CFO Playbook: Agentic AI and the Future of Finance

In this article we discuss:

• What Can an AI Agent Do?
• What Differentiates an AI Agent 
• Critical Considerations for Deployment 
• ‘Jagged Intelligence’ Challenges With Agentic AI 
• Ensuring Trust at PayPal
• How Team Car Care Counts on AI

Agentic AI is rapidly transforming the finance function, from automating repetitive tasks to providing deeper analytical insights. And finance leaders are already finding ways to use the technology to power growth, according to CFOs from PayPal, Salesforce, and Team Car Care.

Silvio Savarese, executive vice president and chief scientist at Salesforce, offered an overview of agentic AI and its profound implications for the future of finance during the latest Fortune Emerging CFO webinar. (The next one will be held November 13.)

Savarese articulated the evolution of AI, the core components of AI agents, their current and future applications, and crucial considerations for their deployment. He was joined by Matt Castonguay, CFO at Team Car Care, and Jamie Miller, CFO and COO at PayPal.

What Can an AI Agent Do?

AI agents have created a new paradigm for the future. “AI agents can act as workflow and workforce multipliers for humans,” Savarese said, “like having a fleet of agents at your disposal, 24/7.” AI agents, he added, are capable of performing “various types of tasks on behalf of humans or augmenting humans’ capabilities.” Such tasks can span functions across finance, sales, marketing, IT, and customer service.

In the context of finance, Savarese identified a few compelling use cases, including in retail banking, where automation could provide “personalized, seamless customer interaction across various channels” and “enhance customer satisfaction.” In mortgage lending, he added, the technology could automate “the entire lending life cycles from applications to approvals, ensuring faster processing or a smoother borrowing experiences.”

Savarese also touched on the question of whether agentic AI should be able to independently make purchases: “We are still at the beginning of this process,” he said, emphasizing the critical importance of trust and ensuring agents operate within established boundaries when making significant decisions.

What Differentiates an AI Agent 

An AI agent differs from other technologies such as large language models (LLMs), Savarese said, describing its four essential components:

• Memory: “A memory allows the agents to be persistent, to have the ability to use information in a proper way,” such as customer data, past conversations, and best practices, Savarese said.
• Brain: “The brain allows the agent to reason, to plan actions, to orchestrate workflows,” he added.
• Actuators (function calls/API calls): Savarese said these traits enable “the agent to mix actions based on the plan that the brain is performing.”
• Sensory systems: This component defines “how the agency is connected with the outside world,” allowing communication through “language, voice, and other type of channels,” Savarese added.

Critical Considerations for Deployment 

For companies thinking about deploying agentic AI, Savarese offered key advice:

• Data infrastructure: Building the appropriate data infrastructure is critical, Savarese said, because an agent “does require access to the right information and to the right data.”
• Accountability and transparency: It's essential, Savarese added, to not just ask “what these agents can do for me, but how we can trust that these agents are making the right decisions?” Users must understand how to make the agents auditable and accountable.
• Governance: Savarese emphasized the importance of gaming out what happens if something goes wrong, particularly in “high-stake workflows like financial forecasting, procurements, and compliance,” he said, adding, “Agents should be able to defer to humans when the ambiguity or risk is high,.”

‘Jagged Intelligence’: Challenges With Agentic AI 

While previous iterations of generative AI models struggled with so-called hallucinations, Savarese introduced the concept  "jagged intelligence."

“Jagged intelligence actually means that AI is not uniformly smart,” Savarese said. “So while AI can perform tasks which are clearly similar—to a certain degree—to tasks that humans can do, it can struggle in an idea where sometimes there is more complex reasoning, some level of structure, or sometimes even common sense reasoning. It’s something that some AI systems can fail to achieve.”

Savarese said it’s of “paramount importance” to understand those limitations and adjust the trust boundaries within which agentic AI operates within an organization.

Ensuring Trust at PayPal

Jamie Miller, CFO and COO at PayPal, understands well the value of trust.

“Compliance and trust is at the heart of everything we do,” she said, noting that the ecommerce giant processes transfers, completes transactions for consumers and merchants, tracks product delivery, and handles various steps along the value chain. 

Miller said it’s also important to understand how many AI agents operate within your organization, just as business leaders would consider the number of employees they hire. “We’re going to have to evolve our brains a little bit to solve some of that and think about how we govern,” she added.

AI agents have already found a place at eBay, which takes invoices from various vendors in multiple formats to “very rapidly pull the data, and get it into one harmonized set” so it can be processed in myriad ways, Miller said. Yet it also happens under very strict parameters, she added. 

“As CFOs, we have responsibilities for the financial statements, we have responsibilities for compliance, and it's really important that we understand the objective of what’s happening, and that we have the right sort of checks and balances and controls to make sure it’s happening the way it should,” Miller said. “But I think with the right framework, with the right sort of oversight, there’s an unlimited amount of space where this can be applicable.”

Miller envisioned AI agents optimizing the operational side of the business, initiating new product offerings, strengthening risk management, and improving customer service. 

“Agentic AI is a huge game changer—and I think we are just now beginning to sort of see some of the art of the possible,” she said.

How Team Car Care Counts on AI

At Team Car Care, operator and franchisee of the Jiffy Lube brand, AI agents have been tasked with reconciling a high volume of inventory transactions. “It’s a lot of data, and getting the agents to be able to process that much data has been our biggest challenge,” said CFO Matt Castonguay. “But it’s promising, and we expect to see ROI pretty soon.”

Additionally, Castonguay said, Team Car Care uses AI “to help customers understand, one to one, what exactly they need to help maintain their car,” both at the time of service and to help keep them coming back.

While new technology often comes with outsized promises, Castonguay expects outsized results from agentic AI. “It’s going to revolutionize all of our teams and change everything,” he said. “Coming from a private equity- backed company, that’s always a focus: How are you finding more cost? How are you cutting it out? That’s critical for us.”

Miller added that the imperative for growth is universal across companies of all types, adding that business leaders should understand AI’s potential to provide that competitive advantage. “If big companies don’t really attack this and start to figure it out and incorporate it soon, it opens up the opportunity for the competition to figure it out first,” she said.

Learn how growth champions leverage a scalable, AI-first technology stack, a focused strategy, and people and cultural best practices to fuel long-term success. Download the report produced by Boston Consulting Group (BCG) and Workday.