Towards A Chatbot Taxonomy

Chatbots are like teenage sex: While everyone is talking about it, nobody is doing it well. In other words, chatbots carry a great promise that they’re yet to fulfill.

There’s lots of confusion surrounding this exciting technology. I’ve worked out seven questions to help us talk about chatbots.

The hype makes sense: We live in an instant economy where customers enjoy on demand movie streaming, rapid commerce, and seamless communication. They’re rightfully demanding companies to react quickly and chatbots are the only solution for instant customer support at scale.

How do you talk about chatbots?

I have yet to find a clear classification of chatbots, and what they actually do.

When talking to prospects, they often ask “how’s this different from IBM Watson?” — which underestimates both its and our own capabilities. Chatbots are so hard to compare because they are much more complex than they appear at first sight:

  • Often bots rely on APIs to connect with other services
  • These technologies are largely hidden behind the marketing efforts of chatbot manufacturers — while you see their claims, you can’t see the tech stack.
  • Meanwhile, the field is incredibly dynamic, with new startups bringing new technologies to the market each month.

My market research on classifications and taxonomies didn’t yield any satisfying results. All articles I read are either overly complex, left out important factors, were hard to remember, just narratives, or not logically sound. Nevertheless, some examples are here: link 1, link 2, link 3, link 4, link 5.

Classification doesn’t work (now).

We need a common wording that helps us tell them apart. But since the field is so dynamic, it would be short-sighted to define categories now and sort the different technologies into predetermined boxes. I believe it makes more sense to classify chatbots through seven specific questions to understand more. Later we can group similar chatbots into categories.

I’ve come up with these questions following my own experience of building Solvemate, a Berlin-based company that uses chatbots to automate customer support. Our own chatbots (which we call virtual agents), are being used by thousands of endusers every day, solving the majority of support requests in less than 12 seconds.

Since sharing is caring, find below the first iteration of the chatbot taxonomy: Seven questions that let you understand different chatbot manufacturers and the scope of their technology.

Seven Questions to Ask

  1. What is the medium where the bot will be used?
  • Text-based (Chat on a website, FB messenger, WhatsApp, Slack, Telegram, WeChat…)
  • Voice (Alexa, Google Home, Siri etc.)

2. What is the general purpose of the chatbot?

  • Sales (Goal: Sell stuff. eCommerce bots, airline bots etc.)
  • Branding (Goal: Engage customers with the brand itself)
  • Support (Goal: Help customers solve their support requests)
  • Other/Fun (Goal: Chatting with a cat, checking the weather etc.)

3. How much of the chat flow is rule-based v. dynamically generated?

  • Rule-based (fixed flow with if/then or other hard coded rules)
  • Dynamic (responses are crafted based on training data and mathematical calculations)

4. Is the flow automatically changed based on usage?

  • The flow doesn’t change (humans need to update if the flow shall be altered)
  • The flow changes automatically (based on machine learning. This is the crucial part, so please ask follow-up questions: What exactly is changing? Why? What are the algorithmic boundaries? Examples? Any good provider should be able to explain that well.)

Key learning: Most of the providers put the “AI” badge on their products. Question 3 and 4 are crucial to figure out if a provider has a “fixed, rule based chatbot” or a “dynamic system with machine learning”.

5. How does the input work?

  • Free text only
  • Multiple choice only
  • Mostly free text
  • Mostly multiple choice

6. What is the Business Model of the chatbot provider?

  • Custom developed (these companies sell projects, not software and make money based on the the amount of work required; a digital agency for chatbots)
  • Software-as-a-Service (they sell software licenses, have developed a “standardized” piece of software that just needs to be “configured” by their customers and usually is paid by the software usage)

Key learning: Beware of custom development — you don’t want to continuously spend resources on maintaining a custom built software.

7. Is the solution applicable to all industries or only specific verticals?

  • Specialized one industry only
  • Applicable to all industries


I know it’s hard to remember so many question. It helps to remember them in terms to seven dimensions:

  1. Medium
  2. Purpose
  3. Flow Generation
  4. Machine Learning
  5. Input Type
  6. Business Model
  7. Industrial Focus

Just by remembering these dimensions, you are already in a good shape to ask questions.

We’ve made a graphic that you can print out or share with your colleagues as visuals always make it easier to remember:

Applying the taxonomy

To show how the questions help you assess a chatbot manufacturer, I have written down the answers for my own business, Solvemate. As you can see, we end up with a simply list of classifications:

1. Solvemate is a text based,

2. support-oriented chatbot provider.

3. Our flow is dynamically generated and

4. optimized by both machine learning and manual improvements

5. We allow only multiple-choice input.

6. Our business is Software-as-a-Service and

7. applicable to all industries.

Let’s apply it to Chatfuel to have another example:

  1. Chatfuel is a text based,
  2. Generalistic chatbot provider (for Facebook Messenger).
  3. Their flow is rule-based and
  4. The flow doesn’t change automatically when users use it.
  5. Mostly multiple choice input is used.
  6. Their business model is Software-as-a-Service and
  7. Is applicable to all industries

And finally, let’s also apply it to IBM’s Watson Conversation Services, to see the difference:

  1. IBM’s Conversation Services is a text based,
  2. generalistic natural language API
  3. The flow is dynamically created (by deep learning algorithms)
  4. but doesn’t change automatically inside your bot
  5. Mostly natural language input is used
  6. Their business model is Software-as-a-Service (charging by API calls )
  7. Their service is applicable to all industries.

We know, it is very tough to answer those questions from the outside just by looking at a webpage, but when you are in a call with a provider, try to ask the questions and if you don’t understand it, ask how or why until you’ve really understood it.

Constant Flux

As I wrote on top, the field of chatbots is still coming into its own, with new providers and solutions popping up almost every day. While this taxonomy should help you classify chatbots and their providers, I’m sure we’ll publish an update with refined questions in half a year.

Have your own feedback? I would love to hear about it in the comments. Every thought helps us come up with a better system to talk about chatbots, cut through the hype, and end the confusion about this promising technology.