Accounting firm leaders who are excited about artificial intelligence have spent the last two years talking about how it is helping them address the pipeline crisis, surface insights and enable new efficiencies. Less common, however, are specific details for how the technology is actually doing any of these things.
This doesn't mean, though, that these things are not happening. While many accountants are determined to stay as far away from AI as possible, others have found ways to use this technology to yield tangible benefits for their firms. Whether they're fully embracing AI or still keeping it at arm's length, accounting leaders are already using these tools to, well, address the pipeline crisis, surface insights and enable new efficiencies — but in some very specific ways.
One of the most common is something that, whether they're going through FASB standards or the Tax Code or audit evidence, has historically been a massive time sink for accountants: document summarization and analysis.
It can be difficult to sift through a giant repository of text files, PDFs and other documents for exactly the information needed, according to Spencer Lourens, managing principal of data science, machine learning and artificial intelligence with CliftonLarsonAllen, which is why this was one of the first things the Top 10 Firm started to address in early 2023.
As a result, instead of poring over dense documents themselves, staff can now ask a generative AI bot specific questions and get the answers they need. This encompasses not only things like policies and procedures for client engagements, but also administrative questions that would normally be answered by an HR professional. This capacity is not only internal; Lourens said they've used it to help clients as well, such as a large nonprofit that had over 800 reports it needed to go through on an annual basis.
(See Accounting Today's special report: "
"So 50% of someone's full-time equivalent was going through these sometimes very lengthy reports and extracting the same information across all of them," Lourens said. "What we did was say, 'We can create this process where we take in the information and have a pipeline run where these questions are asked of the AI models involved, and they will go through and provide answers and try to cite themselves in the process to hopefully shave off hundreds of hours per year.'"
Samantha Bowling, managing partner of Maryland-based GWCPA, similarly uses generative AI to sift through the information that previously took hours for humans to process, which she then uses to offer financial statement analysis. While her firm has always done this, it used to take hours to gather the information and analyze it fully. AI allows her to cut this time down to mere seconds. This has vastly increased the scale at which her firm can perform these analyses, which in turn has allowed them to expand the number of clients they can offer them to.
"We're using it for clients who do bookkeeping work and outsourced CFO type work. We see this being a quicker tool to get answers on stuff and maybe provide this information to people who are just doing bookkeeping and not making financial statements … as opposed to waiting for the year end," said Bowling.
Suma Chander, a PKF O'Connor Davies partner responsible for high-technology strategic advisory, noted that her New York-based Top 100 Firm has developed a tool that not only provides analysis on a large corpus of internal information but goes further in actually formatting it for different uses. This is because, as part of the firm's AI strategy, use cases are encouraged to be functional across practice areas — whether tax or audit or ESG — so they can support firmwide improvements. The ability to not only research information but to format and export it, she said, has led to a 30-40% boost in time efficiency this year.
"So, how do you scrub data from a PDF or Excel or from a Word doc and put it into [the right] format and analyze it quickly? Everyone wants to do that for different reasons," said Chander. "They want to scrub it for a financial statement or for an audit or for an analysis. So, scrubbing data out of a PDF and putting it in Excel and being able to give the end user the ability to analyze it without going through so many invoices is very simple and you're saving 30-40% of your time. So this is something that cuts across the firm."
From consumption to creation
Going further, there are other AI applications that not only analyze information but produce some of their own too. Sergio de la Fe, digital enterprise leader with Top 10 Firm RSM US, noted that the firm is using generative AI to not only conduct high-level tax research, but to actually draw threads together to draft actual tax position documents. Again, while humans have long done this for clients, he said that things "get very complex very fast," so professionals must take into account things like the actual tax law itself, legal precedents about taxes and RSM's own positions on tax strategy. The company's internal AI can do all of this much faster.
"We've been able to create and leverage language models and generative AI to help us write those memos, because it accesses all of those databases and information, and is able to help us generate the response," said de la Fe. "So not just research, it's leveraging the actual end product and how those position documents are written."
Similarly, Carmel Wynkoop, partner-in-charge for AI, analytics and automation at Armanino, said the California-based Top 25 Firm has seen great returns on its house-developed SOC 2 platform, Audit Ally. Leveraging the Claude base model by Anthropic, Audit Ally helps auditors generate content that traditionally has eaten a lot of time, such as evidence requests, which they must write "God knows how many times."
"There's no need to hand-key all that information every single time. So we have a prompt that lets you get the gist of what you need and see the evidence and auto-generate what is intended for clients," she said, adding that while it might only save a few minutes at a time, over the course of a year, between all the staff who are using the tool, it makes a huge difference.
But even classical AI can yield enormous benefits for firms. OJ Laos, AI leader at Armanino, pointed to the firm's 13 Week Cashflow analysis tool as an example. The tool is built not on generative AI but robotic process automation, but is still able to analyze large amounts of information around spending and cash flow to project what a client's cash position will look like over a quarter, something which has proven very popular with clients.
