Most banks still rely on rule-based bots, but few are building real AI agents: Maveric Systems CEO

The banking and financial services industry is at a critical point in its AI journey. As regulatory clarity improves and AI technologies mature, institutions are rethinking how they integrate AI, not just to automate tasks, but to drive core business performance.

Maveric Systems, a digital transformation partner for banks, is closely observing this shift. In a conversation with TechCircle, Ranga Reddy, Co-founder and CEO of the company, explained how banks are moving from isolated AI use cases to aligning AI with business outcomes. He shared insights on where the industry is headed, the challenges around data and talent, and how leading banks are redesigning their operating models to integrate AI into both technology and operations.

Edited Excerpts:

As AI is becoming central to strategy in BFSI. From your perspective, which use cases are leading banks adopting most?

People are focused on use cases, not business cases. While AI technology has matured and it makes sense to experiment with use cases, it's time to shift the focus toward business impact. In banking, regulations around responsible AI are clearer now. So instead of just looking at AI for cost or coverage optimisation, the focus should be on the four core business metrics that AI can influence. This shift, from use case to business case, is what many banks are starting to make.

For example, in areas like lending or payments, banks are now asking how they can lead in those spaces. That requires rethinking technology, processes, and teams. If AI is embedded into that broader transformation, its impact will be more meaningful than if it's simply implemented in isolation.

Most mature banks that have been experimenting with AI over the past 18–24 months are now making this shift. Others, who are just starting out, are still working at the use case level. The more advanced ones are moving toward defining business outcomes first and then aligning AI, technology, and operations to meet those goals.

Functionally, AI adoption is strongest at the front end (like customer service) and back end (operations), where tasks are typically lower in complexity and cost. The middle office, which involves analysis and decision-making, hasn’t yet adopted AI at the same level.

Another area of AI application is in technology operations. Some banks are experimenting with AI-driven software development life cycles (SDLCs), covering everything from requirements to testing. But because of regulatory constraints, most banks still use traditional SDLC processes and only apply AI tools within that structure. A few leading banks have adopted more AI-native SDLC models in specific business areas, helping them develop applications much faster, cutting timelines from around 10 months to just 3 in some cases.

AI is also being used in application maintenance and support, commonly referred to as "run the bank." So broadly, AI is being applied in two main areas: banking operations and technology. In operations, adoption is further along. In technology, AI is being actively explored for support functions, while full integration into development processes is still limited.

Why is the only visible use of AI in customer experience in India still limited to chatbots and automation bots?

As AI continues to evolve, it's important to understand the difference between automation and actual AI. Many chatbots today rely on rule engines, predefined workflows that trigger specific responses based on user input. If a customer asks for A, the system follows a set path to respond. That’s not AI; it’s automation.

In contrast, a true AI agent doesn't just follow preset rules. It can understand context, adapt to situations, and make decisions beyond predefined flows. Most front-end chatbot implementations today are still rooted in rule engines and workflow automation, not full AI.

Here’s an example: A major US bank has deployed a system in customer support that goes beyond standard chatbot functions. It starts by anticipating the customer's reason for reaching out, using data from past interactions and recent communication. When the customer logs in, the system might say, “Are you contacting us about an issue with your billing update?” That’s anticipation based on history and context.

Such a system doesn't just react, it identifies the issue, understands the process needed for resolution, and initiates action. This goes beyond automation and enters the realm of AI-driven agents.

Many companies that initially deployed rule-based chatbots are now exploring how to enable them with real AI capabilities. Current chatbots handle straightforward tasks well, but when enhanced with AI, they can manage more complex scenarios, such as anticipating customer needs and resolving them without human intervention.

While most visible AI work is happening at the front-end, chatbots and virtual assistants, there’s also progress at the back-end. However, middle-office functions are still largely unexplored.

There’s also growing customer frustration. In sectors like banking and airlines, people are noticing that service responsibilities are increasingly shifted onto them through self-service platforms. This shift is creating dissatisfaction, as customers feel they’re being asked to do the work once handled by service staff.

Where is the industry headed with AI, and are banks ready to scale it responsibly given compliance and operational challenges?

Based on what I see among our customers, they generally fall into three categories when it comes to AI adoption. About 40% are in the early stages. They’re making small, exploratory investments aimed at understanding how AI works and what it could mean for them. These aren’t efforts aimed at generating business outcomes yet, the goal here is learning. They’re treating it as an investment in building internal understanding before committing to larger initiatives.

