The Productivity Implications of How CFO Survey Firms are Using AI

If recent developments in generative artificial intelligence (gen AI) and large language models live up to the imagination, they will enable productivity growth and have significant implications for the macroeconomy. Indeed, a host of researchers and economists predict notable productivity and efficiency gains from AI over the coming decade.

However, there is considerable uncertainty around the effect of AI on productivity. First, researchers need to estimate how many occupations will be affected by AI and, within occupations, which tasks will be affected. To estimate effects on overall productivity, we then need to, among other things, estimate the share of those tasks and occupations in the macroeconomy. In all these estimates, there are large margins of error. As Andreas Hornstein discusses in a recent Economic Brief, even studies that use a common conceptual framework can come to widely different conclusions about how much AI will increase productivity over the next decade. Another important factor is timing: Even if we expect large productivity improvements, when will they happen? As economist Zhu Wang wrote in a recent opinion piece, it may take a long time to achieve measurable, large-scale productivity gains from AI.

The CFO Survey can provide insight mostly on the first question: Which tasks (and thus potentially which occupations) will be affected and how? It might also provide insight into the potential for the sorts of AI-enabled systems or organizational change that could spur productivity growth that comes not just from workers producing more with each piece of capital (or machine) but from the combination of labor and capital together producing more, or what economists call total factor productivity (TFP) growth. Some researchers argue that AI has not yet delivered on the promise of technology like the internet has, which even in its infancy enabled low-cost solutions (such as online selling) to disrupt high-cost solutions (such as brick-and-mortar stores). Responses from the financial leaders surveyed in the CFO Survey indicate that although many firms are using AI for automating tasks such as document creation, financial reporting, and customer service, current technology and adoption is growing but remains somewhat limited.

How Many Firms are Currently Exploiting AI Technology?

The first step to knowing how AI might disrupt processes and enable productivity is understanding how many firms are currently adopting AI tools. The CFO Survey results from the second quarter of 2024 on the adoption of AI indicate that of the 60 percent of respondents who introduced automation into their processes in the last 12 months, about 40 percent adopted labor-saving AI tools. Thus, about 20 percent of all firms who responded to the survey reported adopting AI in the last year. Looking forward, more than 30 percent of all respondents intend to adopt new labor-saving AI tools in the next 12 months.

The CFO Survey results are similar to those from a monthly survey of businesses in the Richmond Fed's district, which indicated that about 46 percent of firms automated some processes in the past two years, and about one-third of those adopted AI technology. A recent study by McKinsey Global Institute found an even higher share of firms using AI — in that survey, 65 percent of respondents reported that their organizations are regularly using generative AI. In the CFO Survey responses, adoption by large firms far outstripped that of smaller firms. In the Richmond Fed survey, adoption of AI was slightly higher among service sector firms that reported automating processes.

What Do We Know about the Productivity Potential of AI?

The McKinsey study indicated that respondents are starting to see cost decreases and revenue increases in functions such as HR, supply chain/inventory management, and marketing or sales. But is task automation and revenue enhancement in certain functions enough to get us to the considerable AI-generated GDP growth predicted by, for example, Briggs and Kodnani?

In a May 2024 NBER paper, Daron Acemoglu focuses on two of the channels through which AI can enable productivity gains: (1) automating tasks (e.g., mid-level clerical functions, data classification, etc.); and (2) giving workers information or inputs that would increase their production per hour. Acemoglu assumes that generative AI will have notable, albeit more modest, effects on the macroeconomy by improving productivity and reducing costs in a range of tasks. Through this channel, he suggests a much lower impact on productivity than Briggs and Kodnani, estimating that TFP effects from AI advances will have an upper bound of a 0.66 percent increase in total within 10 years, which translates into a 0.9 percent GDP increase over the same period. (The GDP increase may be more because complementarities will lead to greater investment and new tasks and products will add to GDP.)

Key to these calculations are the type, number, and size of tasks and processes that are being disrupted by AI. Do the ways in which CFO respondents report using, and expecting to use, AI tools tell us anything about the potential productivity gain from AI in the near term?

The Tasks Associated with AI

The second quarter CFO Survey asked firms to describe their use of AI tools. The responses were broadly consistent and indicated a focus on relatively standardized accounting tasks, content creation, data analysis, and HR activities (which is, perhaps, not surprising to Agrawal, Gans, and Goldfarb). When asked how they were using AI tools, a quarter reported using AI for administrative or routine tasks, including routing calls, automated billing, or repetitive data analytics that can reduce employee time spent on a task. Another quarter of firms reported uses like this respondent, who wrote, "[We use AI for] software development/coding ... about 80 percent of the work can now be done by AI with the final 20 percent requiring human review/interaction." That respondent went on to describe another way his firm was using AI, which was in accounting. In fact, many firms (just under 20 percent of respondents to this question) reported that AI was starting to help with financial reporting, budget reconciliation, and accounts payable/receivable.

Another quarter of firms reported using AI for content creation. Although a lot of the responses indicated using AI in developing marketing content, there was other notable content generation using AI tools. As one respondent wrote, "[We use AI for] writing emails, creating presentations, [and] writing policy." Another wrote, "We don't have company-wide AI tools but on individual basis, we ... utilized tools to write/compose email and documents." Many respondents expressed a continued practice for employees to finalize content, e.g., "We have used ChatGPT to draft job descriptions, contracts, press releases, etc. and employees then modify and edit those documents to arrive at the final draft". Many firms also reported using AI for first-line customer service tasks and for fraud detection.

Almost all the firms using AI indicated using it for specific tasks — many fewer firms referred to organizational-level process or system enhancements using AI. An example of a firm that seemed to be using AI tools more broadly reported using them for "forecasting, logistics optimization, call center automation, equipment failure prediction, [and] administrative data analysis." Another firm noted that they have "installed an AI overlay to our building management system at a particular building." Still, even these firms indicated using AI in one process within their organization, not in a way that transformed their organization.

It was also more likely for firms to report process optimization or improvement as something that they plan to implement. But even then, the uses seemed focused and challenges existed. For example, one firm reported, "AI implementation could greatly improve processes. The biggest issue with implementation is the security environment."

Moving Forward

Looking ahead to the next 12 months, many firms described how they are researching opportunities for AI usage, such as "undertaking a pilot to assess the options for where AI could have an impact" and "learning the nuances of AI integration." The transformative possibilities of AI will not happen quickly. New tasks might be created, new systems might emerge, and we might see organizational structures adapt to the new technology — none of which is evident in our surveys yet. As Agrawal, Gans, and Goldfarb argued in a 2021 working paper, the potential for AI to affect productivity is impacted not just by the tasks that can be enabled by AI, but by the organization's structure. Right now, the CFO Survey firms that are using AI are using it to make a set of tasks (such as accounting or job description generation) more efficient. But as everyone agrees, this is only the beginning.