Mmm maybe later ✕
Awesome! 🤘
Shoot me an message on Linkedin or an email below.

Thank you! Your submission has been received!

Oops! Something went wrong while submitting the form

Best Practices for Training and Designing Conversational Experiences

Andrew
|
Chatbots
|
Apr 17, 2020

Orginally published on Chatbots Life

I’ve learned a lot in the past 5 years of testing and designing conversational experiences. From quality assurance of voice interfaces, training AIs, and managing over 10 chatbot projects.

Here are my 6 best practices.

Use plain, but conversational language

After setting the design and plan for a chatbot, we begin to ‘conversationalise’* the content. This means taking existing FAQs, processes, documents, miscellaneous content and adapting it for a chat medium.

The combination of plain English and conversational ready language ensures that you cover the majority of users. Our users have been primarily non-native English speakers, so catering for them is extremely important in terms of content creation and AI training.

Establish trust — honest about capabilities, roles, misunderstanding

In my early experience, we had the challenge of introducing chatbots to users who have never used them before. Failing to introduce a chatbot effectively led to conversations with users who were unaware they were chatting with a chatbot, leading to confusion and mistrust later in the interaction.

We played with chatbot descriptions and introduced the capabilities on each welcome message and greeting messages. Honesty was key as we explained in each chatbot through introduction and onboarding that you are speaking to a chatbot and how to use one effectively.

Keep in touch — language, content, events, training

Keep in touch with your users and stay on top of how your users are interacting. Take into account the language they use and any external events that are impacting them.

Monitor your interactions and be aware of real-time events to ensure your chatbot is up to date and serving your users. Start planning what new content or training could be added. Be proactive.

Train your AI on new language use, as the types of utterances evolve over time.

Think of the channel — multi-channel, apps, devices

When we grew and scaled across other messaging channels we realised how important testing is. It is vital to ensure that your content and conversation designs are consistent across channels, as minor details such as button character lengths and fonts change.

Test your designs on the different messaging apps you use, on different devices, languages and browsers. Things always look a bit different.

Keep a conversation library

Store your common conversational components to return to and re-use in other chatbot projects. I categorise and store these in large Google Sheet. This allows better scaling when you grow as you won’t need to re-write your copy and scripts each time. It will ensure your designs are consistent and keep your team on the same path.

The same applies to training data, keep libraries of common synonyms to use in entities and lists of phrases and utterances that users use frequently.

Nail your NLU processes

AI and natural language understanding technology are always evolving. New features arrive and training methods can change. Ensure your training processes are up-to-date as this allows your team of trainers to be on the same page and provide consistency across your chatbots and languages.

Subscribe to updates and release notes on the tools you use, as your existing training sets may be affected and in some cases, fail without you noticing.

The best chatbots and voice assistants are always the ones that are maintained and looked after. Keeping to best practices like the ones above will help.

Andrew
AndrewF

Recent Blog Posts

Start a conversation
Contact Me
Mmm maybe later ✕
Awesome! 🤘
Shoot me an message on Linkedin or an email below.

Thank you! Your submission has been received!

Oops! Something went wrong while submitting the form