AI Implementation: Crafting Effective Prompts for SMEs and Family Businesses

The Four Building Blocks of an Effective Prompt

In our work with SMEs and family businesses, a straightforward framework has proven particularly effective. It consists of four elements that together help AI systems understand what is actually expected of them. This approach can be applied to virtually any AI application in your organisation.

1. Context – Who Is Speaking, and in What Situation?

Providing the right context is especially important. How else would AI systems acquire the necessary knowledge about your industry, your company, or your customers? The pitfall here is clear: everything you do not tell the AI will be replaced by assumptions, and these assumptions rarely hit the mark.

Consider an example from mechanical engineering: “Write an email about the delivery delay” produces a generic apology that could just as easily come from a furniture retailer. However, the instruction “You are supporting the sales team of a mid-sized special machinery manufacturer. Our customer is a long-standing partner in the automotive supply industry. Draft an email that explains the two-week delay in delivering a specialised milling machine while strengthening the relationship” yields an entirely different result.

The good news is that context does not need to be reformulated with every request. Many businesses create a baseline context containing the most important information: industry, company size, typical customer structure, and communication tone. This context is then supplemented with situation-specific details as needed, representing an important first step toward systematic AI implementation.

2. Task – What Exactly Should Happen?

Precision in describing the task matters more than length. The instruction “Write something about our new product” is vague. “Formulate three key messages for our new product, each highlighting a different customer benefit” is concrete.

A helpful rule of thumb: If you were assigning the same task to a new team member, what follow-up questions would they ask? Your prompt should answer those questions in advance.

Typical questions might include: Who is the intended audience? What is the objective – to inform, persuade, or document? Are there aspects that should be emphasised or deliberately omitted?

A practical tip that has proven valuable: test your prompt first with a rough draft. The gaps in the initial AI output will show you which contextual information is still missing.

3. Format – What Should the Result Look Like?

AI systems can deliver results in virtually any form: flowing prose, bullet points, tables, or drafts with placeholders. Without clear guidance, the system will choose a format that rarely fits your needs optimally.

This is particularly relevant for SMEs and family businesses: many tasks already have established formats. A quotation follows a specific structure, meeting minutes have fixed components, and a complaint response has a proven layout. These existing structures should be described in the prompt or provided as a template.

Specifying “maximum one page” or “in three paragraphs” may seem trivial, but it saves considerable rework in daily operations.

4. Examples – What Would a Good Result Look Like?

This fourth building block is optional but highly effective. When you show the system what a successful result looks like, the hit rate increases significantly from the start. In technical terms, this is called “few-shot prompting” – you provide a few examples from which the system derives the pattern.

For businesses, this means: collect well-crafted texts, emails, and documentation as reference material for the AI. These serve not only as quality benchmarks for your employees but also as training material for your AI applications.

From Individual Queries to a Prompt Library

The real productivity leap comes not from better individual prompts but from systematization. Imagine your sales department has five standard situations in customer communication: responding to initial enquiries, following up on quotations, handling complaints, communicating delivery delays, and preparing annual reviews.

For each of these situations, you can develop a tested prompt containing all four building blocks that every team member can use. The result is not standardisation in a negative sense, but quality assurance: the foundation is consistently good, while individual adjustments remain entirely possible.

This prompt library grows over time. Employees who discover better formulations can contribute them. What works gets documented. What does not work gets refined. This creates collective knowledge about the effective use of artificial intelligence in your organisation.

From the perspective of AI systems, two aspects are particularly crucial: processes and context. When processes are clearly described and context is available, AI can truly deliver the leverage effect you expect.

Common Mistakes and How to Avoid Them

Asking for too much at once: A prompt that simultaneously asks the AI to analyse, summarise, evaluate, and provide recommendations overwhelms many systems. It is better to break complex tasks into discrete steps.

Too little context, too much trust: AI systems are not experts in your industry. They generate plausible-sounding text that may be technically inaccurate. The more relevant context you provide, the lower the risk.

Accepting results without review: Even the best prompt does not guarantee error-free results. Review by a knowledgeable person remains essential – it simply becomes faster because the foundation is already sound.