These Bots Were Made For Talking
January 14, 2020
When evaluating solutions that will allow you to add chatbots as a support channel, you will most likely encounter the following two challenges:
- Vague descriptions of “Conversational AI” and “Intents and Expressions” that fail to adequately explain a solution’s capabilities
- A lack of specificity regarding the time it will take a vendor to build a bot for your business or enable you to build one yourself
It’s extremely difficult to effectively evaluate a solution without knowing the extent of its AI capabilities and the time needed for implementation; however, since these details will vary from chatbot to chatbot, it’s important to understand how different options will support different business objectives and use cases.
The evolution of chatbots
To begin, an examination of chatbot evolution helps to differentiate between three distinct chatbot types. The most simple type of chatbot is called a “Flowbot,” while “Replybots” are slightly more complex and “Conversational Bots” are the most complex.
The Flowbot is the most accessible chatbot and one that works in a similar way to an IVR menu on the phone. After dialing, the caller listens to a voice instructing, “For support, press 1. For accounting issues, press 2.” Instead of working over the phone, Flowbots present a chat window that guides consumers through a preset “Flow.” It works by requesting input in response to a series of questions designed to identify key information and facilitate issue resolution. The Flowbot may employ tools such as “Quick Replies” or “Buttons” that are part of the company’s native support for its main social networks such as Facebook Messenger and Twitter DM. Flowbots can be set up by a single human being who does not need the ability to code. This characteristic reduces the complexity associated with implementation, and the straightforward question/answer format provides a user-friendly experience that effortlessly guides consumers to the solution.
Replybots are a little more advanced than Flowbots. Unlike Flowbots, Replybots don’t present answer choices in a multiple-choice format. Instead, they allow consumers to type out any question, which the Replybot will then answer. You might think: “So, this is where Artificial Intelligence kicks in, right?” Well, not necessarily. The chatbots that do use AI in this case are a cheaper alternative to the ones that don’t.
When opting to go without AI, every question a consumer might ask requires a manual input. This requirement means that support teams will need to build extensive libraries of every probable question and answer. Replybots use an “If… Then…” approach to evaluate a consumer’s response and select the appropriate answer. For example, if the consumer says: “Hi, can you help?,” then the Replybot will answer: “Hi, what can I do for you?” Even though pre-built libraries are sometimes available, the process of building a Replybot requires significant preparation in order to fully customize questions and answers to a specific brand.
Using AI, building the aforementioned libraries can be highly automated. Deep learning allows a Replybot to automatically understand the intent of messages and can immediately pick a suitable reply. It also recognizes “Hi, Good morning” and “Good afternoon” as greetings and is able to reply accordingly. The use of automation dramatically reduces building time, but it will still be necessary to train the AI to work with the specific brands and business processes.
In 1950, Alan Turing, one of the most influential computer scientists, developed a test for Artificial Intelligence that evaluates a machine’s ability to exhibit behavior that is indistinguishable from human behavior. As the most complex of the three bot types, a Conversational Bot should pass the Turing test. It should be able to answer any question a consumer asks in a way that results in the consumer being unable to distinguish the bot from a human being.
While humans possess the ability to build Conversational Bots, they can only do so if they focus on a very specific domain. As a result, these chatbots are very expensive and offer functionality that’s far beyond what is commercially necessary.
So, which chatbot is right for me?
Each chatbot type has its own pros, cons and price tag. Flowbots are readily available and excellent at managing large amounts of incoming volume. Best used to answer FAQs or point to a help article, Flowbots are easy to set up. They are cheap and often included with contact center software. For instance, Clarabridge offers a flowbot builder that easily integrates with Twitter and Facebook free-of-charge.
Meanwhile, Replybots use AI for more advanced functionality and are available for commercial use; yet, the ease of implementation varies greatly depending on the vendor. Consumers consider this type of chatbot to be very innovative and even like to test the chatbot’s limits now and then. These bots are more difficult to set up than Flowbots as building a Replybot requires cross-functional cooperation among customer service, marketing and IT teams. Despite a more complex implementation, Replybots add tremendous value when executed well.
Given the high costs and unnecessary level of functionality, the development of Conversational Bots should not be a priority. Clarabridge does not build this kind of chatbot as it is a very distinct technology; however, Clarabridge does help chatbot builders improve their AI language models. In a commercial context, the use of Flowbots or Replybots will be much more effective.
Whichever bot you set up, it needs to work well with the agents who are operating on the same channels. An agent doesn’t need to see every bot conversation but should be able to pick up where the bot left off the moment a consumer asks for human assistance.
Dimitri Callens (@calldimi) is the director of product management at Clarabridge Engage.