How to make perfect slides, in less time than ever before

3 AI elements guaranteeing a deck that enhances your credibility at ExCo

It is 9pm on a Tuesday. The board pack is due Friday. Your GR lead has built 22 slides, and you are looking at them with a glass of wine you are not enjoying.

Three people have touched this deck. You can tell, because it is written in three voices and set in three fonts. The community lead’s slides are dense and careful. The media manager’s are punchy and thin. The graduate’s have a stock photo of a sunrise behind a wall of text. Slide 7 says the consultation programme is on track. Slide 14 says it is the biggest risk in the pack. Nobody has read the thing end to end, which is why nobody has caught it.

You close the file. You open a blank deck. You start again. You will be finished around 1am, and the pack will be good, because you will have rebuilt it yourself, in one voice, with one set of fonts. You have done this for four Tuesdays running.

This piece is about why that 1am rebuild happens, and how to stop doing it. The short version is that a good board deck needs three things, and most corporate affairs functions have none of them set up properly. You need the thinking. You need a design system. And you need a capable LLM to take the first two and produce the deck, so that no one senior ever has to push a pixel again. I am going to take them in that order, because the order is the point.

This is the second piece in a short series on using AI in corporate affairs without producing rubbish faster. The first was on the thinking. If you have not read it, read it before this one, because this piece assumes it. It is here.

First, the thinking (and the step everyone skips)

That issue laid out how to structure a strategic argument the way the executives around the table do it: Situation, Complication, Question, Answer, and then a pyramid of three to five supporting pillars under the Answer. That is how you build the logic of a board paper. It is the difference between being in the room and being handed the decision to communicate after the room has emptied.

I left one step out of that piece, because it was already long. It is the step that connects the pyramid to the deck, and while every corporate affairs practitioner is good at it, each individual does it slightly differently. It is called storylining.

The pyramid is the logic. The storyline is the order you tell it in. They look similar, because both are built from the Answer and the pillars, so people assume they are the same artefact. They are not, and the difference is what makes a deck readable.

You build the pyramid top-down: Answer first, then the pillars that hold it up. That is how you assemble the thinking. But you do not always deliver it in that order. There are three patterns the consulting firms drill into you.

• Top-down. Lead with the Answer, then the pillars. “We should close Site C over 18 months through three workstreams. Here is why each one is right.” This is the default for a board that has asked for your recommendation and wants the bottom line first.

• SCQA. Open with the Situation, Complication and Question, deliver the Answer, then walk the pillars. This is for an audience that does not yet share your view of the problem and needs the why before they will accept the what. The default for a sceptical executive committee, or a first meeting on a hard topic.

• Inductive. Walk the evidence first, arrive at the Answer last. This is for a room that holds a strong prior against your Answer and would reject it on sight. Use it sparingly. It asks the audience to sit in suspense, which is expensive.

Whichever order you choose, every slide title obeys one rule: it is a complete assertion, not a label. “Workforce transition: six-month notice with retention” is a title. “Workforce” is a label. The test is simple. Read only the titles, top to bottom, and they should tell the whole story without a single body bullet. For a site-closure board paper, the title sequence reads like this:

Site C is our highest-cost producer and has missed its benchmark for six straight quarters.

The competitors who closed comparable sites are now in court and short an 11-month approval.

Close Site C over 18 months through three sequenced workstreams to protect both the NPV and the Site D pipeline.

Workforce transition cuts class-action risk by making the exit predictable and supported.

A CEO-signed community covenant removes the consultation grounds our competitors were sued on.

Early regulator engagement protects Site D’s approval window.

Recommended: endorse the 18-month closure and authorise the CEO to sign the covenant before announcement.

Read those seven lines on their own. They are the titles of each individual slide in the board pack. Everything else on the slides is evidence for a line that already makes the case. That is a finished storyline, and it is the actual handover to the deck. Once you have it, the writing of the deck is done. What is left is the production: turning seven argued lines into seven designed slides. Which is exactly where the 1am rebuild lives.

