicates, split bullets smuggling two points.
  5. Prune to the load-bearing few (rarely more than 4–5 top-level); detail goes to appendix or the spoken track.
  6. Flatten — avoid sub-bullets beyond one level; deep nesting means the slide is doing too much.
  7. Tighten — cut filler openers and hedges; keep numbers.

One message per slide (one-message-slide)

  1. Count the messages — if the title needs "and" or a semicolon, it's two messages; a table supporting three conclusions is three.
  2. Split or cut: essential → its own slide; interesting-but-not-load-bearing → appendix or delete. Not every true thing belongs in the main deck.
  3. One exhibit per message; the other chart moves to backup.
  4. One single-claim action title per slide (no "and").
  5. Re-sequence the split slides; often the overload existed because order wasn't decided.
  6. One focal point per slide; white space is a feature.
  7. Check count vs meeting (~1–2 min per content slide); if it overflows, cut harder.

Slide density (slide-density-balancer)

Tune each slide to the one-message standard: enough to prove the title, not so much the point is lost. Split overloaded slides; merge thin ones (a slide holding one sentence usually belongs combined); move supporting tables/methodology to backup. Glance test: point grasped in 5–10 seconds. Tune to setting — read-ahead tolerates more density than a presented slide. Re-check the title chain after splitting/merging.

Exhibit selection (exhibit-selector) & layout (slide-layout-picker)

Match the message to the form that proves it fastest.

Name the relationship, then pick:

  • Comparison across categories → sorted bar (horizontal if labels long).
  • Composition at a point → stacked or 100%-stacked bar; avoid pies beyond ~3 slices.
  • Composition that builds/bridges → waterfall.
  • Change over time → line (many periods) or column (few).
  • Distribution → histogram / dot plot.
  • Correlation → scatter, with a 2x2 overlay to classify.
  • A structured set of ideas (not numbers) → a framework layout (2x2, value chain, process flow, tree), not a chart.

For a layout, classify the message shape — sequence/process, comparison, categorisation/matrix, composition, relationship/driver, single-point — and map to a layout (stepped row, column-per-option grid, 2x2, stacked/waterfall, driver tree, one big number). One structure per slide. Specify the slots (columns, axes, criteria-rows) as a wireframe, pre-constrain any chart (a comparison wants bars, not a pie), and sketch the reading path (top-left → across → takeaway). Every exhibit needs a comparison anchor (target, benchmark, prior period) and one planned highlight. Sanity-check: could a reader draw the opposite title from the same exhibit? If so, the form is wrong. (Chart specifics → references/chart-craft.md.)

Framework slides (framework-slide-designer)

For a structure rather than a number (positioning 2x2, value chain, maturity model, roadmap).

  1. Match archetype to relationship: two crossing dimensions → 2x2; linear sequence → process/value chain; hierarchy → pyramid/tree; parallel components → columns/wheel; maturity → stairway.
  2. Define axes/stages precisely — ambiguous axes are why a 2x2 fails; name all four quadrants or all stages (verbs, MECE).
  3. Place real content, not just labels — one line of substance per box.
  4. One reading path (L→R for flows, bottom→top for pyramids, a marked start quadrant).
  5. Highlight the conclusion in the accent (the quadrant you're in, the broken stage).
  6. Balance the geometry — equal box sizes, even gaps, consistent corners, grid-aligned; lopsided reads as careless thinking.
  7. Limit arrows to genuine flow/causation. Add a client-specific callout so the framework isn't generic.

2x2 matrix (two-by-two-matrix-builder)

  1. State the decision the matrix serves (what to fund/keep/target) — it disciplines the axes.
  2. Two independent axes that matter and aren't correlated (value vs effort, attractiveness vs ability-to-win, impact vs feasibility). If they move together the matrix collapses to a diagonal and adds nothing.
  3. Define the scales — what low/high mean and the rough threshold (concrete: "payback under 12 months vs over").
  4. Place items honestly by real values; if everything lands top-right, the axes don't discriminate.
  5. Name each quadrant by its action (do-now / drop / quick-wins / big-bets), not by coordinates.
  6. Highlight the recommended set in the accent.
  7. Guard against false precision — place items in zones, not exact coordinates, unless the axes are genuinely quantified; flag contested placements.

Callouts & annotations (callout-and-annotation-designer)

Direct the reader to the insight on an exhibit. Find the insight point (spike, gap, outlier, crossover). Write the annotation as a claim ("Margin crosses zero in Q3"), one short line. Thin leader line to the exact element — no thick arrows, glow, or region boxes. One annotation per exhibit (two if the slide makes two points; more → split). One accent colour. Place in white space, not over data; align to grid. A fresh eye should reach the intended conclusion in a few seconds.

Gestalt & grid (gestalt-alignment-and-grid)

Make a slide read as designed, not assembled.

  1. Lay a grid — margins, title zone, 2/3/4 body columns; every element snaps to a line.
  2. Align to shared edges — left-align text blocks to one vertical, top-align rows. Misalignment is the #1 tell of a rushed slide.
  3. Proximity — related items close, unrelated apart; a label sits next to its element.
  4. Similarity — same-type items get the same size/colour/style (three options look identical so they read as peers).
  5. Common region — group by shared white space or one hairline border, not by boxing each item.
  6. Hierarchy via position/size/accent mirrors the logical structure.
  7. Equalize spacing — identical gaps between repeated elements.
  8. One type scale — a small set of sizes used consistently.
  9. Squint test — groups and hierarchy still obvious as blurry shapes.

Before/after redesign (before-after-redesign)

The slide-rescue pass chaining the above over one weak-but-correct slide: read it cold and note the diagnosis → recover the message as an action title → inventory all content (nothing real gets lost, only moved) → choose the right exhibit → restructure MECE on a grid → strip and direct (remove junk, add the one highlight + annotation) → apply color/icon discipline → write a before/after note (what changed and why) → verify nothing was lost against the inventory.