Practical guide to developing chatbots

Practical guide to developing chatbots

No question is wrong, especially in the case of chatbots, which became part of marketing tools family only two years ago. The web is flooded with various myths about them. The more doubts we get rid of at the very start, the easier it will be to understand how this solution works. Now I will take you behind the scenes of developing chatbots, because this’s what I do every day. 

Chatbots are not born in vacuum

Creating a decent chatbot requires a well-filled brief. This equips us with specific know-how and allows us to start working with several departments on the client’s side. Namely, with graphic designers, so that the bot will match visually the brand’s other activities; developers, if the data are automatically downloaded from the website; and in the case of chatbots for storytelling or a larger campaign, the creative department. Let’s also not forget about the agency that handles the brand in social media. Chatbot complements their efforts, so don’t let them think that the new solution is meant to replace them. 

Stage I

Research and Analysis

The brief’s done and the departments know exactly where things stand, so can we move on to building the chatbot? Not necessarily. While for many the stage of the brand or target group independent research and analysis may seem unnecessary (after all, there’s the brief) or exhausting (can we already move to creation!), I find it one of the most important and interesting stages of the entire process. 

Before we start, it’s worth taking the time to get to know the topic and the client’s brand thoroughly, particularly if we’re not part of its everyday target group. A well-filled brief will give us basic information about the target group and the strategic objectives to be met by the chatbot. Yet, we don’t get what is very important. We won’t learn the language that customers use, their questions and how they ask them, what emoji they use, what makes users respond positively, and what does not. 

Find out what users post in the comments on the brand’s fanpage, ask for a list of FAQs. Perhaps there is a closed group, where interaction differs and resembles a private conversation, like in a chat? We read it all, analyze (and somewhat take over) the style, which we then map into conversation scenarios. This is a crucial stage, because gives me insights that will help me understand users even better and respond to their needs. After all, I’m supposed to design communication not for the client, but for its audience.

Stage II

Design and Drafting

Once I know the target audience and the strategic functions of the chatbot, I move on to determine the main blocks that will appear in it. This is challenging, but especially with more elaborate chatbots, brainstorming is essential. Having tried a few design methods, I’ve found that the best way to design a chatbot is to put your ideas down on paper, and step by step, page by page, you come up with the optimal solution

When developing (not the first) chatbot, you can easily forget that most people who will interact with it have probably not used such a tool before. In fact, they may not even know it exists. Most websites or mobile apps resemble each other more or less and it’s not because their developers are not creative. It’s just that they build solutions for people who are already used to a certain way of presenting menus, shapes or button colors. With chatbots, the challenge is even greater, because such connotations have not yet been defined. If I use a carousel in the bot structure, then I inform in the preceding message that it can be moved to the right and to the left. If a particular command only works in one block, whenever possible, I set up an interaction that redirects to the appropriate module once those words are typed elsewhere. 

Still at this point, I start planning initial activities for the coming months. After all, a chatbot that hasn’t been updated is a slowly dying product, and adding more modules to a ready-made bot can be quite a challenge. This way the tool will be consistent not only at its launch, but also in the following weeks or months, with no need to turn the whole structure upside down.

Stage III

Building a chatbot

Once the structure has been planned, then comes the stage, which I personally find the most frustrating, on the one hand, and the most rewarding, on the other. Transferring ideas from a sheet of paper to a platform at times feels like a cold shower. Especially when the modules don’t fit together as well as they should, and certain solutions aren’t that practical. So I try, test, make adjustments, move modules around, and change paths. Once the structure is finally in place, I write texts, choose emoji and see how the chatbot gradually becomes a full-fledged member of the brand. At this stage, working with graphic designers is crucial, because it’s their designs that make the tool looks great. And with more complex solutions, developers are the key, because with many features their help is indispensable.

While designing communication in chatbots, I’ve learned that there is no such thing as a rhetorical question in the chatbot context. There surely be someone to answer it. I also try not to use open-ended questions with quick replies. They often make users not pay attention to the small buttons under the message. Instead, at the very start I include the word “select,” so that the user is more likely to click on the button. For quick replies, I always set keywords to each reply. This way, once the user types in their message a word related to one of them, the chatbot will redirect the user to the relevant next step. Alternatively, a carousel with no photos or button messages works well, since its very design known from similar solutions guides users to click on something.