Second, the design system (the thing that ends the 1am rebuild)

Go back to the three voices and three fonts in that opening deck. That is not a thinking problem. The thinking might be fine. It is a production problem, and it has a specific cause: three people authored slides with no shared system underneath them, so each one reached for their own fonts, their own spacing, their own version of the logo, their own idea of what a chart should look like. The deck is inconsistent because nothing made it consistent. You rebuild it at 1am to put it back into one voice and one look, by hand, slide by slide.

A design system is the thing that makes that rebuild unnecessary. It is the house style, written down in a form a tool can use: the exact colours with their codes, the fonts and their sizes, the spacing, the logo in its correct versions, the favicon, the drop shadows or the deliberate absence of them, the chart styles, the rule about how the one accent colour gets used. Everything you can see in a branded asset, specified once, in one place.

Here is the part that matters for a corporate affairs function, and it goes well beyond slides. Say you have to announce that site decision. You do not produce one asset.

You produce:

• a board deck, and

• a notice for the TV screens in the mess, and

• an email header for the COO to send it under, and

• a SharePoint page people click through to for the detail.

Four assets, four formats, four different people building them, probably across two days. Without a system, every one of those is a fresh negotiation with the brand guidelines: is this the right orange, should the logo go at the top or the bottom of the site notice, are these the heading fonts, geez the deck has rounded corners while the email header has straight edges, oh my God I earn far too much money to be caring about this. With a system, all four inherit the same colours, the same fonts, the same logos, the same everything, because they are all drawing from the same specified source.

What that buys the corporate affairs person is the most valuable thing they have, which is attention. You get to spend your thinking on the content. Is the message right. Is the sequence right. Will the COO’s email land with the workforce. You stop spending it on whether the box is a few pixels too wide, or whether someone has used last year’s logo, or whether the orange on the SharePoint page is the same orange as the deck. The system holds the consistency so you do not have to.

This is the piece most functions are missing, and it is missing because nobody owns it. The thinking has an owner: you. The assets have owners: your team. The system underneath the assets has no owner, so it does not exist, so every asset improvises its own, so nothing matches, so you rebuild at 1am. Building the system once, and owning it, is what ends that.

Third, a capable LLM (the part that ends the pixel-pushing)

Here is the third thing, and it is the biggest unlock of the three, because it removes the lowest-value work in the entire process.

Give a capable LLM the storyline and the design system together, and it produces the deck. The thinking tells it what every slide has to say. The system tells it exactly how every slide has to look. There is almost nothing left for a person to do. No formatting charts. No nudging a box three pixels left so it lines up with the one above it. No hunting for the right shade of the brand colour. No checking the logo is this year’s version. The action title is already written, because you wrote it in the storyline. The colours are already right, because they came from the system. The chart is already styled, because the system styled it.

Think about what that pixel-pushing actually is. It is soul-destroying work. Moving boxes around a slide at 11pm, matching one shade of orange to another, lining up four logos that someone pasted at four different sizes. It is also work that is completely beneath someone paid $300,000 a year to read the political weather and keep the company out of trouble. And if the business ever saw you doing it, even for one click, they would quietly conclude you do not have your act together. The Head of Corporate Affairs, on the night before the board meeting, manually resizing text boxes.

That is exactly what happens when five people have a hand in one deck. The person who clears it ends up doing the cosmetic clean-up by hand, because someone has to, and they are the last set of eyes. So the most senior person in the chain spends the last two hours not on whether the recommendation is right, but on whether the boxes line up. The strategic thinking gets the time that is left over after the pixel-pushing, which on a Tuesday night is none.

A capable LLM, handed a good story and a real design system, takes the pixel-pushing away entirely. You concentrate on the message, which is the part that matters and the only part that needs you. You get a board-ready deck, in the correct brand guidelines, right the first time. The prize is not “AI writes my deck in ten minutes”, which produces the three-voices mess. The prize is that the thinking and the look are both settled before the deck exists, so building it is no longer a job a senior person has to lower themselves to do.

