AI-Powered Virtual Assistant for Digital Banking
- Customer
- ABB Bank
- Project manager on the customer side
-
- IT Provider
- ABB Bank https://abb-bank.az/
- Year of project completion
- 2025
- Project timeline
- November, 2023 - December, 2025
- Project scope
- 500000 subscribers
- Goals
- The goal of the project was to create a multilingual, AI-powered virtual assistant that could handle the growing volume of customer requests in ABB Mobile without increasing human resources. The solution aimed to automate repetitive support tasks, provide instant and accurate responses 24/7, and enhance customer satisfaction by reducing waiting time from minutes to seconds. Additionally, the project sought to personalize customer interactions — delivering relevant financial insights, proactive recommendations, and product offers tailored to each user. Ultimately, the initiative was designed to improve operational efficiency, reduce service costs, and strengthen ABB’s position as a digital-first, AI-driven bank in Azerbaijan.
- Project Results
-
The project has significantly improved both customer experience and operational efficiency in ABB’s digital ecosystem. The AI assistant now serves over 420,000 users, handling 1.9 million queries annually and resolving 61% of all requests end-to-end without human involvement.
Average response time dropped from 20 minutes to 9 seconds, while accuracy reached 95%. Through automation, ABB saves approximately 79,000 AZN per month, equivalent to 31 agent salaries and 130,000 working hours annually.
Additionally, 60% of negative interactions were converted into neutral or positive ones, increasing customer satisfaction and loyalty. The assistant has evolved from simple FAQ responses to real banking operations such as card blocking/unblocking, balance checks, cashback inquiries, and personalized product offers, positioning ABB as a regional pioneer in AI-powered banking.
The uniqueness of the project
The project’s uniqueness lies in its multilingual GenAI infrastructure, ensuring complete data privacy while delivering human-like conversations in Azerbaijani, English, and Russian — a first in the region. Unlike conventional chatbots, it performs real banking operations such as card blocking/unblocking, balance checks, and transaction inquiries through secure API integration. The assistant can proactively start conversations, generate personalized financial offers, and adapt responses using contextual awareness and sentiment analysis. Additionally, the system was developed entirely in-house using an orchestration and RAG-based architecture fine-tuned on internal data, allowing ABB to achieve a 95% response accuracy rate without reliance on external cloud models.
- Used software
-
The project leverages a combination of AI, data orchestration, and infrastructure technologies to ensure scalability, security, and accuracy:
-
AI & NLP Frameworks: GPT-based transformer models (GPT-4, GPT-4.1, Geminiand etc.), RAG (Retrieval-Augmented Generation), and intent classification models for context adaptation.
-
Integration & Orchestration: Custom-built orchestration layer for multi-intent routing and dynamic API management.
-
Infrastructure: Docker containers orchestrated via OpenShift; PostgreSQL and QlikSense for data storage, monitoring, and KPI visualization.
-
Backend & APIs: Python (FastAPI), internal banking APIs for real transactions (card block/unblock, balance, cashback).
-
Monitoring & Analytics: QlikSense dashboards for AI-to-human ratio, missed queries, accuracy metrics, and sentiment tracking.
-
- Difficulty of implementation
- The main challenge during implementation was building accurate >context understanding> for multilingual customer queries. At the early stage, response accuracy averaged >70–75%> due to the variety and ambiguity of user questions and the similarity of numerous banking products. The team spent extensive time designing >context management>, >intent classification>, and >retrieval-augmented generation (RAG)> logic to ensure precise responses. Developing a robust orchestration layer and aligning customer data across systems also required complex integration efforts. Through continuous fine-tuning and iterative testing, accuracy was improved to >95%>, while maintaining high security and compliance standards.
- Project Description
-
The project represents a major step in ABB’s digital transformation journey — the development of a multilingual, GenAI-powered virtual assistant that fundamentally redefines customer service, engagement, and automation in digital banking.
The assistant supports Azerbaijani, English, and Russian languages and is integrated into ABB Mobile, serving over 420,000 active customers. Its primary goal is to ensure that users receive immediate, accurate, and personalized responses without waiting for live agents. The assistant handles both customer support and sales functions, creating a seamless experience that bridges service automation and intelligent personalization.
The AI assistant currently processes 1.9 million queries annually (out of 3.2M total) and successfully resolves 61% of all requests end-to-end without human intervention. It maintains a 95% response accuracy rate and converts 60% of initially negative customer interactions into neutral or positive ones, contributing to higher satisfaction and loyalty.
Beyond FAQ responses, the system now performs real banking operations through secure internal APIs — including card blocking/unblocking, balance inquiries, cashback information, and transaction-related support. It also generates proactive insights by initiating conversations, reminding customers about product activations, and delivering personalized offers such as pre-approved loans or deposit opportunities.
From a technical standpoint, the project employs retrieval-augmented generation (RAG) and intent recognition models to ensure contextually aware, human-like conversations. The system architecture uses Docker and OpenShift for container orchestration, with FastAPI powering the backend. Data from customer interactions are securely stored and analyzed in PostgreSQL, while QlikSense dashboards track KPIs such as AI-to-human ratio, response time, missed queries, and sentiment trends.
The project was developed entirely in-house by a five-member cross-functional team, combining expertise from AI, product, and business units. Over 4,200 man-hours were dedicated to building, testing, and refining the solution. Continuous improvements and retraining cycles have increased accuracy from 75% in early testing to 95% at present.
Overall, the initiative has delivered measurable operational impact, achieving monthly cost savings of 79,000 AZN (equivalent to 31 agent salaries), reducing average response time from 20 minutes to 9 seconds, and elevating ABB’s position as a leading innovator in AI-driven financial services. Future phases aim to extend the assistant to ABB Biz and ABB Home, enabling even deeper integration into everyday financial operations and expanding toward a fully conversational, AI-first bank experience.
- Project geography
-
The project currently operates nationwide across the Republic of >Azerbaijan>, supporting ABB’s digital customer base of more than >2 million users> through the >ABB Mobile> application. With >420,000 active monthly users>, the AI assistant delivers multilingual (Azerbaijani, English, Russian) customer support and personalized financial guidance across all regions.
In its next phase, ABB aims to scale beyond local success — expanding the solution to >ABB Biz> (corporate clients) and >ABB Home>, while evolving into a >fully conversational AI bank>. This vision positions ABB not only as a digital leader in Azerbaijan but also as a >regional and global innovator> in intelligent, human-like banking interaction.
- Additional presentations:
- Global CIO 28102025.pdf