Author: Liz Peuster, CAE
This article originally appeared in the Fall 2024 issue of the South Carolina CPA Report
As CPAs embrace AI to streamline workflows, one truth stands out: the quality of your AI’s responses depends entirely on the quality of your prompts. Think of AI as a highly skilled intern—it has the knowledge but needs clear instructions to deliver precise results. That’s where prompt engineering comes in.
What is Prompt Engineering?
Prompt engineering refers to the process of designing clear and specific instructions for large language models (LLMs), like ChatGPT. The better your prompts, the better the output. Imagine explaining a task to a new hire—they might be smart and capable, but if you don’t explain what you need clearly, they won’t deliver what you expect.
The same applies to AI. By giving it focused and detailed prompts, you can guide it toward producing relevant, accurate results tailored to your needs.
The 4 Key Components of a Powerful Prompt
When crafting prompts for AI, there are four essential components you should always include to get the best results:
1. Role: Define the AI’s expertise
To get the best results, assign the AI a role. Whether it’s an analyst or a writer, this helps the AI understand what kind of response you’re looking for. When assigning the role, tell the AI it is an expert in the area. For example:
“You are an expert in financial analysis reviewing audited financial statements. Identify key trends and potential financial risks.”
Assigning expertise narrows the focus, ensuring the response is relevant.
2. Context: Provide background information
Just like a new hire, the AI needs context to perform well. Without it, the AI might make assumptions that lead to inaccurate results.
Always include relevant background information:
“Here are the audited financial statements for XYZ Corporation for FY2023. Focus on liquidity ratios and any potential red flags in cash flow.”
Providing context helps tailor the response to your specific needs. This could be the most tedious part of prompting, but I encourage you not to seek a shortcut here. The more information the AI has about the background of the task, the more likely you are to receive a better output.
3. Task(s): Specify what you need
Be clear about the task you want the AI to perform. Whether it’s drafting a report or summarizing data, specificity is key:
“Summarize the key financial trends in XYZ Corporation’s audited financials and provide recommendations for improving cash flow.”
The more specific you are, the better the AI can deliver the exact output you need.
4. Questions: Ask what’s missing
Don’t hesitate to ask follow-up questions. Prompting AI is an iterative process. Once you receive a response, refine your prompt if needed:
“Expand on the cash flow risks and offer additional recommendations for improving liquidity.”
Follow-up questions help you get more detailed and focused results. Ask what else the model needs to know to successfully complete your task.
Take a Test Drive
Let’s look at a practical example. Imagine you’ve just completed an audit of a client’s financials, and you need to draft a client-facing email explaining the findings:
“You are an expert financial analyst preparing a report for XYZ Corporation. Draft an email to the client summarizing key financial insights from their audited financials, focusing on liquidity risks and recommendations for improving cash flow. Remember, the client is a non-financial professional, so your communication must be clear and concise. Before you perform this task, what else do you need to know to be successful?”
This prompt clearly defines the role, provides context, outlines the task, and leaves room for further clarification, ensuring a tailored and accurate response.
Tips for Writing Effective Prompts
- Be specific: Detailed prompts lead to better outputs.
- Set boundaries: If you need concise results or a particular format, make that clear.
- Provide examples: Show the AI what you expect the output to look like.
- Iterate: Keep refining your prompts until you get the result you need.
- Practice: You are learning a new skill, so give yourself time to learn. As you become accustomed to working with different AI models, your prompting skills will improve. Remember: language matters, so pretend you are talking to a peer, colleague, or intern.
Prompt Engineering Checklist
- Role: Have I defined the AI’s expertise?
- Context: Did I provide enough background information?
- Task: Is the desired output clear?
- Questions: Have I asked what’s missing or needed clarification?
By focusing on these four elements, you can unlock the full potential of AI, turning it into a tool that enhances your efficiency and productivity. The key to great AI results is great prompts, and with a little practice, you’ll be crafting them like a pro.
