Conversational AI: Five principles for chatbot UX success

So, you’ve decided to create a chatbot that uses conversational AI to deliver a human-like experience, combining natural language processing, machine learning and big data analytics.

Firstly, congratulations! Once your virtual assistant is up-and-running, you’ll be able to save time, money and deliver a better experience to your users. However, these long-term gains require some short-term effort. At EBM, we’ve experienced the good, the bad and the ugly side of chatbot UX. Here are our top tips to ensure your bot is a success.

1. Make sure you have buy-in from the business

To avoid disappointment further down the line, before starting a project, it’s crucial to analyse where a chatbot can add the most value in your organisation. Once you’ve done this, all stakeholders should agree on both a target use case, and the parameters you’ll use to measure your chatbot’s effectiveness.

Next, ask yourself whether you have the right skills at your disposal. Conversational designers, business analysts, subject matter experts and integration specialists are just some of the people you may need for a successful conversational AI project.

Last but not least, make sure to manage expectations. All stakeholders should understand that chatbot UX will be at its worst when it first launches. Much like any human employee, time, investment and training are needed to develop your digital assistant.

2. Your chatbot isn’t human…

When it comes to building your chatbot, each intent needs to be carefully thought out. It’s important to remember that most chatbots don’t yet “understand” language in the same way humans do. This means that it’s a good idea to ask users to keep messages short and to avoid open-ended questions. Plus, try to gently guide users by giving them an indication of what the bot is expecting. It’s a good idea to make it clear to users this is a bot not a human, as they will interact with a chatbot differently to a human (i.e. users tend to be more concise with a chatbot). Another top tip is not to overdo empathy or humour – the wins from this will be small, but the losses can be considerable.

You’ll also find that grammatical points like negation can challenge your chatbot (e.g. not, no, and the prefixes un-, in-). Plus, opposing phrases like “cancel order delivery” and “order delivery” can confuse it.

Designing a voice-activated digital assistant is typically more complex than a written one – and requires some additional considerations. For instance, concise responses work better, as users will likely lose patience if a chatbot reads out several paragraphs at a time. Additionally, it’s worth disambiguating certain data, and selecting synonyms for commonly misheard phrases – like “overdo” and “overdue” for example.

3. …But it should have a personality

Having said this, try to avoid thinking of your chatbot as a form, or a simple decision tree. The beauty of conversational AI is that you can think in terms of a human-to-human conversation. Like with any conversation, this means you should vary responses, including greetings, and have a strategy for if the conversation gets stuck. This might mean disclosing more information, requesting further clarification or asking users to be more concise.

If a conversation does become too difficult to continue, the bot shouldn’t leave users hanging. To avoid frustration, it’s a good idea to offer a reset option. The same goes for business functions not supported by the bot. These should be gracefully handled, either by directing users to the correct resource or letting them know how to contact human support.

Before you get started, it’s a good idea to decide on your bot’s voice, tone and persona. Think of it the same way as any other customer-facing brand representative. Consider giving your bot a name and an avatar. For example, Canadian children’s mental health charity, Kids Help Phone, opted for a friendly, personable bot named Kip, designed specifically to engage with young people. On the other hand, the Institute of Chartered Accountants (England & Wales) chose a more professional tone for their bot, Mia.

It’s also worth including small talk in your intents. This can be helpful for handling users’ frustrations, and can reinforce your chatbot’s personality. However, try to keep small talk brief, and manage the conversation back to known use cases.

4. Context is important

Often, important concepts like “when”, “who” and “where” are implied in human conversations, rather than overtly stated. It’s therefore important to explicitly manage context in your conversational AI project. One way to do this is by altering the conversation based on who is speaking to your digital assistant. Let’s take the example of a chatbot at a children’s charity. A conversation with a child reaching out for help is likely to require different responses than one with a concerned teacher or neighbour.

When it comes to conversational AI, there are usually many paths to the same destination. For instance, some people will state their problem “lost package” rather than asking for a solution “can you help me find my package?”. You need to train your chatbot on the different ways your users could express their question/intention. This means you’ll likely have lots of training phrases for a single intent and response. This is why it’s also good to keep responses general, i.e. never start with ‘yes’ or ‘no’, as a user may not have asked a question, they could just have stated the problem. 

Additionally, remember that all words carry emotional weight, so you should be aware of how your chatbot sounds to users. For example, the phrase “You neglected to supply…” will likely elicit a negative response, whereas “If you could tell us…” will probably be interpreted more positively.

5. Be prepared to learn as you go

As mentioned above, no chatbot will be perfect when it launches. It takes time to train a digital assistant, and the best data comes from real users, not internal testing.

It’s also important to keep an eye on analytics and insights, so you can identify any gaps in your chatbot’s knowledge. These usually emerge over time, so be prepared to continuously test, fix and repeat.

As you’ve probably realised, conversational AI isn’t a magic wand that solves all problems. But it can be hugely effective if trained correctly. We hope we’ve inspired you to make the most of this exciting tool.  Want to learn more about how Filament CX can support you in developing a smart digital assistant? Schedule a demo.

Previous Article

Next Article