How Nepali Banks Are Using AI for Customer Service and Fraud Detection

A few years ago, "digital banking" in Nepal mostly meant being able to check your balance without visiting a branch. Today, ai in nepali banks means something considerably more specific: chatbots that handle thousands of customer queries simultaneously, fraud systems that can block a suspicious transaction before it completes, and — as of late 2025 — an actual central bank framework governing how all of this is allowed to work.

This post walks through what's actually deployed at named Nepali banks right now, what's still mostly marketing language, and what Nepal Rastra Bank's new draft AI Guidelines mean for where this is all heading.

The State of Banking Automation in Nepal, Honestly

Before getting into specific banks, it's worth setting accurate expectations, because a lot of vendor content makes this sound further along than it currently is.

According to a detailed comparison of Nepali bank chatbots, only about eight commercial banks in Nepal currently have a chatbot deployed at all, and of those eight, the large majority are rule-based systems — meaning they work off pre-set menus and keyword matching rather than genuine natural language understanding. As of the most recent comparison available, only one bank's assistant was identified as a fully AI-powered conversational system, with one additional bank running a hybrid that blends rule-based menus with AI-driven responses.

That doesn't mean banking automation nepal-wide is overstated — it means the maturity curve is wide. Some institutions are running genuinely sophisticated systems; many others have a basic FAQ bot and call it "AI" in their marketing copy. Both things are true at once, and a useful guide should be honest about which is which.

Customer Service: What's Actually Running at Named Banks

Here's a direct look at the customer-facing chatbots actually deployed across Nepal's banking sector, based on published information about each system:

Bank	Assistant	Type
NIC Asia Bank	Saathi	Generative AI — Nepal's first of its kind in banking
Global IME Bank	Viva	Rule-based, with Nepali/English language toggle
Siddhartha Bank	Sid	Hybrid — rule-based menus + AI responses
Nepal Bank Limited	NBL Mitra	Menu-driven
Kumari Bank	Text and voice support

Beyond customer-facing chat, AI is also restructuring internal banking operations. One documented example: a Nepali AI vendor built custom internal employee assistants across 14 different departments for Kamana Sewa Bikas Bank, designed to instantly answer staff questions about HR policy, IT troubleshooting, and the latest Nepal Rastra Bank circulars — the kind of internal knowledge work that previously required staff to track down the right person or document manually.

The KYC and onboarding layer is where AI automation is arguably furthest along industry-wide. Optical character recognition tools can now extract data directly from citizenship cards, passports, or driving licenses, cross-reference it automatically, and flag inconsistencies for human review — turning a process that used to take days into one that takes minutes, before a human finalizes approval.

Fraud Detection: The Less Visible, More Consequential Use Case

Customer service chatbots are the part of bank AI you can see and talk to. Fraud detection is the part that mostly works silently in the background — and arguably matters more, given Nepal's rapidly escalating cybercrime numbers.

According to Nepal's Cyber Bureau, cybercrime cases reported nearly doubled between fiscal year 2022-23 and 2023-24, climbing from 9,013 cases to 19,730. That trajectory is precisely why Nepal Rastra Bank's Financial Intelligence Unit has moved to formally update its own monitoring guidelines.

How AI-based fraud detection actually works in practice, based on how it's implemented across the deployments documented in Nepal's market: the system continuously analyzes transaction behavior in real time, looking for patterns a human reviewer would likely miss at scale — an unusually large payment at an unusual hour, a login attempt from a location inconsistent with the customer's typical behavior, or a transaction velocity that doesn't match established patterns. When the system flags something as suspicious, it can block the transaction automatically and alert both the bank and the customer, often before any money has actually moved.

This is also where alternative credit scoring is starting to matter for financial inclusion specifically. Traditional credit scoring in Nepal is limited by how many people simply don't have an extensive formal credit history. AI-based alternative scoring can incorporate signals like transaction velocity, utility bill payment consistency, and cash flow patterns — giving banks a more complete risk picture for customers who'd otherwise be invisible to conventional underwriting, which matters substantially in a market with Nepal's geography and historical banking-access gaps.

Nepal Rastra Bank's Draft AI Guidelines

This is the most underreported part of the entire story, and it's the part that will likely shape everything described above going forward.

In late 2025, Nepal Rastra Bank released an 11-page draft of formal AI Guidelines for public comment, marking what one detailed analysis described as one of the central bank's most forward-looking regulatory moves in recent decades. The draft's premise is refreshingly direct: NRB explicitly acknowledges that Nepali banks are already using AI, whether formally acknowledged or not — from automated credit scoring to fraud detection engines — and that what's been missing is a clear governance framework for technology that's already quietly embedded in the system.

What the draft guidelines actually cover:

• Scope — applies to all NRB-licensed entities, including commercial banks, microfinance institutions, and payment service providers, across credit scoring, fraud detection, customer support, risk management, and compliance.
• Governance — requires board-level oversight, including an executive AI sponsor and a cross-functional AI risk committee at each institution.
• Risk classification — distinguishes between high-risk and non-high-risk AI applications, with stricter requirements for the higher-risk category.
• Transparency and explainability — AI decisions affecting customers need to be explainable, not opaque "black box" outputs.
• Data privacy — explicit alignment with Nepal's Privacy Act 2075.
• Practical compliance steps — cataloging every AI system in production or pilot, documenting data sources and limitations, defining fairness metrics, and building customer-facing disclosure and appeals processes.

This regulatory framing matters because it reframes the entire customer service and fraud detection discussion above: these aren't isolated bank-by-bank experiments happening in a vacuum. They're increasingly going to operate inside a defined accountability structure — which is good news for customers who've been understandably wary of black-box automated decisions affecting their money.

What This Means for You

The Bottom Line

AI in Nepali banks is real, but unevenly distributed — a handful of institutions like NIC Asia and Global IME Bank are running genuinely sophisticated customer-facing systems, while many others are still at the rule-based-menu stage. Fraud detection is arguably the more consequential application right now, given Nepal's cybercrime cases nearly doubling year-over-year and NRB's Financial Intelligence Unit explicitly updating its monitoring framework to address AI-enabled financial crime. And running underneath all of it, Nepal Rastra Bank's draft AI Guidelines represent a genuinely significant regulatory moment — an explicit acknowledgment that AI is already operating inside the country's banking system, paired with a real attempt to build governance around it before problems emerge rather than after.

For customers, the practical reality is a banking experience that's gradually getting faster and more automated, with real fraud protection running quietly in the background. For the banks themselves, the message from the regulator is increasingly clear: build the governance structure now, because the era of AI in banking operating without a formal Nepali framework is ending.