11 Step Process for a Great Chatbot Design

Designing a bot is like brewing beer; it always takes the same ingredients to make it, but slight changes to the process can transform the whole outcome.

A chatbot needs a good platform, script, name and image for it to work. But it needs a purpose, personality and functionality for it to be great.

Chatbots with performance issues create stigma: According to our research , 60% of participants say that chatbots fail to resolve their issues and users rather speak to a human than a bot to answer service questions. Fair enough, but maybe this perception can be improved.

We’ll show how you can design a chatbot that meets the expectations of your company and customers - including common pitfalls and pro tips by leading experts.

11 tips from chatbot design experts for the creation process

1
Determine your bot’s purpose

Chatbot design starts with a simple question: Why do you need a bot? If you cannot answer this question with conviction, then you may want to rethink if you really need one.

List of anti-reasons for chatbots
Source: chatbotsmagazine.com

UI Designer Saumya Srivastava stresses:

The users are using your chatbot for one reason, and one reason only: to seek an answer to one of their problems.

So what is the purpose? This is what you’ll design your chatbot around.

2
Decide between a rule-based and NLP platform

After the why, comes the how. But let’s get some labels straight, first. Most chatbot platforms call their bot “artificial intelligent (AI),” no matter if it actually uses smart self-learning algorithms or sticks to simple IF-THEN metrics. So the trigger words you are looking for when choosing a building platform are “rule-based,” or “NLP.” These specify how flexible and smart your bot operates within a conversation.

Rule-based bots chat according to defined decision trees. Like a flowchart, conversations are mapped out to anticipate what a customer might ask and how the chatbot should respond.


IF user's input contains 'shop' or 'buy'
THEN send message with product list

Bots with Natural Language Processing (NLP) are able to understand the context even when questions are more complex. Thanks to their ability to learn from their mistakes, they improve with every inquiry.

Rule-based bots are at this point the quicker, cheaper and more effective solution – and users are more likely to engage with them because they are more transparent about their function range .

Screenshot of rule-based chatbot
A rule-based bot can give answer suggestions and react to specific trigger words. Source: NNP Group

UI designer Saumya Srivastava recommends rule-based bots because they create a more fluid chat experience: “I find it’s always a good idea to present your users with options while they’re chatting with the bot. The reason being, it’s a huge time-saver, and also allows the conversation to take place easily, and seamlessly.“

3
Know the limitations of your platform

Once you have decided on the type of platform you want, you still need to find the right one.

Some rule-based platforms solely work on a multiple choice basis without the option to create unique answers. While it’s possible to guide the conversation in specific directions, you can’t write suitable responses to questions that may be asked.

Some (especially younger) platforms like ThinkAutomation expect you to input questions and answers in a coded format, which requires a certain affection for coding to enjoy using them.

So, it might be the better option to choose a thoroughly-tested platform as it comes with exhaustive documentation on tools and tricks. Plus, you can join threads and online communities where these platforms are discussed, making troubleshooting a lot quicker and easier. If you’re a newbie in chatbot design, you might find it useful.

Screenshot of two different communication flows
Does the bot platform allow you to build the desired communication flows? Source

To help you make an informed decision, I’ll link you to two neat chatbot lists: one comparing NLP and rule-based bots and one specifically for the best NLP bots .

4
Define personality and tone

Your chatbot is a representative of your brand and often the first one to say hello to your customers. It’s important to design its language in line with your corporate identity. You might even use the birth of your digital employee as a chance to improve your brand image by giving it a likable persona.

Our recent study of consumer chatbot perceptions made it more than clear that people prefer bots with human traits. We asked:

What would make your experience chatting with a chatbot more positive?

A respondent answered:

If they reacted in a bit more human fashion. If their responses were more true to life or they were more responsive to language cues.

But how do you create a human-like personality for a chatbot? We can simply borrow some screenwriting principles. You first define the role your bot plays (based on the purpose), then you create a little background story: What’s your bot’s motivation, what’s its specialty, what are its striking character traits? And not to forget, what name fits this persona?

If you are in this particular chatbot design phase, I recommend reading our post on finding a catchy bot name .

For now, let’s take a look at a great example of bot design: Siri. Apple has done an excellent job of enlivening their voice assistant. Siri is truly a sweet and funny character.

Screenshot of Siri chat bot assistant

Wittiness works great for trendy Apple, but it really isn’t for every brand, as an article published by The Manifest points out: “Don’t feel pressured to program humor or street-smarts for audiences that don’t crave those traits. A simple, helpful, and polite robot is the best choice for many brands.”

5
Text like a human

Defining a character is one thing, bringing it to life using engaging dialogue is a different story.

Since we learned that users want the interaction to feel human, it’s important to invoke positive emotions during the conversation.

Vaibhav Verma , UI designer for chatbots, suggests the following: “Taking up the user’s name frequently, greeting him, congratulating him etc. are some practices, which make conversations more personalized.”

Another easy way to invoke human emotions is through the element of surprise. Design a chatbot that is surprisingly smart, witty, empathetic or all of the above.

6
Design the flow

While chatting, your bot should use prompts to keep visitors engaged to quickly and efficiently resolve their request. The biggest challenge is identifying all the possible conversation scenarios, and defining how it’ll handle off-topic questions and unclear commands.

“It is actually a good idea to spend a lot of time on this step to get close to defining the experience for your users,” Saumya Srivastava recommends.

