Check our Ultimate Chatbot Guide
How I Create Chatbots – Practical Guide

How I Create Chatbots – Practical Guide

Ola Bilińska 28.02.2019
Ola Bilińska

Chatbots are quite new tools in the marketing world, so they became the subject of many myths floating around the Internet. It’s natural that questions arise, so the more doubts we resolve in the beginning, the easier it will be to understand how chatbots work. Today, I’ll take you behind the scenes and show you how they are created.

By the way, I’d like to say hello to you. This is my debut as an author and I hope that I’ll positively surprise you. Enjoy your reading ?

Chatbots do Not Come From Nowhere

Before you even think about designing an effective chatbot, create a good brief. This document will help you delve deeper into the subject and allow you to take the further step, that is, collaboration with the client’s departments. You’ll cooperate with graphic designers, responsible for the visual design of chatbots and its consistency with the brand’s visual identity. If you need to extract data from websites, you’ll also work with developers,  and in the case of chatbots that use storytelling or that are an element of a bigger campaign, with the creative department. Let’s not also forget about social media agencies. Don’t rely entirely on chatbots, thinking that they can replace a professional agency. Chatbots are one of many social media marketing tools.

STAGE I – Research and Analysis

We have gone through the brief, all the departments know what’s going on, so can we finally move on to building the chatbot? Well, not entirely. Even though carrying out a research and analysis of the brand or target group may seem to be unnecessary (since we have the brief) or tiring (let’s move on to the creation!), I would say that it’s one of the most important and interesting stages of the whole process.

Before you start working on the chatbot, it’s worth spending some time on familiarizing yourself with the world of the brand, especially if you’re not its everyday customer. A good brief provides us with basic information concerning the target group or strategy goals the chatbot has to reach. However, keep in mind that the brief won’t tell you how the clients speak, what questions they ask and why they do so, how they use emojis, what triggers their positive reaction, and what doesn’t.

Find out what comments users post on the brand’s fanpage, ask for a list of questions they frequently ask. Perhaps there is a closed Facebook group that communicates differently, in a way that resembles private conversations, a bit like during a chat? Read all this and analyze (and in some part acquire) the way of their expression and then recreate it in the conversation scenarios. This is a very important stage because it allows to find valuable insights that are helpful in understanding the users and their needs. After all, our goal is to design communication not for our client, but for his recipients.

STAGE II – Designing and Sketching

After learning about the target group, I can see what strategic functions the chatbot should have and what main blocks it will contain. Designing blocks poses a challenge especially in the case of more advanced chatbots that have, for example, complex decision tree hierarchies. A true brainstorm is necessary. I’ve already tried out several design methods but the one that works best is putting ideas down on paper. Step by step, card after card, we can work out an optimal solution.

During the creation process, it’s easy to forget that probably most people who will interact with the chatbot have never used this tool. It’s also quite possible that some won’t even know such a thing exists. Most websites or mobile apps are more or less the same, not because their designers weren’t creative enough but because people are already used to certain menus, colors, shapes, buttons. In the case of chatbots, it’s much harder because such design patterns don’t yet exist. If I use a carousel structure, I need to inform users that it can be swiped to the right or left. If a specific command works only in one block, whenever possible, I create an interaction that will redirect me to the appropriate module after I have written the same command in a different place.

Also, I prepare a plan for the following months. A chatbot with no regular updates belongs to the past; adding further modules to a finished chatbot can turn out to be quite a challenge. Planning will guarantee you that the tool remains the same as it was in the beginning. You won’t have to turn its entire structure upside down during the following weeks or months.


STAGE III – Building the Chatbot

After the structure is planned, I reach the stage which I find the most frustrating, yet at the same time the most satisfactory. Putting my ideas on paper is sometimes like throwing cold water on myself. Especially if the modules don’t connect the way they should and if some solutions don’t turn out to be that practical in reality. I try, test, make improvements, move modules, change paths. Eventually, after I get the structure right, I write texts, choose emojis, and witness how my chatbot turns into a full member of a given brand. Collaboration with graphic designers is key at this stage because thanks to their work, the tool acquires an attractive appearance. In the case of more complex projects, you’ll need to cooperate with experienced developers who will help you add more advanced features to your chatbot.

