Case study: Learn how a "millennial-focused" bank uses AI to drive customer engagement Payjo

Blog > How Intelligent are Banking Chatbots in Reality?

How Intelligent are Banking Chatbots in Reality?

“Tell me a joke.”
How would chatbots in banking websites respond?
Here is a screenshot from a chat with SBI Intelligent Assistant (SIA).

Chatbots in Banking – A Screenshot of a Conversation

State Bank of India is India’s largest bank and its chatbot SIA is powered by Payjo.

Does this mean that chatbots in banking are ready to have meaningful conversations with humans? Not yet. As promising as it sounds, we are yet to reach precision in AI technology. Businesses that go over the top with their expectations from a chatbot often find themselves let down. And, so do their customers.

Even after multiple promising releases of AI chatbot APIs by tech mammoths such as Google, Microsoft, Amazon, Apple, and IBM the so-called smart bots are yet to reach a fraction of human to human communication capabilities. Aren’t we still making our calls ourselves after the Google I/O 2018?

Now the question is how will you find a perfect balance between a bot that is intelligent and a bot that walks the talk? In this article, we will discuss parameters that can help you evaluate and decide on an AI solution you want:

1. Scope

The scope of the bot is its Job Description. These are points on which it needs to deliver. A good scope definition for a bank AI chatbot will look like this:

Answer basic customer and prospect queries from the FAQ section. Fallback: customer care
For advanced customer queries, ask the customer to sign in
Associate a ticket to the customer’s issue

Case 1: If the response is available in the database
Respond and ask the customer to confirm resolution.
If yes end conversation else, re-engage.

Case 2: No predefined response available in the database
Send an email to the customer care team with the ticket and include conversation as an attachment.

Notify customer of the expected time within which they’ll receive a call from a representative of the bank.

Note that all performance metrics (KPIs) should be based on the process defined for the scope of the bot. In the above case, the bot needs to be updated with the user’s communication history, understand the context and offer remedies based on predefined responses. It is unjustifiable to expect the bot to think and respond to undefined client’s queries.

2. Data Source

The number of data sources along with the quality and quantity of data decide two things:
1. Basis available for developers to create machine learning models for the future.
2. Streaming data that will help the bot to handle unusual requests.

Since financial institutions have a tremendous amount of data, they are allowed to expect detailed answers from bots. Note that most of the data might be available only after the user logs in. Make sure that your bot doesn’t try to solve a concern without access to appropriate data.

3. Channels

Initially, companies expect to have a dedicated bot for each channel – Facebook Messenger, Google Assistant, Website etc. But having a different chatbot for each channel would create complexities. For instance:

– The customer would start a conversation on one platform and expect the bot to refer to history while using another platform

– All the natural language APIs might not be available on distinct platforms

– Audience type differs with each channel. Facebook is concentrated with more casual and digital-savvy users than bank website. Hence, you’ll need to tailor responses to each platform.

– It will increase the time and cost involved in development and testing; the load on the server increases manifold.

You might want to start with one platform depending on your target audience. It’s always easier and simpler to expand your reach once you figure out a winning formula.

Payjo has successfully implemented omni channel AI chatbots in banking that are context-aware when switching channels, even when the conversation moves from a text channel (website chat) to a voice channel (call center).

4. Flow vs Intent

Flow-based bots are those that expect you to follow a pre-defined way of communication. It gives the user a series of multiple-choice questions where each question is based on the option chosen in the last answer. These don’t need artificial intelligence as the level of options are narrow.

Intent-based bots, on the other hand, serve on the basis of user requests. Since the user can ask any query the possibilities of failure are much more. Thus, these bots need contextual understanding and advanced AI to handle complex communications.

For financial institutions, we recommend a hybrid model. You can give customer options after they initiate a conversation and fallback to intent-based model when there aren’t any pre-defined options available to the user.

These pointers together will help you determine the level of intelligence and complexity of conversations your bot should be able to handle. Now that we have discussed the major parameters for deciding the level of intelligence, let’s discuss the failures. What if your bot fails to deliver or worse keeps circulating the user in an endless loop?

This is where your exit plan kicks in. Exit feature is the default fallback option available to both the user and the bot. In case if either one is unable to conclude the conversation or arrive at something meaningful this option should be readily available so the conversation can be moved to a human agent.

Leave a Reply

Your email address will not be published. Required fields are marked *

Similar Stories

Top banks that provide Intelligent Personal Assistant (IPA) for their customers and employees

timestamps May 16, 2018

Top banks that provide Intelligent Personal Assistant (IPA) for their customers and employees

Read full story >