Artificial Intelligence in Digital Banking: A Systematic Literature Review and a Novel Cross-Domain Resilience Framework for Customer Experience, Fraud Detection, and Risk Management
This systematic literature review synthesizes fragmented research on AI in digital banking to propose the novel AIDBRM 2.0 framework, which integrates customer experience, fraud detection, and risk management through cross-cutting mechanisms like a Unified Continuous Trust Score, a Generative AI Governance Sandbox, and Federated Cross-Institution Learning to address existing gaps in explainability and architectural alignment.
Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
Imagine the world of banking as a massive, bustling city. For decades, this city ran on a very old, very rigid map. If you wanted to cross a street, you had to follow a specific, painted line. If you wanted to buy something, you filled out a paper form. This was the "first wave" of digital banking: just turning paper into pixels, but keeping the same slow, rule-bound logic underneath.
But now, the city is changing. It's being rebuilt with a living, breathing nervous system. This is the "second wave," powered by Artificial Intelligence (AI). In this new city, the systems don't just follow rules; they learn, sense, and react in real-time. Think of it like a smart traffic light that doesn't just turn green on a timer, but actually watches the cars, predicts where they are going, and adjusts instantly to prevent jams.
However, building this smart city is tricky. The city planners (the researchers) have been working in separate neighborhoods. One group is obsessed with making the "Customer Experience" neighborhood friendly and fun, using chatbots that sound like friends. Another group is guarding the "Fraud Detection" fortress, building high-tech sensors to catch thieves. A third group is managing the "Risk" district, making sure loans are safe and explaining why a loan was approved or denied. The problem is, in the real city, these neighborhoods aren't separate. A single person walking down the street might be chatting with a robot, buying something, and applying for a loan all at once. If the neighborhoods don't talk to each other, the city gets confused, the security is patchy, and the rules don't make sense. This paper is a detective story that brings these three neighborhoods together to see how they can work as one team.
The Detective Work: What the Paper Found
The author, Durga Prasad Dāsepalli, acted like a super-organized librarian. They didn't just guess; they went on a mission to read and sort through 36 serious, peer-reviewed studies published between 2020 and 2025. They used a strict checklist called PRISMA to make sure they only looked at the best, most reliable research.
After reading all these studies, the detective found three big problems that were holding the city back:
- The "Silos" Problem: The researchers studying friendly chatbots rarely talked to the researchers studying fraud catchers. But in real life, a bank's computer system is one big machine. If a customer is chatting with a bot while a fraud alarm goes off, the system needs to know both things at the same time. Currently, the research treats them as if they are in different universes.
- The "Explainability" Imbalance: This is a fancy word for "being able to explain your answer." The researchers found that when banks say "No" to a loan, they have gotten really good at explaining why (using tools like SHAP and LIME). But when they say "Stop!" to a suspicious transaction or "Here's a special offer," they often use "black box" AI that can't explain itself. It's like having a teacher who explains their math homework perfectly but refuses to explain why they gave you a detention. The paper suggests this is unfair and risky, especially since regulations are starting to demand explanations for all AI decisions, not just loans.
- The "Blueprint" Disconnect: There are architects designing the buildings (the cloud and microservices that run the software) and engineers building the robots (the AI models). But the architects and engineers aren't talking. The architects are worried about speed and stability, while the engineers are worried about accuracy. The paper argues that you can't have a great AI robot if the building it lives in isn't ready to support it.
The New Blueprint: AIDBRM 2.0
To fix these messy neighborhoods, the paper proposes a new, unified blueprint called AIDBRM 2.0 (AI-Enabled Digital Banking Resilience Model, version 2.0). Think of this as a new city plan that connects all the districts with special bridges and shared tools.
The paper suggests three "super-tools" that haven't been used together before in this way:
- The "Unified Trust Score" (The All-Seeing Badge): Imagine every customer has a single, glowing badge that updates every second. This badge doesn't just show if they are a good borrower; it combines their credit score, how likely they are to be a fraudster, and how happy they are with the bank's service. If a customer usually has a high trust score but suddenly makes a weird transaction, the system can look at their whole history and decide, "Oh, this is probably safe," instead of just blocking them immediately. This stops the annoying "false alarms" that make customers angry.
- The "Generative AI Sandbox" (The Safety Net): Banks are starting to use "Generative AI" (the kind that writes stories or answers complex questions) to talk to customers. But these AI models can sometimes "hallucinate" (make things up) or accidentally leak secrets. The paper suggests building a special "sandbox" or testing room. Before any AI answer goes out to a real customer, it gets tested in this room to make sure it's not lying, leaking data, or getting tricked by bad guys. It's like a flight simulator for the bank's AI before it flies with real passengers.
- The "Federated Learning" Club (The Secret Handshake): Fraudsters often work in networks that cross different banks. One bank might see part of a scam, and another bank sees the rest, but they can't share the customer's private data. The paper suggests a "Federated Learning" module. This is like a secret club where banks share their lessons learned (the math behind the AI) without sharing the secret ingredients (the actual customer data). This way, they can all get smarter about catching money-mule scams without breaking privacy rules.
What This Means for Everyone
The paper is very clear about what it hasn't done yet. It hasn't built this new city or tested it with real money. It's a conceptual framework—a really strong, well-researched idea based on what we already know. The author admits that while the blueprint looks great on paper, we need to actually build it and see if it works in the real world.
However, the implications are huge. For the people running the banks, it means they can't just buy a cool AI chatbot and call it a day. They need to upgrade their whole building (the architecture) and make sure their security guards (fraud detection) and loan officers (risk management) are all using the same playbook.
For the regulators, it's a wake-up call: if you demand banks explain their loan decisions, you need to demand they explain their fraud blocks and their product recommendations too. And for customers, the hope is that one day, the bank will feel less like a rigid robot and more like a smart, helpful partner that understands the whole picture of your life, not just one tiny slice of it.
In short, this paper argues that the future of banking isn't about having the smartest single AI; it's about having a team of AIs that can talk to each other, explain their work, and work together to keep the city safe, fair, and friendly.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.