Chatbots are new for many.
In the hope of adding weight to the chatbot definition, I thought I would put together a cheat sheet on the types of chat bots available.
I thought I will stay outside of any complicated technical terms, but you should be prepared for a little more advanced content than I usually share on this blog.
Brace yourself, we have quite a bit of content to cover.
Disclaimer: chatbots, chat bots - for this article we will use both interchangeably. We tend to stick to 'chatbot', though.
Who defines these types of chatbots?
Truth be told, no one.
There is no content out there that is the 'definitive guide to chatbot types' or anything like that.
This puts me in the fun position of being allowed to give my own definition. I get to define what chat bot types there are and how I differentiate them.
We are not new to this. Being at the forefront of the chatbot industry, we got to define a lot of trends, nomenclature, and other definitions.
So, who defines these types of bots? Right now, we do.
The different types of chat bots
For the purpose of clarity, I will split this into four sections, each of which will contain at least two chatbot types. I have grouped them by features and uniqueness.
This section is all about the backend of the chatbots. What is powering them?
Flow chatbots are tree-based. The user is driven down a specific path, a path previously defined by the chatbot's developer.
Have you ever read one of those roleplaying books (I used to love these... nerd alert!) where you are the hero, make decisions, and jump from page to page based on what you decide to do? Although you have the liberty to pick your own path, you remain within the confined rules of the book (i.e. you can't make up new pages).
Flow chatbots are the same. As the hero of this chatbot, the user gets to make decisions, but can never stray away from the path(s) the chatbot developer has made available.
Usually, flow chatbots involve lots of buttons and keywords as opposed to free typing. These buttons drive the user down the predefined paths.
Artificially intelligent chatbots
As opposed to flow chatbots, artificially intelligent chatbots rely on artificial intelligence (duh!) to deliver the user experience.
As we will see further down this article, this can be divided into chatbot types even more. For now, though, let's keep it simple.
Artificially intelligent chatbots allow the user to engage in a much freer way, with text, back and forth answers; a real discussion. This is not possible with flow-based chatbots.
Hybrids are the most common type of chatbots you will find. Like the name suggests, hybrids have a bit of both worlds.
They drive the user down a specific pre-defined path but also allow free text and interactions. Or, they allow free text and interactions but make use of pre-defined triage whenever necessary.
It is rare to encounter a chatbot that is solely based on artificial intelligence at the moment. Most functional (see below) chatbots use some sort of flow interaction at some point, to improve UX and goal completion.
'What? I just read your section about artificially intelligent chatbots. What's this?'
I know I know. I am making this section to drill down a little further in the types of artificially intelligent chatbots you can find. Bear with me.
One-way AI chatbots use artificial intelligence on only one side of the user to chatbot interaction. The goal of the AI for these ones? To understand what the user is saying - that's it.
As promised, I am not going to go too technical here (mostly because I am not the right person to do so). Instead, I will do what I do best: present you with a horrible drawing.
As you can see, the magic here happens on the right-hand side. The AI is here to make sense of whatever the user said. Once it has associated an intent to that input, all it does it fetch the associated answer from a pool of answers prepared by the chatbot developer.
As opposed to one-way AI, these chatbots use artificial intelligence to feed the information back to the user.
Instead of matching an intent and cherry picking an answer from a pool of pre-made answers, two-way AI chatbots have the capability to build answers on the fly - so to speak.
Most two-way AI chatbots work by connecting to a database of available information they can learn from. This all gets complicated, so once again let's keep it simple.
Two-way AI chatbots have the capability to compute user input, understand the intent, then construct the most accurate answer before delivering it to the user.
As you can imagine, this is high-level super-smart stuff. These chatbots can learn over time (trial and error, better database, better information to work with, etc.). Sometimes, they fire back (hi Tay).
A great way to divide chatbot types is by purpose. What do they do?
I believe, as the industry develops, we will differentiate chatbot types by purpose more and more. Once all the technology aligns and reaches status quo, this is what will make more sense.
So, at a high level, what is the purpose of your chatbot?
Functional chatbots serve a real purpose.
More than that, they optimise a process. Functional chatbots 'one-up' the tech currently used to perform the same task. They take away pain.
Think of KLM's chatbot. It takes away the pain of browsing a horrible website full of obscure filter options. It transforms this experience into a conversation.
You could argue you would prefer to use the website instead of the chatbot. The quality of the chatbot here is not evaluated, rather the functional aspect of it. If you'd like to learn more about reviewing chatbots, we published the document we use internally to do just that.
Grab the Chatbot Review Checklist.
As opposed to functional chatbots, fun chatbots are there for... fun (who knew). They do not serve a purpose greater than the amusement they give their users.
Fun chatbots could be divided into even further sub-categories (social chatbots, brand persona chatbots, etc.). The point to remember here is that though a chatbot can have a fun aspect and still be functional, a fun chatbot has no function.
As an example, think of the (pretty horrendous) Catbot. What did it do? Reply with cat sounds to anything you way say to it.
This is a bad example because, let's be honest, this is no fun at all - but you get the gist.
The final section of this article is the audience. It is also something that has been talked about before by some of our peers, so I am sure this will resonate with you.
A generalist chatbot is a chatbot that tries to do it all. The best way to describe it is through an example, so let's take Siri.
Siri 'does it all'. You can ask it questions about general culture ('who is the president of the United States?'), questions about your calendar ('what meetings do I have today?') or even to do stuff for you ('call mum').
Generalist chatbots are, as you can imagine, extremely complicated. It is no surprise, then, that they are developed by the most innovative companies such as Google, Apple, or Amazon.
To be able to 'do it all' and do it all well, a generalist chatbot has to have access to humongous amounts of data.
A specialist chatbot is David's answer to Goliath.
It may seem impossible for any business (David) to compete with the massively advanced companies (Goliath) in terms of AI, chatbot, and pretty much everything else. Yet, there is one thing these businesses can do: focus.
Specialist chatbots focus on one constrained thing and do that one thing extremely well.
Sure, Siri can give you information about different types of cars; but can you use it to go through a car renting offer, proposal, and delivery?
Specialist chatbots have the unique advantage of knowing exactly what they are here for. Having a sole goal to achieve is a welcomed restriction in the AI world. It reduces the error margin, makes it easier to understand user intent, and offers convenient fallbacks.
All the chatbots we have developed have been specialists. We focus on specialist, one or two-way AI, functional chatbots. Now you know what this means!