Agentic AI in Banking: Mapping the Autonomous AI Frontier

As the financial sector grapples with the pressure to digitize, the limitations of early AI adoption have become clear. While Generative AI ("GenAI") has transformed content creation, it cannot single-handedly manage the rigorous, high-stakes workflows of a modern bank back-office. This operational challenge was the focus of a recent panel discussion at Techsauce Global Summit 2025, "The Autonomous Future of Banking Powered by Agentic AI," which brought together experts from corporate venture capital and enterprise technology. The consensus was clear: "Agentic AI"—systems that execute actions rather than just generating text—is no longer a futuristic concept but a strategic necessity for achieving speed, accuracy, and scalability in banking.

The panel featured two key experts: Palida Artispong, Acting Managing Director of Krungsri Finnovate, and Danny Centen, Key Accounts Director of fileAI. Sathapon Patanakuha, CEO of Guardian AI, moderated the discussion.

Why the "Human Loop" Is No Longer Effective

Before the panel explored solutions, they first painted a stark picture of the current banking predicament. The industry is caught in what Danny Centen illustrated as an efficiency crisis driven by "unstructured data"—the messy reality of PDFs, images, ID scans, and handwritten forms that traditional automation cannot handle.

The problem is that current workflows force highly skilled humans to act like robots. The first major challenge is the "Unstructured Data" bottleneck. Palida Artispong revealed that nearly 50% of customer-submitted bank statements fail to meet basic requirements, such as correct date ranges or formats. This forces staff to manually check documents, creating massive delays.

Danny Centen also provided another anecdote to illustrate this friction: a personal health insurance claim that took 32 days to process. The delay wasn't due to complex decision-making, but simply "human availability." Every time a document needed a basic verification, the workflow halted.

My customer experience was extremely hampered... I am 100% sure if we could automate these low-value tasks, the customer satisfaction was a lot higher,

— Danny Centen

This leaves banks with a critical, unsolved gap: a need for a workflow that is fast and digital, but capable of understanding messy data like a human without the fatigue or error rate. This, the panel argued, is the precise challenge that Agentic AI is positioned to solve.

Agentic AI: Bridging the Gap Between Insight and Action

For many, the term "AI" currently evokes images of chatbots writing poetry and reports. The panel explained why Agentic AI is a completely different proposition. Unlike GenAI, which functions as a creative co-pilot, Agentic AI functions as a "Process Steward."

It's not just technology tools anymore but rather... partners that hold hands in hand with business,

— Palida Artispong

Key advantages that Agentic AI could offer include:

• Autonomy: It doesn't just read a file; it executes the workflow, injecting data directly into ERP systems.
• Localization: A major issue in Southeast Asia is the local language and script. Through Krungsri Finnovate’s strategic support, fileAI has mastered the ability to read Thai handwriting, a feat that global models often fail to achieve.
• Scalability: With Virtual Banking licenses becoming a reality in Thailand, banks can no longer hire 10,000 back-office staff to scale. Agentic AI allows for exponential customer growth without linear headcount growth.

The Investment and Strategic Horizon

The race for autonomous banking is already on. Palida Artispong detailed Krungsri Finnovate’s role not just as an investor, but as an "orchestrator." By backing fileAI, the CVC acts as a gateway, bringing global-tier technology to solve hyper-local operational problems across Vietnam, Indonesia, and the Philippines.

Danny Centen noted that this investment serves as a "ginormous vote of confidence." It signals a shift where banks are moving beyond testing AI to hardening it for enterprise use. The goal is "hands-free automation," allowing the bank to handle sudden spikes in volume—such as an acquisition or a new product launch—without needing to retrain the entire workforce.

The 5 Metrics of Success

The conversation turned to how banks should measure this new technology. The panel rejected vanity metrics in favor of hard operational KPIs. Danny Centen outlined the five metrics that define success in 2025:

1. Granular Cost Per File: Drastically lowering operational expense by reducing processing time from 12 minutes to 20 seconds.
2. Speed of Execution: Eliminating the wait times of conventional approaches.
3. Agility: The ability to ingest a new invoice format instantly, regardless of whether the data is upside down or unstructured.
4. Human Override Rate: Keeping the error rate below 0.5%.
5. Straight-Through Processing (STP): The keystone metric of success

If you cannot achieve a straight-through processing rate of over 90%, then I would ask myself the question, why did we automate this?

— Danny Centen

A Symbiosis of Value

Ultimately, the discussion painted a picture of a transformative technology that is arriving at a critical moment. The consensus was that Agentic AI is not about replacing bankers, but about liberating them from "low-value tasks."

By adhering to the 80/20 rule (let AI handle the 80% of operational grunt work), humans can focus on the 20% of high-value strategy and relationship building. As Palida Artispong urged in her closing remarks, "Doing nothing is not an option." The technology is moving fast, and banks must start orchestrating their autonomous future today to stay relevant.

Based on the session: “The Autonomous Future of Banking Powered by Agentic AI” at the Techsauce Global Summit 2025.