The way you make an LLM do this is two prompts, run in order, with one rule between them: the first prompt builds the design system and writes no content, the second fills it and changes no design. The first does the look. The second does the words. Neither does both, because asking one prompt to do both is what produces four fonts under a wall of text in the first place. I have put both prompts at the end of this piece, with a step-by-step guide to filling them in for your own company.

Then I tested which LLMs could actually do it.

Five tools, the same brief, in BHP’s colours

I gave five AI tools the same job, in a house style everyone in this sector knows on sight: BHP’s. One orange accent used once per slide, one typeface, charcoal text, a lot of white space. Inside your own company you would hand the tool your real brand guidelines and template. I used BHP so that you can look at a screenshot and judge for yourself how close the AI got, without taking my word for it.

I ran it in two phases. First the design system, empty. Then the same site-closure board paper, filled into it.

Phase one, the design. Claude Design returned a complete, reusable system: the colour codes, the type rules, the spacing, two reusable components, twelve empty slide layouts, a master slide documenting the grid, and a small instruction file to produce the next asset in the same style. It held the brief exactly. One orange, one typeface, the restraint. This was the moment the whole argument landed, because what came back was the system itself, the thing you build once and reuse across the deck, the mess notice, the email header and the SharePoint page. Codex returned a clean, editable deck, which was good, except it quietly leaked a second font and a second orange in from PowerPoint’s own defaults, which is the three-voices problem arriving through the back door. Gamma and Copilot could not hold the brief.

Phase two, the content. Claude Design won again, and the exhibits are why. It built a real cost curve, the target site in orange above the benchmark line. It built a timeline showing how an 11-month delay at a competitor would consume the entire approval window for the next project. A covenant table. An engagement schedule. Argued exhibits, not boxes with “chart goes here” inside them. It was not perfect, and the flaws are worth showing rather than hiding: on three detail slides the third bullet ran into the footer, and one slide had a placeholder caption sitting over the title. Both are a five-minute human fix. Both are also a tidy demonstration of why you keep the two prompts separate, because the design held and the writing mostly flowed, but a couple of slots had been sized before the real words arrived. You can see exactly which job each flaw belongs to, which is the point of doing the jobs separately.

Claude Design's Storylined Deck in the style of BHP

Codex also produced a board-grade deck. Its exhibits were more schematic, and the same defaults leaked into the file again. Gamma could be made to work, but only by rebuilding the whole system as a Gamma “theme” by hand, which took 20 to 30 minutes and was harder to adjust afterwards than handing an LLM the written specification. That is the tell worth keeping. A system is valuable because it travels between tools and assets, and Gamma is the point where travelling broke.

I am not telling you to go and buy Claude Design. These tools change every month, and by the time you read this the order may have shuffled. The durable part is the method: do the thinking, build the system, and hand both to whichever LLM does the production best this month. The pixel-pushing goes away. Your attention goes back to the message.

Build the design system once

This is the part you do a single time and reuse forever. Here is how I built the BHP one, in Claude Design, the tool that won my test. The two prompts work in any capable LLM, so if you use a different one the clicks will differ, but the shape is the same: give it your brand assets, give it the design prompt, let it build the system, then export it so you can reuse it.

1. Go to claude.ai/design.

2. Click Design Systems in the header.

3. Click Create, then Create here.

4. Fill in your company name.

5. Click the button to add fonts, logos and assets.

6. Upload everything you would ever want in an asset: your brand guide, a PowerPoint template, your logos in colour and in white, any company photography you are cleared to use.

7. In the notes field, paste your tailored version of Prompt 1 (below). The Try This Week section at the end gives you a prompt that writes the tailored version for you from your brand guide, so you do not have to edit it by hand.

8. Wait a few minutes while the system builds.

9. Open the finished design system, click Share (top right), open the Export tab, and download the Project archive. That archive is your reusable system. Keep it. Next week’s paper starts from here, not from a blank deck.

