How Accounting Firms Should (and Should Not) Use ChatGPT

By Christopher Stark, Founder & CEO, Cetrom.

ChatGPT ushered in a new era of automation in accounting (link goes to YouTube), for better or worse. While technology has made many content-based tasks much easier, questions remain about its accuracy and security.

This machine learning-powered chatbot with the GPT-3 engine at its core has captivated people’s attention with its rapidly evolving ability to interpret questions and commands to produce generally coherent natural-language responses. Within five days of its release in November 2022, more than a million people had given ChatGPT a spin. That number ballooned to over a hundred million users in less than a year, making it the fastest-growing application ever.

The AI has shown itself capable of composing everything from college essays to copywriting. ChatGPT waxes philosophical in intriguing user exchanges and has even won art contests. In finance and accounting, early adopters have seen the potential of language-based AI interfaces to create scripts and formulas and draft responses to clients.

The best to come from this early AI boom for the public is mostly time-saving mechanisms. It’s also raised intriguing questions, and perhaps more worrying ones, about the future, especially for professions whose labor could be replaced by more polished AI language models with greater capabilities.

This article will better explain how your accounting firm should (and should not) use tools like ChatGPT.

The Good – ChatGPT’s Abilities to Boost Efficiency for CPA Firms

ChatGPT is the culmination of recent advances that put powerful new AI tools into the hands of general users. Image-generating AI can create images in multiple styles, responding to users’ prompts. There are comparable services that focus on audio, text, and video. OpenAI, the organization behind the GPT model used by ChatGPT, created many such innovations.

ChatGPT is deceptively simple. After creating an OpenAI account, users can input any text they want. When the message is sent, ChatGPT responds quickly in written form to respond to the user’s query.

For all its simplicity, though, accountants have already discovered that AI has some surprising capabilities, especially in handling programming and Excel data.

For example, unorganized information like addresses or phone numbers that have been messily condensed into columns in Excel is tough to work with. A general, human-language spreadsheet description could be enough for ChatGPT to produce an Excel macro that can remedy the problem by automatically sorting and splitting the data into separate columns.

It may take some fine-tuning within ChatGPT, but the chatbot has responded with successful coding creations that are functional. The system can also explain why certain functions are being used and how they help achieve the desired fix.

Instead of spending hours creating a macro or potentially longer learning to script, the GPT model can use its knowledge of the programming interface of a separate technology platform to do the job in minutes. It’s clear that generative AI is useful for making custom tools, such as the kind that might be used by CPA firms to perform otherwise laborious tasks without requiring intensive research or development time.

Because of ChatGPT’s written communications capabilities, many professionals use it to speed up creating drafts for formulaic emails. ChatGPT can write and rewrite, and it is often worth doing so to get the right tone for a particular message.

Unlike how you might with a human being, it’s as simple as pasting text into the chat form and asking it for a revision.

Similarly, ChatGPT can produce copywriting and content writing, which could improve an agency’s marketing efforts. The details or style of current marketing materials can be modified through ChatGPT and translated to speak to a new audience.

Conversations with ChatGPT may yield new ideas, unexpected connections, or even just simplify a complex topic. Users can explore its huge volume of aggregated information to simplify the research process (with caution not to treat it as a final authority).

The Bad – Limitations and Weaknesses of ChatGPT

Generative AI, like ChatGPT, has critical weaknesses for all these uses. Most egregiously, it can be confidently incorrect.

ChatGPT will sometimes give the wrong answers to math problems. More complex arithmetic proves a challenge. It will come up with reasonable-sounding responses that are simply wrong. Interestingly, generative AI can also “hallucinate” — a self-generated Mandela effect where a non-truth becomes accepted as fact for no apparent data-driven reason. In some cases, ChatGPT incorrectly fuses disparate data together or relies on sketchy sources to validate its claims.

That should concern CPAs planning on using AI to generate client-facing content. Anything ChatGPT-generated must be carefully vetted for accuracy.

The system is generally good at creating code or instructions for Excel, Python, and other products and programming languages. However, the platform can also be mistake-prone, just as it is when creating natural-language responses — generating code that produces errors or using functions it hallucinated into existence that do not exist in the target program or language.

It’s apparent that relying on ChatGPT alone to code important assets for an agency is ill-advised, as its code could disrupt systems or at least create headaches.

But ChatGPT can also pose a security risk. OpenAI can use data that users input to ChatGPT to improve future language models per the end user agreement. OpenAI claims that it removes personally identifying elements from such data. Even so, if accountants input sensitive data in ChatGPT, it could become part of the database it is trained on — and potentially resurface in other users’ interactions.

The Uncertain – Future Developments and Expectations in AI for Accounting

As GPT is adapted from a general-purpose model to more specific sub-models, greater accounting-relevant advances will come. For example, a trusted future model might be granted access to client files, allowing it to become a customer-facing chatbot.

With the many questions clients direct at CPA firms, simply pointing an AI at a firm’s practice management system and having it answer them seems like an exciting use for generative AI in the not-too-distant future.

Of course, natural language processing is already being used for customer service — think of the automated chat windows on many websites. However, future iterations of the technology will have greater capabilities actually to answer questions.

Advances in AI in the next decade or so will likely lead to serious upgrades in the internal tools at a firm’s disposal. Legal and compliance teams could have enhanced abilities to use AI tools to scrutinize regulations and develop ways for their agencies to meet those requirements.

Data-driven models that simplify the vast fields of numbers and information that constitute many processes in the accounting profession could become standard implementations in practice management software.

The more disparate tools can interact with one another, the more powerful their potential.

Best Practices for ChatGPT

For now, ChatGPT and similar models remain accessible largely in limited prototypes or proof-of-concept versions. This is why accounting agencies should proceed carefully.

While ChatGPT may be convenient, the bottom line is that it isn’t always accurate. AI builds off user interactions to build a database, which means it’s always learning.

Remembering the security risks behind that function should keep accountants from blindly dumping swaths of text containing confidential information in ChatGPT. It’s unlikely you fully understand how the AI works, including when handling confidential data or intellectual property, so treat it with caution.

Vet ChatGPT like you would any third-party and with the same scrutiny you would use for any other technology for security and privacy. Don’t trust it based on prestige, novelty, or popularity alone.

Consider going further and establishing an agency-wide acceptable usage policy for generative AI. Ensure that employees are aware of the public-sourced nature of ChatGPT and encourage them to withhold private information to keep it from entering the public database.

Conclusion

Predicting how AI will advance is difficult, especially considering its enormous progress in recent years. Indeed, GPT-4 is here and already displaying a range of new capabilities.

In the meantime, the best course of action is cautious experimentation with a healthy dose of skepticism. Don’t be afraid to be imaginative; who knows what kind of Excel macro you could come up with? Some of ChatGPT’s best uses are yet to be discovered, and its relevance to accounting, like all fields, will continue changing.

===

Christopher Stark is the President and CEO of Cetrom, founded in 2001. Today, Cetrom is recognized as the Top Hosting Provider for CPA Firms (Since 2015). Learn more.