Businesses across the globe are at the beginning of the artificial intelligence revolution. As they’re exposed to various forms of AI, they’re experiencing how impactful the technology can be.
In the accounting industry, firms are using AI to complete repetitive tasks, analyze financial information quickly, identify anomalies, and create initial drafts of professional writing. A
Major accounting firms have announced billion-dollar investments in AI use, but much less has been heard about how small and midsized firms are leveraging this technology.
Without breaking the bank, small and midsized firms can make smart investments to adopt AI and tailor it for specific uses. With the release of generative AI, it has never been easier for firms to own AI models that uniquely serve their business processes.
Picking the right AI model
There are three different approaches that firms can go about to adopting generative AI: out-of-the-box, custom-developed, and assembling.
1. Out-of-the-box: Purchase the technology from a variety of vendors.
To date, many small and midsized firms are leveraging an out-of-the-box model. These generative AI applications, like ChatGPT, can be a cost-efficient option for those firms, but have their limitations.
Out-of-the-box models can be unreliable, lacking the practical context needed to aid accounting firms. They’re trained on publicly available information without access to specialized articles with paywalls or whitepapers produced by industry experts, and their access to public information can have time limits, leading the model to provide outdated information.
Additionally, data security is a concern when using out-of-the-box AI models. Closely evaluate the AI company you’re partnering with to ensure you maintain compliance with applicable professional standards, regulations, and contractual commitments. Sensitive data should generally stay protected within your own network. Be sure allowable data sent to the generative AI model outside your own network isn’t being kept any longer than absolutely necessary, and that your client contracts disclose the use of this technology and provide consent to use their information where applicable.
2. Custom development: Build an in-house model with custom generative AI.
Businesses could choose to develop their own generative AI model from scratch, but that could prove costly. For an accounting firm, building a model would require a lot of proprietary data that could be difficult for some firms to access, especially if the cost of acquiring the data is high.
3. Assembling: Adopt offerings created by companies such as Microsoft, OpenAI, Amazon, and Google, and integrate the technology with current business processes.
Small and midsized firms should take advantage of the technology by partnering with an AI company to assemble their own generative AI model on top of a pre-existing application.
![artificial-intelligence-eight.jpg](https://arizent.brightspotcdn.com/dims4/default/4d0a460/2147483647/strip/true/crop/204x195+0+0/resize/740x707!/quality/90/?url=https%3A%2F%2Fsource-media-brightspot.s3.us-east-1.amazonaws.com%2Fb3%2F96%2F24a47d1a4d8ba855754ee804bffc%2Fartificial-intelligence-eight.jpg)
Andrew Ostrovsky/agsandrew – Fotolia
Assembling and maintaining a generative AI model
Traditional machine learning models require thousands or even millions of data records for the technology to become highly accurate. Generative AI typically requires no training. Firms can use the technology out-of-the-box in numerous ways to streamline processes.
There are already a variety of companies integrating generative AI into their product offerings that small and midsized firms can utilize. ChatGPT has an enterprise version and Microsoft has a variety of generative AI offerings rolling out through familiar products like Excel, PowerPoint, and Microsoft Dynamics.
Firms can take the technology a step further by building on top of it with prompting. Prompting can be used to provide extra context to a generative AI model, or to give it a persona, which allows it to access specific information in its corpus.
If you want to add extra context to your generative AI model, you could, for example, feed it an article about updates to lease accounting in 2023, and ask the model questions about the article, or to summarize it. If you want to give your model a persona, you could tell it to take on the role of a tax partner and to write a memo about Internal Revenue Code Section 163(j), including citations.
Assembling a generative AI model on top of existing models could be a cost-effective option for many firms, and tools such as Microsoft’s Azure Cognitive Services or DataRobot are a starting point, though these require a close eye when it comes to maintenance.
A firm that assembles a generative AI model is essentially building its technology on top of someone else’s, possibly entangling firm data with the AI company’s and putting it at risk of being collected. Just like with an outsourced model, closely evaluate the AI companies you work with to secure data and not violate any applicable commitments, regulations, or professional standards.
Firms should also consider the costs of managing the generative AI model. The lack of AI training and education can be a major barrier to finding talent with the necessary skillset to manage AI models, but with the right governance, you can make small investments in management by building AI capability between IT and client services.
Making small investments will prolong the building process, but it grants small and midsized firms the ability to gradually invest in long-term AI development.
Small and midsized firms don’t need to spend millions of dollars to start adopting and experimenting with custom AI solutions. Partnering with AI companies or building a small skillset internally is a great way for them to leverage this technology.