Prompt 1 — build the design system

This produces the master and the twelve empty layouts. It writes no content. The colour-and-type block is set to BHP’s house style as the worked example. Replace that block with your own (or use the Try This Week prompt to do it for you).

You are designing an executive SLIDE MASTER and an EMPTY 12-slide SKELETON for a
board paper produced by a senior corporate affairs function. You design the LAYOUT
SYSTEM and the empty skeleton ONLY. You do NOT write slide content, headlines, or
data. Where content would go, leave a clearly labelled placeholder.

[WHAT THIS IS FOR]
Asset: a reusable board-deck master + 12 empty slide layouts a Head of Corporate
Affairs can fill for any board-level recommendation.
Audience: CEO, executive committee, and board.
Strategic job: make a structured argument legible at a glance — situation,
complication, recommendation, evidence, action — with board-pack discipline.

[HOW IT SHOULD READ]
Reads as: authoritative, measured, board-ready.
Never: glossy, "designed", startup-slick.
One line a stranger could repeat to a designer: it should look like a paper a serious
board would accept, not a marketing deck or a consultancy template.

[REFERENCES — borrow the behaviour, do not copy the brand]
1. Board risk dashboards and investor-relations fact packs, for decision-led
   hierarchy: issue, implication, evidence, owner, next action on every page.
2. Financial Times / Economist editorial discipline, for restrained charts, one idea
   per slide, source-labelled evidence, generous white space.
3. AEMO / Infrastructure Australia system diagrams, for sober, low-ornament exhibits
   that read in a boardroom.

[ANTI-REFERENCES]
Do NOT look like: a generic SaaS deck, blue-and-white consultancy boilerplate, an ESG
gloss report, a LinkedIn-guru carousel, or a public-sector PDF.
Forbidden: gradient fills, drop shadows, glassmorphism, stock photography, decorative
icons on every bullet, centred-everything hero slides, fake dashboard chrome, two
competing accent colours on one slide.

[COLOUR AND TYPE — replace this whole block with your own brand guidelines]
Colour (use these exact codes):
- Primary accent: BHP orange #E35205 (Pantone 166 C). ONE accent only, used for
  emphasis or signal, never as a fill behind text. Roughly one orange element per slide.
- Text: near-black charcoal #1A1A1A for all body and titles.
- Labels and source lines: warm grey #6B6B6B.
- Background: white #FFFFFF, or very light warm paper #FAF9F7 if supported.
- Rule lines: light grey #D9D6D2 at hairline weight.
- Status colours, used only where a slide genuinely shows status, always with a word
  not colour alone: amber #C77A00 open, green #2E7D5B on track, red #B3261E high risk.

Type (name a real substitute if a font is unavailable):
- Titles: Arial Bold (or Helvetica Neue Bold), sentence case, left-aligned, 28–30pt,
  two lines maximum.
- Body: Arial Regular, 16–18pt, left-aligned, no bullet over two lines.
- Labels, sources, page numbers: Arial Narrow, 10–11pt, warm grey.
- Exactly ONE typeface family across the whole deck.

[THE MASTER — define and render]
- Title zone: top, one action-title line, room for 12–15 words.
- Body zone: one message per slide, three supporting points maximum.
- Exhibit zone: one chart, diagram or table.
- Footer: source bottom-left, page number bottom-right, confidentiality marker.
- One consistent grid and margin system on every layout.

[THE 12 EMPTY LAYOUTS]
1 Cover. 2 Executive summary (the answer, up top). 3 Situation. 4 Complication.
5 Governing question. 6 Recommendation (with three pillar labels). 7–9 Pillar detail
(the SAME layout three times, to prove the system repeats). 10 Evidence (one chart,
action title, one "so what" line). 11 Implementation (timeline). 12 Closing
recommendation and decision sought.

[OUTPUT]
Render the master and all 12 empty layouts. Keep every placeholder generic:
"[ACTION TITLE]", "[supporting point]", "[exhibit: chart type]", "[source]". Note the
export formats available (web, PPTX, Canva).

