Page 38 - CMA Journal (July-August 2025)

Focus Section

• A third digital agent pulls campaign data from modernize infrastructure. In other words, legacy
marketing. systems need to be phased out, and data moved to
the cloud. This is essential for building the right
• A fourth digital agent synthesizes the work of the
pipelines to migrate and store data efficiently.
three into a report and strategy.

Traditional AI vs. Agentic AI

Traditional AI Agentic AI
Requires prompt from Takes action to achieve
human and/or a defined outcome
explicitly programmed often without direct
human input.
rules.
• Perceiving environment
• Pattern recognition
• Reasoning
• Prediction • Executing actions
• Classification within • Learning from outcomes
Structured datasets

These steps will flow much like human workflows, with 3) Drivers of Momentum: Start small, releasing AI
digital agents working together and double-checking agents to take on low-risk responsibilities; e.g.
their work. They will escalate an issue only when it’s truly workflow coordination for financial planning or
needed. The result is faster, more accurate decisions with period-end closing. Humans can validate any risk in
fewer manual steps. This new technology allows finance the chain of operations. Gradual confidence and
teams to move from being simple scorekeepers to experience can lead to broader application.
strategic orchestrators.

Organizations such as Alaska Airlines, Puma, and Best Buy How Fintech Is utilizing Agentic AI
are already tapping into this capacity to re-engineer
operations and improve customer experiences. The While many industries are only starting to explore
intersection of LLMs with multimodal inputs (text, Agentic AI, fintech is already ahead. It is showing what’s
images, audio, video) is accelerating this change. possible through real-world applications:
Laying the Groundwork for Agentic Finance 1) More Intelligent Lending: Companies such as
Upstart are using Agentic AI to look at many data
Agentic AI will not be plug-and-play. CFOs can set the
points beyond traditional credit. Their agents access
stage in following ways to foster successful adoption:
all sorts of data about a person, including but not
1) Strategic Roadmap: Avoid the trap of isolated, limited to, education, employment, gender, age, etc.
self-contained pilot programs. Instead, build a clear This allows Upstart to make better-informed lending
and compelling vision for your organization's digital decisions for many individuals who currently get
transformation—and position Agentic AI as one of excluded from systems used to identify
the key tools to deliver on those ambitions. Map out creditworthiness.
where you are today versus where you want to be,
and establish a roadmap that connects the dots 2) Proactive Fraud Detection: Rather than simply
with intention. notifying you of a suspicious transaction after it has
occurred, Agentic AI systems are able to monitor
2) Data Infrastructure: Poor data quality remains the
behaviors and detect fraudulent actions in real-time.
biggest barrier to successful AI deployment.
They can detect new types of fraud because they are
According to KPMG’s 2025 AI Pulse Survey, 85% of
designed to learn and can decide to collaborate with
business leaders believe that unreliable data will
other agents and take automatic action. For instance,
hold them back from realizing AI’s full potential.
temporarily block a card and notify the customer in
CFOs must play a central role in shaping a data
advance before the fraud could take place.
strategy—collaborating with IT and business units to
36 ICMA’s Chartered Management Accountant, Jul-Aug 2025