"With RPA, the gains are massive. Everyone's talking about generative AI, but when you talk about real cold hard facts of ROI and time savings, those automations [are providing real benefits]" he said.
Indeed, Wynkoop said the firm has made a number of "digital workers" built mainly on RPA. For instance, there was a client who spent 15 hours a week reconciling funds across 70 entities and hundreds of bank transfers. The firm created a "digital worker" that interacts with those web systems and does all the consolidation and reconciliation for them, as well as providing a full audit trail. Another goes into more than 100 investment statements and pulls information on a monthly and quarterly basis.
"They were spending 50 hours a month doing this and creating journals. The digital worker now goes and does all of this and prepares a journal entry, [flags] anomalies and routes it for approval. They save 500 hours a year by utilizing the RPA," said Wynkoop.
Adding generative AI to the mix can produce even more powerful results to automation. RSM's de la Fe also talked about their bespoke solution, RSM Automation Compliance. The tool takes clients' regulatory requirements on the one hand and their internal control and process document on the other, and brings them together to map where their clients are or are not meeting compliance obligations. This used to be an exhausting process for human accountants, especially for complex clients, but now they can near-instantly analyze their regulatory compliance to see "for these three regs you don't have a procedure or control, or you have 10 controls over this one thing, so maybe distribute that better."
"Think of a client who operates in 10 states — the regulatory requirements that follow could be very diverse and then how do you know you capture everything, how do you know all the controls you got, how do you map each control to a compliance item but with generative AI we can reduce that time significantly, so significantly I'm scared to say," he said, adding that they also built in ability to ask the model to write a control or a process that meets the regulatory requirements.
Client communications is another area where accountants are seeing improvements via generative AI. Adolfo Marquez, marketing manager for Fresno, California-based MBS Accounting, said the firm has been using Karbon's AI capacities to summarize client communications, including lead follow-up. He said that, for certain accountants, writing and grammar might not be their strong point, but with the generative AI tool they can not only summarize client communications, they can also improve tone and maintain a consistent professional voice that fits with the firm.
Similarly, the firm is in the middle of a website redesign, and so has been using generative AI to not only produce landing pages for industry specializations but also optimize the content for search engines to generate more leads. However, he said this content is supplemented with their own unique perspective, as the language of these models can be very generic.
"Our philosophy has evolved to blending the generic boilerplate content AI produces with our unique point of view. For example, AI can produce ad nauseum about monthly reconciliations, but what are the specific ways we do it and the things we want our clients to know? … Why pay attention to us if they can get [generic information] anywhere?" he said.
Facing fears
While these accounting leaders have found value in AI, many others have not. Accounting Today's
Marquez conceded there is merit in some of these concerns, noting that the firm stopped using generative AI for marketing content because the amount of editing it often required erased a lot of the time savings. Overall, he said, leaders at the firm say, "It's hit or miss," so they need to tread very carefully around this technology.
"You put a lot of power into these bots, but really they're like a five-year-old kid," he said. "Having human review is so critical; it's the middleman between testing and improving efficiency and putting stuff into production because we have a solid reference network and very respectable organizations and we can't be a churn-and-burn niche firm. So balancing that is vital."
Overall, human review is a major portion of how firms protect themselves while using AI. For instance, James Watson, CLA's managing principal of service for the Pennsylvania region, noted that the firm goes through extensive testing and validation of its AI models' accuracy through a team of subject matter experts who evaluate the results.
"We don't build a system and say we're done; we go through arduous user acceptance tests with our team" he said.
Beyond human review, Armanino's Laos said the accuracy issue can be at least partially mitigated by restricting the model's corpus to only highly specific subject data. "You can very easily create a selective model that is used for a very narrow use case that we're able to control," he said.
PKFOD's Chander noted that AI tools, if used correctly, can actually improve, not degrade, accuracy. She noted that many widely used solutions, such as Sage or QuickBooks, already have AI built into them today, and these capacities have helped users be more accurate, not less. She pointed to bank reconciliations as an example: With AI identifying anomalies that humans might miss, they can be even more confident in their calculations.
"So we are adopting and using it in our own ways since we have these tools, and I think it has increased accuracy," said Chander, though she stressed that it all depends on how AI is used. "If you try to load a large sum of data into a chatbot and ask it to do something, that is a different story."
In terms of protecting client data, meanwhile, firms seem to have adapted by using private models that operate internally on the firm's own servers, versus public models where data gets sent to a third party. RSM's de la Fe, for example, noted that its AI draws mostly from internal databases — the firm even worked with Microsoft to get a protected version of Copilot that masks any client information leaving the firm.
"We've got to make sure we're not putting data out there that isn't appropriate, or getting data in that isn't what we should have," he said.
Understanding how to navigate such challenges will be essential in the future, as de la Fe says that even if an accounting firm isn't very interested in AI, its clients inevitably will be. He noted that a survey the firm conducted of its clients found that three out of four are using AI right now, and 85% of RSM clients who use AI say it has benefitted them more than they thought.
"So when we have a client base saying that we have no other option but to be brave and to really lean into the AI opportunities," he said, "this is really important."
