Prompt Engineering: The Business Skill That Determines How Useful Your AI Is

The same AI model given different inputs will produce very different outputs. This is not unique to AI: the same expert consultant given different briefs will produce different work. But with AI, the gap between a vague input and a specific, well-structured input can be extraordinary.

A prompt that asks an AI to 'summarise this report' will get a summary. A prompt that asks it to 'summarise this report for a board audience, focusing on the three most significant strategic risks and the recommended mitigations, in no more than 300 words, using plain language without jargon' will get something substantially more useful. The difference between these two prompts reflects the quality of briefing, not the quality of the AI.

Effective prompts in business contexts tend to share several elements.

Context: who is asking, what organisation or function they are in, and what background is relevant to the task. AI models do not have access to your specific context unless you provide it.

Task specification: a clear, specific description of what you want the AI to do. The more specific, the better the output. 'Write a proposal' is a weak task specification. 'Write a 500-word executive proposal for implementing a new employee AI literacy programme, aimed at a CFO who is concerned about ROI, covering business case, implementation approach, and success metrics' is a much stronger one.

Constraints: length, format, tone, audience, and any specific things to include or avoid. Constraints help the AI produce output that is immediately usable rather than requiring significant editing.

Examples: where you have examples of what good output looks like, including them dramatically improves quality. 'Write in a style similar to the attached example' consistently outperforms descriptions of style alone.

In organisations that are using AI effectively, prompt engineering is becoming a shared team capability rather than an individual skill. Teams develop prompt libraries: collections of tested, effective prompts for common tasks that any team member can use.

This is particularly valuable for maintaining consistency. If a team develops a highly effective prompt for producing a weekly board update, sharing that prompt means all team members produce outputs of comparable quality, rather than the quality varying with individual prompting skill.

Microsoft Copilot's Copilot Lab feature is a specific implementation of this concept: a shared library of prompts that organisations can develop and share across their user community. Organisations that invest in developing a prompt library as part of their Copilot deployment consistently see higher utilisation and better output quality than those that leave prompting to individual initiative.

Prompt engineering is powerful but not unlimited. It can significantly improve the quality of AI outputs within the model's capability. It cannot make the model capable of things it fundamentally cannot do: accessing real-time information it does not have, producing reliable specific factual data it was not trained on, or reasoning through genuinely novel problems that require first-principles thinking.

There is also a point of diminishing returns. Prompt engineering effort should be proportionate to the value of the task and the frequency with which it is performed. Spending hours crafting the perfect prompt for a task you do only once is not a good investment. Investing in developing and testing a prompt for a task your team performs fifty times a week is.