How to Enhance Your Customer Service Chatbot
One of the interesting questions we get is how do bots and live agents work together in customer service. While chatbots are capable of handling many different inbound questions, they sometimes need the support of a live agent to move a conversation along.
By this point, you are likely familiar with bots and their capabilities. In short, a chatbot is designed to have conversations with humans. Typically embedding some level of natural language understanding and decisioning into its capabilities.
At a high level a chatbot works in three steps:
Understand - The bot processes the input given by the user in an utterance from a voice or text message
Retrieve - The bot takes the processed information and searches available databases to find a result it calculates to be the result the user is looking for
Respond - The bot responds with an answer based on an assessment of the database results. The output or response comes in the form of written text or audio message.
This is a very simple description for how a bot hears and responds to a user’s question or input. But sometimes the bot misses the point of the human’s prompt. It is in these cases that a human in the loop strategy can augment the bots performance.
Why Do We Need Human-in-the-Loop
Human in the loop is an approach that bot makers insert into a chatbot’s capability to increase its efficiency. The goal of human-in-the-loop approaches is to enhance the end user’s experience.
The human in the loop capability is a requirement for most bots particularly those asked to provide Tier 1 customer support. Imagine a use case where a customer is looking to simply retrieve their account balance and recognize a potential issue with their account that they would like to fix. While the bot is capable of staging the conversation for a live agent to handle, the bot may continue to listen to quickly answer questions or wait while the human agent resolves more complex issues.
Many chatbots rely on AI and natural language processing to respond to queries. Natural language processing or understanding is based on a predefined structured set of algorithms. Unfortunately, there are times when the data that informs the algorithm is insufficient or the algorithm itself fails to understand the user’s request either due to:
- Scope limitations of the bot
- Language barrier, either from an accent in a voice based interaction or misspellings
- Request is processed unsuccessfully
- Human recognizes they are speaking with a bot and wants to connect with a live agent instead.
Thinking back to the understand, retrieve, respond cycle, the bot may not understand an utterance or have access to the necessary database and may not be able to respond. Bot builders can insert a human option in these instances.
Who Offers Human In the Loop Capabilities
Several of the analytics vendors within the chatbot ecosystem have introduced solutions to this problem. In some cases making it part of their premium offering.
A handful of the dialog management systems have also created features (or suggested the feature will be part of their technical roadmaps) that give customers the ability to “switch” between the bot and a human interface.
These vendors’ features today offer bot builders the ability to pause the bot. This approach allows a human interface to step in and manage conversations, effectively creating a human in the loop experience.
We can think of a number of examples where it may not be necessary or desirable to pause the entire bot. Sometimes bot makers and administrators need to have the ability to step in and address a single thread while letting the bot continue to interact with other users.
As a chatbot developer, we have developed solutions for our customers that allow them to pause individual conversations with the bot and switch that conversation to a human interface. This approach allows the bot to continue to interact with other users while a human addresses the instance that requires differentiated support.
We have developed capability that allows the administrator or human in the loop to redirect conversations to a new flow with the bot that can more efficiently handle their storyline creating an augmented experience for the user while pulling just-in-time human support for cases where the interactions may have been sent down the wrong path.
What Verticals Benefit Most
Bots create a new channel for companies to interact with their customers, business partners, employees and other stakeholders. They are capable of handling complex conversations across a number of use cases. Here are just a few places we can see bots perform and excel as standalone agents but also with a human augmented capability.
Insurance - bots today are helping insurance companies handle customer claims, automate portions of the underwriting, and serve to adjust portions of the policy
Banking - bots today allow customers to ask questions about the financial accounts, initiate transactions and get financial advice. While many of these today are limited to answering basic questions such as account balance, users oftentimes what more.
Real Estate - these bots can offer a host of capabilities today including helping house hunters find local listings and agents.
Travel - bots in the hospitality sector have taken off in popularity. But given the complexities of travel booking and rebooking creates challenges that often times require human intervention.
Personal Digital Assistants - this is my favorite use case and one we use often at Azumo. Booking meetings and venues requires a lot interactions but often times can break down due to slow responses or unclear directions.
The list of potential verticals that bots can support while benefitting from Human in the Loop are endless as chatbots represent a disruptive force that most people are ready for today.
While many AI as a service vendors are working hard to improve their algorithms so that human intervention can be minimized. These services rely on “intelligence at scale” so to speak to create AI services to support their bot customers.
As you can imagine, improving algorithms are heavily dependent upon the data available to train the systems. Lack of data coverage to handle more complex inbound questions is what creates the opportunity for continued live agent support. That is why we implement solutions and encourage our customers to think about including human in the loop capability.