{"id":2518235,"date":"2025-03-25T19:23:02","date_gmt":"2025-03-25T23:23:02","guid":{"rendered":"https:\/\/www.pymnts.com\/?p=2518235"},"modified":"2025-03-26T20:58:01","modified_gmt":"2025-03-27T00:58:01","slug":"beyond-the-hype-what-cfos-should-know-about-ai-agents","status":"publish","type":"post","link":"https:\/\/www.pymnts.com\/news\/artificial-intelligence\/2025\/beyond-the-hype-what-cfos-should-know-about-ai-agents\/","title":{"rendered":"Beyond the Hype: What CFOs Should Know About AI Agents"},"content":{"rendered":"

When chief financial officers (CFOs) consider implementing agentic artificial intelligence (AI), they should treat it like other technologies their companies have adopted in the past, with some adjustments to mitigate specific risks that come with the new system.<\/p>\n

There\u2019s a good reason for this approach: Agentic systems have been around for a long time, according to George Westerman<\/a>, senior lecturer at the MIT Sloan School of Management, in an interview with PYMNTS.<\/p>\n

What is agentic AI? Westerman said there\u2019s not yet one agreed upon definition, but at its simplest, it is software code that will process information and make a decision or take an action.<\/p>\n

\u201cThe concept is not new,\u201d Westerman said. For example, many processes launch and relaunch in Microsoft Windows automatically on a regular basis. Also, automated trading on Wall Street has been around for decades.<\/p>\n

Read more:<\/strong> Inside Goldman Sachs\u2019 Big Bet on AI at Scale<\/a><\/p>\n

Not Like Traditional Computing<\/strong><\/h2>\n

What\u2019s different is that instead of rules-based bots, where actions are pre-determined, AI agents built on top of generative AI models can dynamically generate responses, understand, adapt and learn while having autonomy and decision-making abilities to complete a task given by the user.<\/p>\n

\u201cIt can do things that traditional computing could not do: It can hold a conversation with a client. It can process unstructured data and documents,\u201d Westerman said. However, \u201cwe still want to use the same processes to evaluate how to do it.\u201d<\/p>\n

That means CFOs should approach agentic AI like past forms of automation they\u2019ve incorporated \u2014 evaluating what processes can benefit, identifying costs that can be removed, find potential benefits from accelerating work, and assessing risks to finances and reputation, according to Westerman.<\/p>\n

CFOs also should use their existing governance policies and risk frameworks to evaluate the new technology \u2014 and revise as necessary.<\/p>\n

\u201cYou should already have very strong governance and policy approaches to automation, and you\u2019d want to apply those same policy approaches to agentic AI but also see where those policies may need to change,\u201d he said.<\/p>\n

See also:<\/strong> Nvidia CEO: Why the Next Stage of AI Needs A Lot More Computing Power<\/a><\/p>\n

Assessing ROI<\/strong><\/h2>\n

The same approach goes for evaluating the ROI of agentic AI, Westerman said.<\/p>\n

Niall Byrne<\/a>, CFO at\u00a0Qatar Investment Authority (QIA)<\/a>, told the World Economic Forum<\/a> on Tuesday (March 25) that \u201cwe are exploring pilot projects with clear metrics to help quantify the return on investment on AI investments, including looking at adoption rates, data processing speed, value creation and employee productivity.\u201d<\/p>\n

Westerman said CFOs must be clear-eyed about costs, since implementing AI is \u201cexpensive,\u201d like other technology integrations. It was the same case with cloud computing, which is pay-as-you-go, but costs could climb with higher usage in contrast to using your own servers.<\/p>\n

For AI, the costs include compute charges, cleaning of data, for licenses and changing processes, which is often the larger expense, he said.<\/p>\n

Westerman sees the biggest agentic AI opportunities for finance in the following areas:<\/p>\n