Demystifying Prompt Engineering For Finance Teams (Hint, Anyone Can Do It)

This article was originally posted on Forbes.com, March 2024.

Dubbed one of the hottest tech jobs in 2023, prompt engineering ensures that AI models produce reliable and relevant outputs, but the term is misleading. Basically, if you’re asking a large language model service such as ChatGPT to write your resume, a typical prompt might be “make it shorter.” That doesn’t take any engineering skills.

That’s why I really don’t like the term prompt engineering. Certainly, it is part of a technical field, but to successfully instruct your AI assistant, you need to provide relevant context and apply language in a very precise, logical way. I see prompting as the art of crafting effective commands that guide your AI assistant’s behavior, ensuring effective human-AI communication.

How can you prompt AI to perform effectively?

In my last article, I explained that you can trust any AI assistant as much as you can trust any member of your finance team—as both will only perform as well as they have been enabled and instructed. Keep in mind that these LLMs have already been trained, so they have certain capabilities. At this stage, they need information, instructions and enablement to deliver relevant results.

Providing these instructions is the essence of prompting. It’s basically a conversation. You tell your AI what it should do and what the outputs should be. After it generates useful responses, you move on to the next round of instruction.

It starts with having a clear understanding of what you want to achieve. Let’s say you want your AI to go through your expense reports and ensure they are all compliant with the expense policies of your company: for example, not reimbursing alcoholic beverages.

The AI’s first task is to capture all the data from all the expense receipts that have been submitted by your employees. Next, it has to do line-by-line compliance checks and identify any items that are alcoholic beverages. The strength of an LLM is that it will already know what drinks are alcoholic, probably even better than you when it comes to a red wine brand or the name of an exotic Manhattan variation. You could create the following prompt for the task:

“The uploaded document may contain a line item that is an Alcoholic Beverage. If there is one, set the Yes/No Alcoholic Beverage data point to Yes; if not, set it to No. If you recognize an item as being alcoholic, flag the document for review”

Your human employee’s job will be to review the flagged documents.

After identifying the alcoholic beverages in the expense reports, the next task might be to check whether the incoming invoices comply with your local sales or value-added tax laws and regulations. This involves extracting the given rate on the invoice, categorizing the invoiced product and services, and ensuring the tax rate is correctly applied in accordance with all formal requirements. Providing such complex instructions in clear, simple terms is the job of the AI prompter.

Will you have to hire an AI prompter to run your new tool?

Ask yourself, which human team member would have the right skills to be that prompter? It would be someone who has already performed the task countless times. It would be someone who knows all the exemptions and would know what to do if they encounter an unexpected situation. It would be someone who is already performing the job today.

This person would be finding answers to complex questions by asking a series of questions: Should I reject this invoice? For what reason? What are the alternatives? What instructions should I give the tool in case the situation happens again?

This person would not be writing software code. This employee’s job would be to write specific instructions for completing specific accounting tasks and processes. This job requires linguistic proficiency, problem solving and critical thinking, and creative writing skills.

According to the Future of Jobs Report 2023 published by the World Economic Forum, over 85% of organizations believe adopting new technologies and increasing digital access will drive transformation in their organization.

WEF notes that the fastest-growing roles today “are driven by technology, digitalization and sustainability,” with AI and machine learning specialists topping the list of fast-growing jobs. They state that the fastest-declining roles today are also driven by technology and digitalization. These include clerical or administrative roles such as bank tellers and data entry clerks, as such jobs will be replaced by automation.

The report also points out that six in 10 workers will need training before 2027. Top training and workforce development priorities include analytical thinking, creative thinking and using AI and big data to achieve business objectives.

What does this mean for your workforce?

Implementing AI in the finance function does not mean you need to hire specialized engineers. It means you now have an opportunity to transform your data entry clerks into data analysts with specialized AI prompting skills. This is a true transformative shift in responsibilities.

The accountants of the future will need holistic insight into the company business, so they can interpret data and act as advisors. They will need to engage with people and AI tools, be part of an AI/human team and hone their communication skills.

‍