How to build a chatbot in 5 easy steps

Interested in building a chatbot but don’t know where to start?
Then you are in the right place!

In this blog post I will talk about how you are going to build your own chatbot in 5 easy steps.

Build your own chatbot – The model

When building a chatbot, you need to think about the following steps:

1. Analyze your needs
2. Build your chatbot
3. Test your chatbot
4. Launch your chatbot
5. Analyze the Analytics of your chatbot

In the rest of this article, I will explain each step very carefully. But first, I will provide some background information about the model itself.

Some background information

After having interviews with six experts in the field of chatbots, I noticed a common approach in the building of chatbots. What I noticed is that all the experts use more or less the same approach to developing a chatbot, without following a certain guideline.

The six chatbot experts I interviewed were:

• UX Designer at a large Dutch bank
• Senior Designer at a large Dutch bank
• Former psychologist and entrepreneur developing chatbots for SMEs
• Co-founder and CEO of a software company for building chatbots
• Chatbot Engineer at a large Dutch Bank and former chatbot engineer at a Greek chatbot company
• Conversational UX Designer at a large Dutch bank

Analyze your needs

In the first phase, we are going to analyze your needs. We can divide this analysis in three categories:

1. What is the goal of the chatbot?
2. What chatbot elements should we use?
3. On which platform should we launch the platform?

What is the goal of the chatbot?

What do you want to achieve with the chatbot? Does it only need to answer simple questions or does it need to do more sophisticated tasks?

This is a really important question to answer, because it determines all the other things of your chatbot. The advice on this is to do the following:

When you begin with developing a chatbot, first think of one or two purposes for which you could use the chatbot. For example, you could start with giving answers to standard questions. So then the goal becomes: I want that my chatbot gives answers to standard questions.

Whenever your chatbot has “mastered” this goal, you can move on to the next goal of your chatbot.

What is important is here, is that you need to make a trade-off between two things:

1. What does your customer want?
   It is always best to have a customer-driven approach and trying to provide value for your customers
2. What is feasible given the amount of time/money for building the first chatbot?
   But at the same time, you need to be realistic and determine what is feasible

When you decide for which goal you are going to build the chatbot, it is important to get into the details of that goal. For example, when you decide that the chatbot needs to give answers to standard questions, determine all the questions the chatbot should have an answer to.

What chatbot elements should we use?

In the Analyze phase you also need to determine the chatbot elements of your chatbot. Possible questions you could ask are:

• How should the chatbot look like?
• Should users use buttons? Or can they use free text input as well?
• Shall we use a back button (so the customer can go back to the previous question)?
• Should we use a persistent menu?
• How should the chatbot present itself? As a chatbot or as a human?

Which platform should we use?

Shall the chatbot be used in one channel (for example on the website) or on multiple channels (for example on the website and on Facebook Messenger)?

The best advice to determine this question is: go where your customers are. Are your customers primarily on Facebook? Use Facebook to launch your chatbot. Are they primarily on Twitter? Use Twitter to launch your chatbot.

But it is not only limited to what your customers want. You also need to consider the technical implementations. It is really easy to launch a chatbot on Facebook Messenger, but it is extremely difficult to implement a chatbot within a mobile app. Maybe you should use a third party agency to implement the chatbot. All important questions to consider.

Build your chatbot

In the Build phase, the chatbot will be built according to the requirements stated in the Analyze phase.

Important things to consider are:

• Are we going to build a chatbot ourself?
  • And if so, do we code it or use a chatbot platform?
• Are we going to use a third-party agency?
• What is the budget we have for building the chatbot?
• What is the time we have for building the chatbot?

Test your chatbot

In the Test phase multiple types of testing could be used. Based on the content and the interface of the chatbot, a certain type of testing is used. With the interviews we noticed three types of testing:

1. Question/Answer Testing
2. Usability Testing
3. Guerilla Testing

Question/Answer Testing

This type of testing is simple. With question/answer testing someone is checking if the chatbot is giving the right answers to particular questions. If the chatbot doesn’t give the right answers, the team will return back to the Build phase and build a solution for this.

Usability testing

This type of testing is more sophisticated. With usability testing you are evaluating the chatbot with real customers. This could be as simple as sitting down with a customer and let him interact with the chatbot, but could also be as hard as using a complete Usability Lab (with cameras, eye trackers and other fancy equipment).

The goal of usability testing is to determine whether the chatbot is usable by the user.

Based on the results of the usability test, the team will decide whether it goes back to the Build phase and improve the chatbot, or goes on to the Launch phase.

Guerilla testing

Guerilla testing is not specifically focused on usability or question, but more on the total experience of the chatbot. When performing guerilla testing you go out on the streets and ask people who pass by for their thoughts.

Based on the results of the guerilla test, the team will decide whether it goes back to the Build phase and improve the chatbot, or goes on to the Launch phase.

Launch your chatbot

In this phase the website is launched on a website, a social channel (such as Facebook Messenger) or inside a mobile app, depending on the chosen platforms.

Real customers can now interact with the chatbot, which will lead to actual results.

Analyze the Analytics of your chatbot

I cannot emphasize the importance of analytics of your chatbot software. If your chatbot software does not provide analytics, don’t use it at all.

Analytics allow you to analyze how real customers are interacting with your chatbot. This allows you to improve your chatbot tremendously.

For example, with Analytics, you can identify which questions were asked by customers which did not get an answer. By analyzing this, now you can provide answers to these questions.

Important measures you need to consider in this phase:

• User input
  Is there an overview where you can see all the user input? And more specifically, the input where the chatbot could not give an answer to?
• User Retention
  Do users come back to use the chatbot?
• Sent messages
  How many messages have your chatbot sent?
• Open rate
  What percentage of messages gets opened by users?
• Click rate
  What percentages of messages get clicked on?
• Conversation rate
  How many users get to the final goal of your chatbot?
• Fall Back Rate
  How many times need the chatbot to fall back (when he cannot give an answer)? You want that this metric is as low as possible. A high fall back rate means that your chatbot needs a lot of improvement.
• Volunteer users
  How many users come to your chatbot without getting notified?

After analyzing the analytics of your chatbot, you can go further to the Analyze phase and determine which new things you can implement in your chatbot. Then the whole process begins from the start.

Conclusion

With these hands-on plan you can now start building your first chatbot!

Did I miss something? Or would you add something? Let me know by leaving a comment!

Motivated to make your first chatbot, but don’t know which chatbot software to use? Then check out this post!