Hello friends! Thanks for being with us and we hope Stella could help you build greater user experience through Chatbot. We’ve compiled some best practices that lead you all the way to building a good chatbot. We’ve also selected some chatbot showcases for your reference.
There’s only one golden rule you should always bear in mind:
Be user-friendly. You are developing for your end users. We all want to provide a smooth and swift experience for our users. Your chatbot is more likely to help users achieve what they want if you take into account of different use cases. Here are the best practices we recommend when developing:
- Outline all use cases
- Conversation guidelines
- Character of your bot
- A mix of robot and real person
- Source of intelligence
- Stress test
Outline all use cases
Your users are very likely not quite familiar with the use of chatbot. Therefore, you should prepare for a lot of unexpected use cases. Apart from logical use cases that follow through your whole conversation flow, here are some outstanding use cases we find common:
User are usually impatient, in particular when your chatbot content is not interesting enough to keep them in the loop. If most of your users give up in midway, it implies you bot has a bad performance on user experience which would result in a low conversion. We suggest to keep a short, closed and creative conversation.
Multiple choices do have a better performance in guiding users and collecting users’ preferences. However, not all users follow “guidance” and select answers. It is very often that they type out their answers or enquiries in the decision points. We suggest to include a fail-safe in important decision points or throughout the whole conversation.
And instead of typing, some of them just love to click buttons. Sometimes when your users click too quickly and make a wrong choice, they will go back to the previous conversation and click another button they originally want to select. However, it would easily cause an interruption to the chatbot flow if users frequently make repeated choices. We suggest using quick replies to replace buttons in this case.
Highlighted above are some common scenarios that exist in general. There may be use cases which only apply to your chatbot so you should pay enough attention when drafting the use cases.
Only a smooth conversation can power up the user experience of your bot. Here’re some tips on building an effective and clear conversation:
The best conversation length of a direct flow is within 5 nodes. People lose patience if the conversation length is too long and they will easily give up in the middle of conversation. Make sure you could lead your users to conversation in less than 5 steps.
You could interact with your users with either open questions or closed questions. Our experience tells us it’s always better to keep it closed. Give options to your users so they know what to answer.
There is always a purpose behind each decision point. In case of one-off action, such as guiding users to pick an answer or asking users to select their preference, it is best to use quick replies as it disallows users to go back and click the options again. This avoids users making repeated choices.
However, for scenarios you want the users to be able to click the options again, it would be best to use buttons, e.g. newsletter subscription form and official website. Buttons are usually the best fit for conversion actions.
But of course, if you want the buttons to stay always-on-top, you should put it in the persistent menu where users can always access to it.
Users’ actions are although varying, they are usually predictable. When users are not following the designated flow, you should direct them to the correct flow they are looking for. Generally, we would suggest two solutions:
- Go back to previous node
- Direct them to live chat support
Whenever users are deviating from the original path, i.e. not selecting any options provided but typing in their enquiry directly, a fail-safe message should pop out and guide them to land on the correct node.
Sometimes users are just too busy to finish the conversation at once and they will forget they have started the chat with your bot. It is important for your bot to be proactive and take the first step to re-initiate the conversation with the users. Set a recovery message to re-engage your users. The best frequency will be 30 minutes after conversation lying idle.
Character of your bot
Your bot is like a front-line staff representing your company. You should give it a unique personality so it would give everyone a vivid and remarkable impression. Be it friendly, cool, witty, cute or serious - just make it is consistent. Here are some tips to make your bot more lively:
If you plan to build a cute and friendly character for your chatbot, this would be the rule of thumb: use a lot of emojis! Use emojis in text, in buttons, in quick replies. Use emojis to replace simple words and make your chatbot looks more lively!
It’s reasonable to use full stop to end a sentence but we are not writing an essay on our bot. Most of the text messages don’t necessarily end with a full stop. Try to mimic the way you chat with your friends and make your bot sound less formal.
A mix of robot and real person
Chatbot is a good tool for you to automatically engage with your users but you should be aware that a sense of human touch is also important in providing a good user experience. For confined conversation flow, you could leave it to your Chatbot for handling; for complicated enquiries that need real person customer support, you may wish to connect them with your live chat team.
It is best to put the live chat button always-on-top on the persistent menu so users could quickly connect to customer support at anytime.
Source of intelligence
Who doesn’t want to have an intelligent assistant like J.A.R.V.I.S.? There are surely ways to make your bot cleverer and the most common way is to improve through language learning. There are a number of language learning engines you may adopt online:
- Microsoft LUIS
- Google Dialogflow (Which is supported already on our Stella)
You have to be aware that language learning is a long process which requires a heavy input of language data and linguistic expertise on semantics and pragmatics study. With more data and more precise semantics, your bot will become cleverer and can handle more variety of responses.