Designing communication for chatbots made me realize that there is no such thing as a rhetorical question – someone will always answer ? I try not to use open questions that require answers in the form of quick replies because users usually don’t pay attention to small buttons under their message. Instead, I place the word “Choose” to increase the possibility that a user will click on it. In the case of quick replies, I always match answers with specific key phrases, so if a user writes a word similar to the key phrase, the chatbot will be able to answer. Optionally, you can use a carousel without photos or a message with buttons. Users are familiar with their shapes, so they’ll know that they can click on them.

I feel that contact with moderators is often omitted in chatbot structures. Even though it’s an advanced technology, chatbots won’t be able to entirely replace humans. They solve simple, common problems but they won’t give the right answer to more complex and precise questions. From my own experience as a Messenger user, nothing frustrates me more than a chatbot that obtrusively tries to convince me to try out some of its features. If we allow followers to contact a moderator, there’s a good chance that after they receive what they have come for, that is, a clear and precise answer, they will want to try out other features of your chatbot out of sheer curiosity, not out of necessity.

Why is everybody talking about chatbots?
Because brands who use them in business are getting results
Let's talk

STAGE IV – Testing and Interactions

The chatbot structure is already prepared on a platform but there are still several steps we have to make, unless the chatbot is supposed to resemble an app more than a chat. If, however, you want to allow users to send messages, you’re entering the stage over which I often lose sleep – interaction design. This is the moment where extensive research about your target group and the brand really pays off. After a user enters a message, the chatbot, by default, will reply that he doesn’t understand him. But who would want his conversation with a chatbot look like this? Chats abide by their own rules but people love to talk. It’s worth giving them this possibility. Interactions (or intelligent answers) are created on the basis of keywords. If a user writes a specific word or sentence, the chatbot will answer with a block we have already chosen. However, people talk in a variety of ways. They use different words, they make spelling mistakes or stylistic errors. I always try to remember to save key phrases with misspellings and if I make one myself, I leave it, since there’s a high possibility that the user will also make the same mistake.

Another important thing in interaction design (which seems to be quite obvious, yet is still often forgotten), is that a single word can be used in dozens of questions about all kinds of things, and even the words themselves are often ambiguous in meaning. The best example for this would be the word thanks – it makes you want to connect it to the thank you interaction group (it’s still worth being prepared for such messages since people are polite even during chatbot conversations :)). And yet answering the message “thanks to your company, I had the worst vacation of my life” with “we’re glad we could help” won’t improve customer experience ?

Testing is equally important as interaction design. The more people test the chatbot with unpredictable questions, the better. Next, we hand the chatbot over to our client and his moderator team. They will know perfectly well what users write in private messages, so their feedback will be very valuable. Keep calm and prepare yourself for regular optimizations. There is high possibility that the user will write the 201st message, even though we have predicted 200 ?

STAGE V – Implementation and Optimization

After so much preparation, work, and testing, there comes the moment we have all been waiting for – implementation, the day that reminds me of my exit examinations. This moment will verify how carefully you have made the previous steps. We launch the chatbot on the fanpage, the brand posts a message inviting followers to try the chatbot out, and messages start pouring in. This is the time when I can’t take my eyes off the screen and keep checking what decisions users make. When the user, out of two options, chooses a third one, this is the moment when all the “of course” thoughts appear and when my pile of key word notes keeps getting bigger. It’s worth encouraging moderators to control the situation, especially in the beginning, even before the first optimization. If someone gets lost in the chatbot structure, they can take over the conversation for a moment and guide the user. For us, it’s an opportunity to draw conclusions and determine what steps to take to prevent these kinds of situations from repeating in the future.  

It gets pretty hot during the first days – the users want to try out all the features of the chatbot and click on anything they can. It’s hard to keep track, especially if we’re launching the chatbot on a fanpage that has thousands of followers. You may consider introducing the chatbot on a closed Facebook group of the brand, just as we have done it in the case of the Semilac chatbot. This will help you optimize your chatbot more efficiently – less messages to analyze in the first days of implementation, and an additional benefit for more engaged followers.

What Comes Next?

Days pass by, things calm down, and fewer messages appear in the inbox. In this situation, it’s easy to forget about checking on the chatbot. This negligence cannot happen because we not only have to know how users use the chatbot in the beginning but also how they behave after we have made improvements. Even though our chatbot path has begun much earlier, this is just the beginning for its users ?