My impression is that a simple interaction with a moderator is quite often overlooked in chatbot structures. It’s still a long way, before chatbots can replace a one-on-one conversation with a human. They solve simple, common issues, but won’t provide the right answer to a more complex or precise question. If I were a user writing on brand Messenger with a specific question and expecting a quick answer, nothing would discourage me more than a chatbot, which tried to forcefully convince me to use more of its features. If we allow fans to simply interact with a moderator, there’s a good chance that once they get what they came for, i.e. a specific answer, they’ll stay and try other features more eagerly and out of sheer curiosity rather than necessity.

Stage IV

Testing and Interactions

The chatbot structure is already implemented on the platform, yet there are still a few steps ahead, unless our chatbot is meant to resemble more of an app than a chat. If we want to allow users to send messages, however, there starts the stage that I sometimes dream about at night, namely designing interactions. This is the moment when it pays off to know your target audience and the brand’s universe very well. By default, once a user types in a message, the chatbot will reply that it doesn’t understand. But who wants a chat conversation to be like that, even with a bot? The chat has its own rules, and people love to talk. So it makes sense to give them the opportunity, as much as it’s currently possible. We set up interactions (or otherwise intelligent responses) based on keywords. Whenever a user types in a particular word or phrase, the chatbot will respond to the message with a block of our choice. But people talk in different ways, after all. They use different words, make grammar or stylistic mistakes or typos. I try to remember to write down keywords with some mistakes, and if I make a typo myself, I leave it, because it’s possible that another user writes the word the same way.

When designing interactions, there’s one more important thing that seems obvious, but is so easily forgotten: one word can be used in dozens of questions about all sorts of things, and words at times can be ambiguous. The best example is the word “thanks.” You’d like to include it into interactions together with “thank you” (and you should be prepared for such messages, as people are polite even when talking to a chatbot :)). And yet, responding to “thanks to your company I had the worst vacation of my life” with “We’re glad we could help” is unlikely to have a positive impact on customer experience.

While designing interactions is the first step, the second one is just as important, because it’s about testing. At this stage, the more people trying to ask the chatbot a question with their own words, the better. They will certainly use expressions that we didn’t think of. The next stage is to hand the chatbot over to the client and their moderators team. They know perfectly well what users type in private messages and their feedback is invaluable. Still, I advise you to approach the interactions with calm and a mind set for regular optimizations. Chances are that if we predict 200 interactions at the start, the first one someone types in will still be 201 variation.

Stage V

Implementation and Streamlining

Once the preparations, work and tests are done, the moment we’ve been waiting for arrives: the day of implementation, which feels a bit like the graduation day. It’s a kind of exam, which verifies how diligently we have taken the previous steps. We link the chatbot to the fanpage, the brand shares a post on its wall inviting to use the new solution, and dozens of messages flow into the inbox. It’s that moment when I cannot stop staring at the screen and checking out another user’s choices. This is when I use to exclaim another “well, sure!,” when a user selects a third option out of two, and the keyword notes grow. Even before the first optimization, you should encourage moderators to check things regularly: if someone’s not doing well in the chatbot structure, let them take over the conversation for a while and guide the user further. This gives us a chance to draw conclusions and decide what we can do, so that such assistance is not needed again.

It can be hectic in the first days. Users want to check out all the chatbot features and they click on everything they can. It’s difficult to keep up, especially if you’re implementing a chatbot on a fanpage with hundreds of thousands of users. A solution to consider is to present the chatbot to a closed brand group,for example, as we did recently with the Semilac chatbot. Not only does this allow the chatbot to be optimized more efficiently, with fewer messages to analyze on the first, busiest days. But it also provides a bonus for the most engaged fans.

What’s next?

Days go by, the excitement subsides, and there are fewer messages arriving in the inbox. At that point it’s easy to neglect to regularly check the chatbot’s performance. And that’s absolutely necessary, because we need to know which features fans use beyond the first moments, and to monitor whether – once adjustments are made – new users stick to the path we’ve designed. Even though the chatbot’s journey started much earlier for us, for users it is just the beginning.

Contact with us

You are one step away from separating yourself from the communication community in your organization

You can contact us and we will prepare a dedicated offer for you.