[GUARDRAIL]
Design only. The first real word of slide content you write is a failure of this task.

Produce the deck

Now the system exists, producing the deck is the easy part, and it is the part that used to cost you the night. Five steps.

1. Open your LLM of choice.

2. Add the design-system archive you downloaded earlier.

3. Add Prompt 2 below, with your own storyline pasted into it. (Your SCQA, pyramid and storyline from the first section of this piece. The titles are the deck; the prompt just sets them in the system.)

4. Press enter.

5. Download the slides and spend five minutes tweaking.

One rule on that last step. If the tweaking is taking you more than ten minutes, the problem is not the deck. It is the storyline. Stop, go back, fix the titles so they carry the argument on their own, and run it again. The deck is only ever as good as the story you fed it, and ten minutes of fiddling is the tool telling you the story is not finished.

Prompt 2 — fill the system with your storyline

Run this against the design system. It writes the content and changes nothing about the look.

You are an executive communications specialist trained in the Pyramid Principle, SCQA,
and action-title slide writing. You FILL the existing 12-slide skeleton with content.
You do NOT change the design, layout, fonts, or colours. You do NOT add or remove
slides. You work only from the thinking provided below. You do not invent data.

[THE THINKING — paste your own SCQA, pyramid and storyline here]
Situation: [your situation, in 2–4 plain sentences]
Complication: [the change that forces the decision now]
Question: [the single question the paper answers]
Answer / governing thought: [your recommendation in one or two sentences]
Pillars (3): [pillar 1 label + one line] / [pillar 2] / [pillar 3]
Evidence to use: [the specific facts, figures and precedents — and only these]
Storyline (the title sequence): [paste your 6–7 action titles in delivery order]

[RULES FOR FILLING]
- Action titles are complete assertions, not labels. "Workforce transition cuts
  class-action risk" — not "Workforce". Read the titles top to bottom: they must tell
  the whole story on their own.
- One message per slide. Three supporting points maximum. No bullet over two lines.
- Every evidence slide ends with a one-line "So what:" tied to the recommendation.
- The three pillar slides use the same layout, so the system visibly repeats.
- Where the evidence is thin, label it "directional". Never invent a number, a
  headcount, a date or an NPV that was not given to you above.

[OUTPUT]
Fill all 12 slides in place. Keep the master untouched. For each slide give the action
title, the body points, and a one-line description of the exhibit.

[GUARDRAIL]
Content only. If you restyle, recolour, refont or re-lay-out anything, you have failed
this task.

Try this week: a prompt that tailors Prompt 1 for you

The one fiddly part of all this is the colour-and-type block in Prompt 1. You could edit it by hand, copying your hex codes and fonts out of the brand guide. Easier to make the LLM do it.

This week, find your company brand guideline (the PDF your marketing team uses, the one with the colour and typography pages). Open your LLM, attach that PDF, and paste this:

Attached is my company's brand guideline. Read it and rewrite ONLY the
[COLOUR AND TYPE] block of the prompt below so it uses my company's real
details instead of BHP's:

- Replace every colour with my brand's exact hex codes, and say where each
  colour is allowed to be used (which is the accent, which is text, which is
  background). If my brand has more than one colour, give each one a job, so
  the tool does not use all of them at once.
- Replace the fonts with my brand's heading and body fonts and their sizes.
- If the guideline specifies logo rules, drop-shadow rules, or anything else
  visual, add a short line for each.

Leave every other section of the prompt exactly as it is. Then give me the
full rewritten prompt, ready to paste.

[paste the whole of Prompt 1 here]

What comes back is your own Prompt 1, in your own house style. Take it through the steps above: build the system once, keep the archive, then produce decks from it whenever you need one. Build it this week and you never push a pixel on a board deck again.

---

Richard Wilson is the MD of Transformation Partners, a boutique management consultancy helping Heads of Corporate Affairs get a set at the table. He is a former BCG Principal and a former Head of Corporate Affairs at a Nasdaq-listed energy company.