AI is rapidly reshaping how organisations operate and make decisions. For directors, understanding AI no longer means just approving a digital strategy, it increasingly requires knowing how these systems think, learn and respond. At the core of this is prompt engineering — the art of asking AI the right questions to get useful, reliable answers.
What exactly is prompt engineering, and why should directors care? According to Sonia Petering FAICD, an experienced non-executive director and chair of ASX-listed Vitrafy Life Sciences, it’s less about coding and more about communication. Below she shares how she became a better “prompt engineer”. As with most new skills, it’s all about building muscle memory through repetition.
What are the basics of prompt engineering?
“It’s about how you communicate effectively with the AI tool you’re using, so you get the outputs that assist you. You need to refine the results returned from your prompts. My learning on this is that it is about practising how to craft clear instructions and giving the tool context. Structure your requests of AI in a way that helps AI understand what you really need. The more specific you can be, the better. You can provide examples to help narrow its output and once you’ve got to the output you were after, ask it, ‘How could I have crafted this prompt to be better from the beginning?’ so you can learn from it.” (The skill of prompt engineering will soon become redundant!)
How long are these prompts or questions?
“I tend to start with a detailed prompt and then iterate from there. I’m promoting a more detailed prompt over a less detailed one because once you receive the first response, you can subsequently break up your subsequent queries into further refinement and iterate later responses. I think of it as like briefing a really capable graduate – but one who doesn’t need to sleep. Be clear about what you want and what you don’t want, then keep tweaking your questions until you receive the output you’re after, in the right form, the right tone. For example, you can ask AI for the output it generates to be rephrased to be more optimistic, more sophisticated, more casual, in shorter sentences or the opposite.”
What key things have you learned since honing your prompt engineering skills?
“At the start of this year I did a Quantium AI Board Edge masterclass — Elevating Board Impact with GenAI. They suggested a prompt should cover five key areas. The first is the background context. For example, I’m a company director and I’m looking for some analysis ofcompetitors. The second is the goal. What are you trying to achieve? Do you want a strategic analysis with some priorities, a timeline, a list, a heat map? The third is the perspective you want the answers to take. Do you want it to respond as an adviser, a regulator, an aggressive journalist? Next, the task. Do you want the AI tool to analyse and give recommendations? Do you want it to produce an artefact, a clickable interface such as a spreadsheet, a proposal or a visualisation output? The final key area suggested was putting explicit constraints in your prompt. For example, ‘Provide strategic analysis that balances threats and opportunities, and be sure to consider the commercial or technical constraints of x’.”
What’s the place of prompt engineering in the boardroom?
“I have an important caveat on this. You must not put confidential information into these GenAI tools. You must only use publicly available information. I’ve heard stories of directors uploading board papers without permission into generative AI Large Language Models (LLMs) with no privacy settings. That’s scary stuff.
With that caveat, it can be useful in a strategic decision-making context. You can be reading your papers preparing for a board meeting and ask the AI tool to compare two perspectives, analyse competitor strategies or suggest challenges to assumptions. For example, if a particular strategy has been recommended, you could write a prompt to ask the GenAI tool, ‘What are the counter arguments for this approach?’, (on a de-identified basis). Or, ‘What would a skeptical investor say about this?’ It can help by giving you some challenges to consider.
I also like that it puts things into frameworks. For example, here are five ways to approach this. You can get it to analyse competitor moves by asking it what the competitor’s business model is, their funding model, the background of their leadership team, their strategic position on a specific topic.
AI reviews publicly available information to generate its outputs. It can also help you quickly demystify things. Perhaps there’s an unfamiliar topic in your board papers. You can ask it, ‘Please explain this to me like I’m a financially literate director without a technical background in this area’ and it will give you a response to help you understand the topic a bit better in preparation for a board meeting.”
.jpg)
Regardless of the quality of your prompt, what shouldn’t you rely on GenAI for?
“That goes to my second caveat. All of these things GenAI can do are fantastic, but it must never replace your own director fiduciary duty or judgement. GenAI amplifies our ability to be well-informed and thoughtful, but you have to apply judgement.
We know GenAI hallucinates. It could tell you, ‘Competitor is doing X’, and you will review the response it provides and from your own knowledge you know that doesn’t look right. You have to remember this ‘graduate’ gets things wrong. Think of the ‘graduate’ asking you ‘Is this helpful?’ and you’ll look at it with a critical lens and say, ‘Yes, this is 80 per cent helpful’. Humans still need to curate and critique the AI outputs — framing the problems, spotting weak logic, seeing wrong emphasis in the summary generated. Humans have the social skills to take the insight and do something with it.”
How often do you use Anthropic’s Claude AI?
“I’m ostensibly using it as my search engine. The more I use it, the more practised I’m becoming. I practise being specific in my prompts. I get it to provide scenarios by asking it to, ‘Think about this from an alternative perspective’. I see it as bringing expertise to many, because now someone non-technical can engage meaningfully on something technical or use it as an educational tool — again with the proviso it might hallucinate, so we have to be careful. Verification and validation are still really important. It can’t replace the wisdom, experience and judgement of humans. However, especially as we navigate such a complex environment, it can amplify our contribution.”
Key takeaways for high-performing boards
- Start with a detailed prompt, then refine it through iteration to improve the results.
- Treat your GenAI tool as a capable graduate who has lots to learn.
- The best prompt engineering — and importantly, the resulting AI output — must never replace your director fiduciary duty and judgement.
Latest news
Already a member?
Login to view this content