TL;DR. Open with the bottom line, not the build-up. Frame every technical claim as a business consequence (revenue, cost, risk, reputation), use one analogy per concept, and repeat the core message ~7 times. For an AI security talk specifically: map threats onto risk categories the audience already owns (data protection, third-party, compliance, reputation) and translate findings into dollars + a decision the audience must make [1] [5].
The five moves that matter most
| Move | What it looks like | Why it works |
|---|---|---|
| Lead with BLUF | First 2–3 sentences carry situation, complication, and your recommendation [2] | Executives are problem-solvers; they decide whether to keep listening in ~30 seconds [1] |
| Audience is the hero, you’re the mentor | Use Duarte’s Sparkline: toggle between “what is” and “what could be,” ending on a call to action [3] | Empathy frames the talk around their stakes, not your work |
| One analogy per concept, no jargon | “Prompt injection is SQL injection for English” — anchor every term in something they already know [4] | Buzzwords and acronyms make non-technical audiences tune out fast [1] |
| 10-7 rule | Identify the most essential 10% of the message, repeat it ~7 times across the talk [1] | Recall is driven by repetition, not novelty |
| One idea per slide, visual over text | ≤3–4 short bullets, replace bullets with a diagram when systems are involved [1] [3] | Slides are scaffolding, not the script |
AI-security-specific reframing
A non-technical audience doesn’t need the threat model. They need:
- Translation into money. “We found 47 unauthorized AI tools” lands as noise; “$8.4M in exposure via 47 AI tools touching customer PII” lands as a decision [5].
- Mapping to risk categories they already own. Data protection, third-party/vendor risk, compliance, and reputation — AI rarely needs a new bucket [5] [6].
- Practical risks before exotic ones. Employees pasting customer data into public LLMs beats data-poisoning and model theft as your opener — it’s what’s happening this week [5].
- Take a position. Boards want to know what you recommend, not a balanced survey of options [5] [7].
The pre-flight checklist
- Can a stranger state your one-line recommendation after slide 1? If no, rewrite the opener.
- Does every technical term have an analogy or a concrete example within 10 seconds of first use? [4]
- Does every claim end in “so what?” — a cost, risk, or decision?
- Have you cut the slide that’s there because it took you a long time to make? (Sunk-cost slides are the most common reason talks bloat.)
- Watch faces in the first three minutes — confused → slow down; nodding → advance [1].