CASE STUDY: City Union Bank Envisions its Integrated Robo-Banking Ambitions

City union bank deployed a conversational AI platform that can not only assist its customers online but also train its offsite branch counterpart Robot using the training data collected from the conversational AI platform.

The Solution Highlights

CUB deployed what they term as a conversational AI banking assistant-also named Lakshmi with the help of Avaamo’s multi-layered conversational platform. The conversational platform is powered by proprietary AI algorithms combined with Deep Domain ML models that learn and perform multi-turn conversations and execute judgment intensive tasks just like humans.

Lakshmi can currently handle queries ranging from fraud reporting, ID verification, processing of e-wallet transactions, product advice on loans as well as more mundane tasks like updating personal information or locating an ATM.

Notions of Old World Banking

• Rule-based Anti-Money Laundering (AML) has the high false positive rate
• Fraud detection mechanisms adapt poorly to ever-changing fraud patterns

New World Banking Assistance

• Pattern-based AML algorithms reduce false positives
• Virtual help desk improves the customer experience
• Fraudulent behaviour is detected faster and with better accuracy
• Product recommendations are personalized and offered in real-time

More About the Solution:

One of the Major highlights of City Union Bank’s conversational AI banking assistant is the strength of its cognitive platform which is a multi-layered highly differentiated set of technology modules addressing Natural Language Understanding, conversation logic, and machine learning.

1. Understanding Intent

The cognitive platform is designed to correctly understand the questions by identifying the user’s intent. The Cognitive platform can parse the question for intent, syntax, spell check and even emotional intonation like hate, frustration, exasperation to understand context.

“Help! My card is stolen!”

“How do I update my address”

“I want to know more about the personal loan offers!”

2. Countering Ambiguity

Generating a relevant, accurate, and useful response may involve different disambiguation techniques. Avaamo’s AI platform can reference conversational context, utilize background information, or actively seek user clarification to quickly discern the user’s true intent.

3. The response in Context and establishing identity

This means usually generating a response that is relevant to the intent. This will involve understanding the users' actually need in some cases using (1) Disambiguation and (2) driving complex multi-turn actions (3) detecting and repairing a temporarily derailed conversation.

The AI platform’s learning engine based on past actions will create a response in context. An important part of the process is to ensure that the assistant is thoroughly tested on customer authentication. This process requires the AI assistant to ask probing questions and establish customer’s right identity before providing the appropriate responses.

“Would you like to talk to a live agent?”

“Can you authenticate your details to update your street address?”

4. Executing an Action

This means usually inputting data using a conversational interface into a transaction based back-end system to record an inquiry, complaint or in some cases like Banking –initiate a transaction, transfer funds, or in personal account management- execute a change of address or change in plans. Integration to third-party services and transactions systems, ensure the questions are now answered, responded and “closed”.

The conversational AI platform leverages multi-modal integrations to interface with enterprise systems and executes these actions.

Enabling the Platform to Easily Record Conversations and Train the Banking AI Models

Machine learning allows the platform to look for data and decipher nuances to generate intelligent conversations much like a human.

• Classifier: Allows the platform to auto-generate, curate, and classify variations of training data by feeding chat transcripts, previous resolutions, help guides or other historical information.
• Entity Extractor: Identifies specific business objects or topics that bots can be trained to recognize to ensure intelligent conversations.
• Assisted Learning: Enables the platform to present users with a set of alternative choices

to resolve ambiguity and commit this to user“ assisted” learning

• Relationship Extractor: Analyze conversations to find relationships between nouns, verbs, and subjects. For example,“How do I make international payments?”The AI platform can now understand the users' primary goal is to send money based on the sentence and adjust responses.

Results and the road ahead-

Handling over 1,00,000 queries per month

Lakshmi is currently handling her range of roles well with a response accuracy of over 95% on the content that she is trained on. She is also continually improving as she addresses at least over 1,00,000 customer queries per month.

Reduced Operating Expenses

Since the time she was deployed, Lakshmi has drastically improved customer experience, customer support quality and reduced operating expenses by nearly 55%. She is expected to sharpen her skills and further reduce CUB’s operating expenses by nearly 60% in the next couple of years.

Reuse the conversations for its Branch Based Robots

Lakshmi’s previous conversations and trained AI models are being reused to train City Union Bank’s branch-based robots. These robots have immensely gained knowledge from its online counterpart by applying techniques of empathy, disambiguation and multi-turn conversations to deliver an easy and engaging banking experience.

In the next few months, City Union bank plans to program both the conversational AI platform and its banking robot to converse with customers in Tamil. Overall, Lakshmi has enabled City Union bank to further its robotics vision and secure a competitive advantage over other banks and financial institutions in India.