Define the key terms that your bot is able to “understand,” how it responds and which options you give your visitor to move forward. Chatbot expert Vaibhav Verma explains “If the user does not know what to say, the chatbot must come up with suggested tasks that he can perform for the user.”

7
Keep it flowing

Chatbots, like real service agents, sometimes need to wait for downloads or feedback from other parties. Instead of radio silence, fill the waiting gap with fun facts or news and updates about your service or products.

This’ll make the wait for search results or a taxi confirmation a lot more bearable.

8
Integrate visuals

Besides the text, visuals are the second most important and useful element of your chatbot design. According to research conducted by 3M , the company behind those famous yellow Post-its, visuals are processed 60,000 times faster than text. This means using images to illustrate your chatbot’s messages are likely to hook your user’s interest.

Your chatbot can show your customer a map of the closest stores based on their request or send a picture of the driver who is going to take them there. These are simple ways to sprinkle visuals with purpose into your design.

In addition to being helpful, visuals are a great way to put a smile on your visitor’s face. Use them in a clever way to add some humor to conversations.

Screenshot of visual reaction of chatbot

But if your bot design is a voice-based assistant like Alexa or Google Assistant, pictures and GIFs are obviously not an option. Try to inject some quips into your bot’s speech to make the exchange more casual.

9
Educate users on bot cheats

A bot conversation can be draining if the user talks in bits and bites. Give them pro tips on how to quickly solve their request.

Good and bad examples of chatbot usage

The simplest way to educate your users is to let them know at the start which command elements are required for a quick search, or to simply ask “What can you do?”

If the user goes silent for a few seconds during the conversation, the bot can remind him of cheat commands or possible requests.

Some real life examples:

Both companies used a different approach, but were able to convey the scope of their bot’s ability in as few words as possible. This helps avoid overwhelming your customer with options.

10
Find a balance between proactive and reactive

Chatbots can be smart promoters, if you don’t push it.

They can invite visitors to start a conversation about deals and promotions, like after they’ve lingered on a page for 30 seconds: “Hey, just letting you know - if you order within 24 hours, shipping is free. Want a promo code?”

Or let them know about special deals on products they’ve viewed:

“Did you know that we have Nike shoes on sale? Click here to browse :)”

If you decide to use a proactive approach, it’s best to have the chat window pop up in an unobtrusive spot. According to the Gutenberg Diagram , the bottom right corner works best. This will help keep visitors from closing the window before the chatbot can do its thing.

iIllustration of Gutenberg diagram
The human brain scans a web page from top left to bottom right. Source

This is also a good opportunity to offer products and services after your customer has accepted your chatbot’s help.

Seems salesy and stumpy?

You are right.

That’s why it’s all about the balance between responding to the customer’s needs and offering a comprehensive service experience.

For example, if the bot helps me find a new computer monitor, but then starts recommending expensive gaming keyboards and graphics cards, I would be annoyed. These products are potentially relevant, but it’s purely making assumptions about what I need. On the other hand, if a chatbot recommended a warranty plan or HDMI cables, I might be interested. This is useful to me in the moment, and within my price range.

Finding this balance and coming up with targeted campaigns and promotions takes some good thought. That’s why Living Actor , a company specialized in chatbots, suggests creating a new position in your company:

“To be effective, your conversational robot has to be configured by a Chatbot Manager, based on relevant scenarios that are adapted to the actions performed by the users.”

For some more tips, check out our post on how to use proactive chat effectively.

11
Make it easy to switch from bot to human support

Your bot cannot help with every possible inquiry, especially if it comes to complaints or exceptional cases. That’s when your customer needs to talk to a human.

When you design your chatbot, allow for an easy shift from bot to agent. Here’s what the conversation flow could look like:

The most reasonable option is to integrate the bot platform into your live chat software, which makes conversation handovers simple.

Prototyping

Before your bot comes to life, you might need to present the design to an investor, employer or contractor. A prototype is useful for finding clarity and direction during the chatbot design process. Luckily, you have many options for creating professional bot prototypes to showcase your design. Here is a list of bot prototyping tools .

Testing & Data Analysis

Let me stress once again that chatbots are like perfectly brewed beer. The creation takes time to perfect. Your bot design is not going to be great overnight, which is why you need to test it within a closed group (e.g. employees, existing customers, testers) and analyze test data to improve the performance.

This is a helpful guideline for chatbot testing, for both developers and testers.

You will not get around involving real humans in the test process, but you can partly automate testing by using tools designed for this purpose. Botium , one of the first companies offering automated chatbot testing, claims that you can run the same 200,000 tests that a team of 30 people run within six weeks in two hours.

These are some useful chatbot testing tools:

Creativity and tech go hand-in-hand for a great chatbot design

Chatbot design requires a shift in paradigm for a UI designer because 90% of the focus is on human-like, engaging and educational conversation flows. It can take quite some brain stretching to create clear, cohesive and useful scenarios.

Even though bot platforms simplify the chatbot designing process, it is wise to spend a good amount of time with flow creation and later, testing and adjusting.

The best way to know if your chatbot design was successful is to not only check if it works technically, but to make sure that it meets the functional design requirements. Ask yourself honestly: “Does the chatbot actually do what I want it to do?”

Learn more best practices and see examples for chatbots in customer service here .