Another 40% have moved beyond that initial phase. They’ve learned enough to start putting the right foundations in place. This group focuses on three main areas. First, they’re setting clear internal policies around the responsible use of AI, and working to ensure these policies can be enforced automatically across projects. Second, they’re reorganising their data so it’s in a form that AI systems can use effectively. Today, much of the data, including policy documents, PDFs, and other unstructured formats, needs to be vectorised and made accessible to AI models. Third, they’re consolidating platforms and tools. After experimenting with various AI tools and platforms, they’re now narrowing down to a core set that will support future AI work, selecting what will support operations, development, and automation across the enterprise.

Then there’s a third group, around 20%, who are focused on applying AI to drive measurable business performance. They’re looking at how AI can reshape specific lines of business, such as payments, lending, or mortgages. This includes rethinking roles, processes, and the underlying technology stack. They’re revisiting how people work, how processes are structured, and what systems need to change in order to fully integrate AI and automation into operations. This group is not just adopting AI, they’re using it to redefine how the business runs.

Importantly, this maturity isn’t limited to large banks. Fintechs and regional players are also represented in the most advanced group. It’s not about size, it’s about focus and direction.

What challenges are banks and fintechs facing in using AI for regulated areas like AML, KYC, or transaction monitoring, and how are they addressing them?

Among the top 20% of organizations that are serious about driving business results through changes in process, people, and technology using AI, the main challenge is talent. They typically have a strategy and funding in place, but progress stalls without the right people.

The second challenge is data. Specifically, they aren’t ready to organize and prepare data in a way that AI systems can use effectively. This includes both structuring the data for AI consumption and ensuring appropriate safeguards are in place to protect customer information.

The third issue is the gap between technical and industry knowledge. Many understand AI well, but lack a deep understanding of the banking sector. The problem is applying AI in a way that fits the banking environment. AI has developed mostly in big tech, and translating those approaches to a banking context is difficult.

Where are the biggest data bottlenecks in traditional banks, and how are they modernizing to address them?

The core issue with data in AI applications, especially in banking, is that while data exists both within and outside the organization, insight does not. For AI to function effectively, especially for open-domain, context-aware reasoning using large language models (LLMs), the data needs to be curated in a way that turns raw information into usable insight.

Current data structures are not suited for natural language queries. For example, asking in plain English for a list of customers who have defaulted isn't possible without preparing the data in a format that supports such queries. This requires both structuring and enriching the data.

Information about a customer is available internally, within the bank, and externally, such as from credit rating agencies, anti-money laundering authorities, or records of political exposure. To make effective decisions, both sources must be combined and processed in real time.

Access control is another challenge. Different roles within a bank require access to different subsets of data. While internal data access is somewhat managed, access to external data is more limited. Ensuring appropriate data access, while maintaining compliance and privacy, is a key consideration.

Most publicly available LLMs are trained on open data. For banking use cases, these models need to be trained on enterprise-specific data. This adds complexity around integrating structured and unstructured data, combining internal and external sources, and enforcing access controls for different AI agents while maintaining data security.

In simplified terms, the challenges fall into three areas: handling both structured and unstructured data, combining enterprise and public data, and enforcing data access controls to ensure that AI agents only access what they are allowed to and maintain security.

To support AI initiatives, banks need a robust data architecture that includes responsible AI practices, validation, automation, privacy, and transparency. Public and proprietary data must be managed appropriately. A more detailed discussion on AI for data can be explored further if needed.

How are tier-one banks rethinking data architecture to enable real-time intelligence across siloed systems?

Some are following a portfolio-based approach. For example, a bank might start with wealth management and look at how its current data setup compares to what's needed for that specific portfolio. This is common among tier-one banks. They’re reorganising data with a focus on entire business lines.

Others are taking a use-case-driven approach. These banks haven't yet applied AI for broad business impact. Instead, they handle data separately for each use case. If a use case needs five data sources, two external and three internal, they only reorganise data for that specific need. Once that use case is done, they repeat the process for the next one. This leads to siloed efforts and a lack of a unified data strategy.

Where do you see banking operating models, teams, or platforms heading in the next 3–5 years as AI continues to evolve?

AI is helping drive innovation in business models. But the real value comes when businesses rethink their processes to improve performance and take the lead. Right now, many organizations are unsure how to apply AI effectively. The confusion often lies in treating AI as the goal, rather than using it to support a clear business objective. 

The current shift is toward identifying key business metrics to improve. Once those are clear, companies can reassess roles, operations, and technology to support those goals. This means aligning people, processes, and technology around business performance, not AI for its own sake.

Some advanced AI adopters are already moving in this direction, where business performance drives how AI is applied.