Today there is a lot of hype and buzz around Chatbots, artificial intelligence, and cognitive computing making it difficult to understand what is actually feasible and what can actually have an impact on your business.
TechCrunch recently said, “What’s really surprising is that, despite all this hoopla, it’s hard to find a single chatbot that’s actually a really good product.
Of course, we can wrangle over the definition of what makes a good product, but in its simplest terms, a great product would have three traits:
- It’s simple and easy to use;
- It works well 99 percent of the time; and
- It removes or reduces friction in whatever it was you want it to do. “
In this post we show a simple framework for thinking about chatbots and cutting through the hype. We also apply the framework to a customer service environment and show examples that might even be a really good products.
This post is written for business leaders and goes light on technical details. Talk to your data scientist for more details.
What Can A Chatbot Do?
We have a conversation with a Chatbot and get it to do a job for us. This blog post focuses on two questions:
- What can I ask the Chabot to do?
- What are the responses the Chatbot can give?
Pretend you're hiring a Chatbot to do a job. These questions help us understand how this technology can be deployed in our contact center to perform specific jobs. Think of these as interview questions for the Chatbot.
How Does A Chatbot Do Work?
The technology used to do the job by the Chatbot can be grouped into two broad categories: rules-based and smart.
Rules Based Work
The Chatbot can use rules and heuristics to do its job. Rules are how an IVR operates and does its job. A Chatbot could use a rule set similar to an IVR or web self-service to do its work and then use a chat channel instead of a telephony channel to communicate.
Smart Machine Based Work
Smart machines are also called cognitive computing and AI (Artificial Intelligence.) They use technologies like machine learning to do work. We don’t need to know the details of how smart machines work at this point, just what they can do so we can give them the right task.
Gartner says smart machine technologies:
- adapt their behavior based on experience;
- are not dependent on instructions from people (eg, they can learn on their own); and
- are able to come up with unanticipated results.
So a Chatbot is going to use some combination of rules-based and smart machine-based intelligence to do work for us. Let’s see how Chatbot operates and where each of these technologies fits in.
A Chatbot Framework
We built a framework by looking at (1) if the questions asked to the Chatbot are restricted in scope and (2) how the Chatbot generates a response to those questions.
Open and Closed Domain Questions
A Chatbot provides responses to questions you ask it or performs tasks you ask it to do. Questions and responses are either:
Closed Domain: You can ask a limited set of questions on specific topics. (Easier). What is the Weather in Miami?
Open Domain: I can ask a question about any topic… and expect a relevant response. (Harder) Think of a long conversation around refinancing my mortgage where I could ask anything.
A good example of a closed domain is a website’s FAQs. The FAQs was built from questions most often asked by customers. Going further, a Knowledge Management (KM) system will provide a broader set of responses, but it is still a closed domain system. A traditional IVR is also a good example of a closed domain system with clearly defined goals and menus.
An open domain is very difficult because the customer can take the conversation anywhere. An example a Twitter conversation which can go anywhere. There is a lot of research being done in the open domain area and we'll likely see technologies we can put into operations in the future.
Here we are talking about the topics of the conversation. Understanding the question asked is another challenge we leave for another time.
If you like this post, subscribe to get email updates for this blog.
Retrieval and Generative Responses
Once a question is asked, the Chatbot needs to provide an answer and potentially do a task. There are two main ways to generate a response:
Retrieval based system: The Chatbot uses a repository of pre-defined responses from which to pick an answer. Such repository can be an FAQ or KM or other source that houses the response. Picking the best answer can range from rule-based selection to use of smart machine technology. One important point is retrieval systems are not generating new text to provide answers. (Easier)
Generative based system: The Chatbot does not use a pre-defined response but rather generates responses from scratch. Only smart machines are capable of generating responses from scratch. (Harder)
Chatbot Conversation Framework
So putting these two factors into a simple diagram we have our framework.
Closed Domain Question with Retrieval Responses: This is a low hanging fruit area for chatbots to provide business value. We process over 1.5 billion calls a year for IVR business, which is the function that is occurring in Square 1. Chatbots could mirror the IVR functionality with an improved UI for a much better customer experience. We are also using this Square 1 approach to develop content delivery for mobile self-service capabilities when our My:Time digital engagement solution includes a mobile app.
Another good example of a Chatbot using Square 1 approach is booking an airline flight. During a conversation with a human agent, I might ask about 300 possible things. Its limited, pretty well scoped and can built into a conversational engine.
Square 1 is a great first step for a chatbot because it is contained, may not require the complexity of smart machines and can deliver both business and user value.
Closed Domain Question with Generative Responses: In Square 2, questions are asked and the Chatbot has smart machine technology that generates responses. Generated responses allow the Chatbot to handle both the common questions and some unforeseen cases for which there are no predefined responses. The smart machine can handle longer conversations and appear to be more human-like. But generative response increases complexity, often by a lot.
The way we get around this problem in the contact center today is when there is an unforeseen case for which there is no predefined responses in self-service, we pass the call to an agent.
The agent is the smart machine in today’s contact center.
Open Domain Question with Generative Responses: In Square 3, we get to ask any question and expect a response. This is AGI. Artificial general intelligence (AGI) is the intelligence of a smart machine that could successfully perform any intellectual task that a human being can. A lot of money and research is being poured into this area but we are a long way off from achieving anything we can operationalize in our business.
I can have a Chatbot answer my questions using a defined FAQ, or provide my bank balance but to explain a mortgage loan product to me (a Square 3 service)... likely not. At least not in the near term.
What you can look for is more restrictive uses of this smart technology in specific domains. But again, its expense and complexity and might make it difficult to operationalize.
Open Domain Question with Retrieval Responses: This will not work because responses cannot be defined for any question. Square 4 is a no-go.
We provide this framework as a starting place to explore how you can use chatbots and smart machines in your customer service environment. Remember this is a starting place and the lines will get burred by things like hybrid smart machines.
Begin by thinking about jobs to be done and how Chatbots can help. Use this framework to help cut through the hype and focus you on areas where there you can gain near-term benefit.
Here are the resources I used to research this blog post. Enjoy.
- Deep Learning for Chatbots, Part 1 – blog post by Denny Britz
- How to Define and Use Smart Machine Terms Effectively - Gartner 2016
- Build a Chatbot - ML for Hackers #6 – video by Siraj Raval
If you want to talk more about using Chatbot to automate jobs in the contact center and see examples of work we're doing, let me know.