3 ways to use generative AI to save time today

Questions about accuracy and security could be slowing the adoption of generative AI in business, especially in such an exacting field as accounting.

However, a pair of CPAs who recently presented to colleagues at AICPA & CIMA ENGAGE 2024 said there’s a way accountants can gain from generative AI now, without most of the risk.

“Everyone is getting worked up about data privacy and all that, which is important, but you can get a lot of value from generative AI without putting in a single word of private data,” John Higgins, CPA/CITP, founder and CEO of Higgins Advisory LLC, said during the session “Artificial Intelligence — How to Unleash the Power of ChatGPT, Copilot and Other AI Tools.”

As co-presenter Tom Trainor explained, ChatGPT, Microsoft Copilot, and similar forms of generative AI don’t need access to potentially sensitive data to help streamline your workflow processes.

“When you’re thinking about using AI, you can work on your firm as opposed to in your firm,” said Trainor, managing director of Hanover Private Client Corporation. “You can do all of that stuff with AI now, today.”

3 time-savers for today

“There are a lot of work processes you can rethink,” said Trainor, who likened the work that generative AI is currently producing to that of “a really smart, fast first- or second-year associate. It’s really critical that you use professional judgment when you’re doing this.”

Trainor continued: “Whenever you’re stuck and in the old days you would lean over and ask your colleague, think about instead asking ChatGPT or Microsoft Copilot to do it. Simple things like when writing, finishing a sentence or paragraph that you’re not sure how to finish, or getting an email you don’t know how to respond to — those are some simple things.”

The presenters showed ENGAGE attendees, with detailed examples, how to:

• Create a job posting with ChatGPT

A task like this can be done in the free version of ChatGPT, although the knowledge base that it will pull from will be more dated than the paid version (starting at $20 per month for individuals).

“Prompts” are your instructions to the generative AI tool. In this case, Higgins typed in the prompt:

“Write an engaging job description for an online ad to recruit a staff accountant. This is a hybrid position located in Michigan. Our primary software applications are SAP, Salesforce and Concur.”

In mere seconds, ChatGPT produced a formatted 400-word job description that could be copied and pasted as a baseline document. Some editing still would be required, of course; Higgins pointed out how ChatGPT went ahead and listed benefits.

“I didn’t say anything about this job having benefits,” Higgins said, eliciting laughter from attendees.

• Use Copilot to summarize emails, videos

Using a paid subscription to Microsoft 365 Copilot for business users ($30 per month), Trainor clicked on the “Notebook” function in the upper-left corner of the webpage — which allows prompts up to 18,000 characters — copied and pasted a chain of back-and-forth emails related to a work proposal, then wrote this prompt at the top:

“Please summarize the following email thread. Format as a table with two columns for the speaker and their contributors. Also include an assessment of each participant’s enthusiasm for the proposal.”

Copilot quickly and effectively shared a useful response.

The free version of Copilot could have produced similar results and is a good option for getting familiar with the power of generative AI, but it is limited in terms of security and functionality. With a subscription to Microsoft 365 Copilot, which keeps data more secure relative to ChatGPT, you could prompt the AI tool to, for example, scan your Outlook emails, Teams messages, Microsoft documents, and summaries of other relevant information.

Microsoft Copilot 365 also could create a summary of Teams meetings related to the proposal, and that summary shortcut also extends to some videos and PDFs. According to Trainor, Microsoft Copilot 365 can summarize YouTube and Vimeo videos that have transcripts and can do the same with PDFs opened in the Microsoft Edge browser.

Higgins recommended that Microsoft Copilot 365 subscribers perform as many work functions as possible using Microsoft apps for maximum impact. For example, subscribers should conduct online meetings via Teams as opposed to Zoom.

• Create a macro in Excel

“All that time you’ve been spending in Excel training, you don’t have to do that so much anymore. This can be your real-time training for Excel,” Higgins said. “ChatGPT can’t create the macro, but it can tell you how to.”

Higgins, wanting a shortcut for printing a specific segment of a 30-year amortization schedule for clients, prompted ChatGPT to: “Create an Excel macro to print the amortization schedule through to a specified payment date entered via prompt to the user, including the mortgage input variables.”

ChatGPT quickly responded with a step-by-step guide, complete with the coding needed to create the macro.

Soon after, Higgins clicked on “macros” on the developer tab of Excel, accessed the macro, and printed out a four-year segment of the amortization schedule.

“I can clearly see a future where you and I are our own programmers,” Higgins said. “Can you imagine that?”

The future of generative AI

Generative AI’s “extreme state of flux,” as Trainor called it, is highlighted by the fact that half of the ENGAGE presentation featured AI capabilities (thanks to the recent introduction of Copilot) that weren’t even available earlier this year.

Beyond workflow processes, Trainor and Higgins also advocate for using generative AI now for data analytics, but they do so with warnings.

“My biggest concern is that if you spend a lot of time in this area, you get intellectually lazy. You just say, ‘Good. That’s great. Good job, ChatGPT. We’re good to go.’ That would be almost malpractice,” Higgins said. “If you generate any facts or figures from these tools, you have to get additional verification that it’s accurate. Don’t just rely on it. You can get yourself in a really embarrassing situation or, even worse, litigation if someone is really relying on the output.”

Higgins added that a lot less due diligence will be necessary in “30 to 40 months” as the models continue to rapidly evolve. Still, whether for now you choose to use generative AI just for workflow automation, for data analytics, or for both, Higgins believes the only choice that isn’t acceptable is “neither.”

“Not only in your own organization, but if you’re helping your clients, if your clients are mystified by this, they’re going to be looking to you to give at least some insight,” he said.

Editor’s note: Those who purchased an all-access pass to ENGAGE can view this and other archived sessions and a new lineup of live ENGAGE+ sessions on July 17, Sept. 18, and Nov. 14 for additional CPE. If you didn’t attend ENGAGE, you still can access this session.

— To comment on this article or to suggest an idea for another article, contact Bryan Strickland at Bryan.Strickland@aicpa-cima.com.