Chatbots: What you need to know

What is a Chatbot?

A chatbot (bot for short) aims to have a conversation by mimicking interactions between humans. In other words, a chatbot is a software that tries to automate conversations. Chatbots are typically available on a chat popup on a website, on a messenger application like WhatsApp or on a mobile app.

Customer behaviour and companies’ necessity to remain competitive have supported the rising need for chatbots. After all, there are many applications for chatbots these days. Some of them are:

• Sales and Lead Generation
• Customer Service
• Personal Assistants

Many different companies are offering this kind of technology. This means that there is a variety of tools, platforms and software for chatbot development services, deployment and maintenance:

• BigTech like Microsoft (LUIS and Bot Framework), Google (DialogFlow), IBM (Watson)
• Numerous startups with open source projects like RASA or Botkit, or closed source
• Customer service software companies, like FreshDesk, that extended their offering
• Live chat product companies, like Intercom, that extended their chat functionality to include chatbots
• CRM providers, like Salesforce or Hubspot, that added chatbot functionality to their offering

Is a Chatbot an Artificial Intelligence?

Chatbots are sometimes referred to as a Conversational AI. Or as an application of Conversational AI. But, actually, how much AI is there in a chatbot?

To answer this question, we need to de-construct a chatbot. So, what happens when you interact with a chatbot?

1. You type a message in a chat window and send it
2. Your message is “classified” using a so-called “Natural Language Understanding” system, typically Machine Learning based (there is AI involved here). Some entities may be extracted
3. Next, the software decides what to answer, based on logic rules (in most cases, as simple as question -> answer)
4. The bot answers (in many cases with something like “please rephrase”)

In other words, a chatbot may be a combination of AI with classic computing. Many of them don’t even involve AI, since they are “clickbots”, just look for specific keywords in what you write, or offer you buttons to press, guiding users through a maze.

This might seem too simplistic, but it’s the reality. Most chatbots are not more complex than what is described above. Conversational AI still has a long way to go to become really conversational.

A modern chatbot, if well built, is made of different parts:

• Natural Language Understanding (NLU) – this is typically a Machine Learning based part that is trained with examples and can classify sentences, as well as identifying entities (like a product or a location) in the text.
• Context – a context is information available before a conversation starts, like for example the location of the user, and information collected during the conversation.
• Dialogue model – this is where the chatbot is “taught” what to answer in each situation, based on the analysis of the NLU, and the context.
• Answer Library is a repository of answers, that could work like a Content Management System
• Knowledge – a knowledge base that the chatbot can use to search for information, in case it has not been explicitly trained to answer a question

Additionally, there can be other components to a chatbot, like integrations to product catalogs, CRM systems, and others.

A modern chatbot, in order to be useful and offer a great customer experience, needs not only to be well trained, but also be integrated with other IT systems.

The Challenges with chatbots

You have probably interacted with chatbots before and most likely were disappointed. The reality is that people don’t like using chatbots, because most of them don’t offer a good experience. The outcome is, customers are not satisfied, and companies that invested in them are stuck with a low quality solution.

There is good technology out there to build high quality solutions, but only few companies can truly build high quality Conversational AI. Building an experience that truly adds value to the customer.

There are several reasons why customers don’t like interacting with chatbots, and consequently, chatbots don’t produce the desired results for the business.

To me, the main reason is lack of context. Have you ever interacted with a bot that:

• doesn’t know where it is
• doesn’t know you (even though you are known to the brand)
• doesn’t know where you are
• doesn’t know what device you are on
• doesn’t remember their own last sentence
• asks you the same thing over and over
• …. the list could go on.

How to make the experience more pleasant

Despite all this, even a simple, not so conversational chatbot can be useful to both customers and organizations. As long as expectations are set right, on both sides. A couple of useful basic guidelines are:

• Setting the customer’s expectations correctly – revealing that you are in the presence of a “not so smart” bot
• Giving hints to what the covered topics are
• Never, EVER, leave the customer hanging. If the chatbot does not know the answer, offer them an alternative like speaking to a human or  ask the customer to provide their contact information
• Train, train, train! Even a simple chatbot can become useful if trained properly.
• If you don’t know where to start, talk to experts of chatbot development services. They can save you a lot of time.

By the way, these principles are independent of the technology used.

The Future – Bringing things to the Next Level

If we really want to offer an extraordinary experience to consumers with a Conversational AI, then a chatbot must make use of a very rich context, including information like:

• Geo Location
• Customer Profile
• Sentiment of the conversation
• Conversation History
• Products/Services the person owns (if applicable)
• Code of Conduct and Ethics
• Brand Identity

Anytime a chatbot answers a customer, all these aspects need to be taken into account. Only then, we can talk about Conversational AI.

To know more about Conversational AI and chatbots, check our media section and watch (or listen) the Real AI. Now. podcast episode about Conversational AI, with Paulo Nunes and Nuno Galandim.