Conversational User Interfaces (CUI) are often a tough nut to crack for UX designers. On the one hand, they provide a clear structure that UXers have to follow, but on the other, the petite window and message bubbles may impede the ideation process. Despite its size and form, CUI encounters the challenges of standard interfaces. When approaching a chatbot project, you will step on information architecture, UX writing and interaction design. We took the liberty of compiling a shortlist of tips, which make the experience more engaging and handier for the end-user.
1. Follow the KISS principle with the content.
In essence — keep it short & simple. When designing the copy for a conversational interface, consider certain informality that comes with this touchpoint. The chatbot should quickly get down to the nitty-gritty, similarly to a friend chatting with you via a messaging app. While trimming down wordy messages, ensure that the substance gets conveyed. Your chatbot aims to be your customers’ assistant — its responses should be serviceable. Remember to take advantage of the content available on the channel where your assistant sits. You can link parts of a message to your website or external sources so your user can easily access relevant information.
The bot reply should indicate an understanding of a customer’s intent. Before giving detailed instructions, confirm that they apply to a searched topic, e.g.:
- “To renew your annual membership, follow these simple steps: …”.
- “It’s great to hear that you would like to join our team. Please send …”
- “I’m sorry that your card was lost. To order a new one…”
Once your chatbot is up and running in the production environment, it is worth analysing user reactions to a response. If you experience follow-up questions or misunderstanding, it is a clear sign to rephrase the bot’s copy.
2. Make it personalised.
Chatbots aim to curb the distance between you and the customer and make your brand more human. Therefore, it is crucial to inform your audience they communicate with artificial intelligence, yet make sure that the bot’s language expresses empathy. That is when personalisation enters the game. You are off to a great start if you can authenticate the user on your channel. Personalisation might be as straightforward as using a customer’s name in the greeting.
You may upgrade the experience by referring to the user's purchase history or previous dialogue.
- “Hi, Mary! How are you enjoying your new headphones?” can be a method to welcome a recurring user.
- “Hello Andy, these shirts will go great with the jacket you’ve just bought!”
- “Last time we spoke, you felt overwhelmed. Is it better now?”
CRM data might enable tailor-made product or service suggestions. Your company’s virtual agent should resemble the assistant your customer meets at their favourite brick-and-mortar store. It is the one who knows their needs and preferences by heart.
3. Ask follow-up questions and clarify ambiguity.
Neither NLP (Natural Language Processing) engine nor sterling bot training guarantees 100% intent coverage. There is always this one form you did not consider in your user examples or a query completely outside the knowledge base.
The core of chatbot UX is how well it replies to user problems. Make sure that your assistant clarifies broad questions. If there is an entity missing, prompt for it:
- “We have a wide insurance offer. Are you interested in travel, health, or property insurance?”.
- “Which office location are you looking for? Please choose the city.”
Does your chatbot engine allow a disambiguation feature? Fantastic, you may also consider it! Let us briefly explain the way it works in IBM Watson. If more than one bot response matches the user’s intent, the bot asks for a user’s help. Criteria — the “top” and “runner-up” answers have similar confidence, and the top intent confidence score surpasses 20%. If supervised, disambiguation may help your users get the most out of your chatbot’s vast knowledge.
4. Add rich media.
We would not take up conversational AI, having not believed that messages are a natural form of communication. Words assemble chatbots, that is for sure. But with more complex answers and processes, rich media offer invaluable help. If some response in the dialogue needs to enumerate different options, consider carousels. Remember that you can reuse splendid graphic materials compiled by the design teams for company-wide purposes, such as videos, images, or gifs. If you are keen on improving the overall user experience, let the voice of your end customers be heard. An engaging survey of thumbs, fields or numeric scales will help collect priceless feedback. Anytime the dialogue seems too overwhelming, brainstorm with the UX team on utilising the power of symbols and imagery.
5. Consider accessibility.
Accessibility is sometimes perceived as a “nice-to-have” within digital products, though this trend luckily is diminishing. Want to improve your UX design? Make it equitable! Ensure that historically underrepresented groups of users can enjoy your product. Not sure where to start? Web Content Accessibility Guidelines (WCAG) offer a helping hand by clearly explaining the requirements. Guidelines enclose but do not limit to contrast and font settings. To include people with vision impairments in your designs, a virtual assistant must be compatible with screen readers. Check the details of accessibility features in our University of Silesia implementation.
6. Provide human handoff.
Let us face the truth — you have happened to get annoyed at a virtual assistant at least once within your lifetime. Despite being the greatest advocates of AI-based agents, we do feel you. It is tough not to lose your temper when you see “I’m sorry, I don’t understand your question” for the 3rd or (God forbid) 5th time. Our general suggestion would be — do not design dead-ends in your dialogue! But more specifically, make sure that your customer reaches a contact point they are undoubtedly seeking. You may consider providing a Live Chat handover that automatically connects a user with your company rep. If not, a plain text response can work wonders too! After a second failed attempt to understand the query, your assistant may display contact details, ideally department-specific — adjusted to topics covered in the dialogue.
And what are your thoughts? What actions can upgrade the conversational UX? If you want to learn more about our approach, or plan your deployment with us, visit our website.
