USER EXPERIENCE IS EVERYTHING, IT ALWAYS HAS BEEN BUT IT’S UNDERVALUED AND UNDERINVESTED IN – EVAN WILLIAMS
In layman terms, UX can be defined as the process of improving user satisfaction of a product by enhancing its usability, accessibility, and pleasure provided while interaction.
Whether it’s a website, application or chatbot or any digital product, UX holds the same importance as the product’s visual identity. In fact in a study, it was even found that companies with highly effective UX have increased their revenue by 37%. The effectiveness of the user’s intuition, emotion and delight can be enhanced with a great UX design.
With the latest advancements in Natural Language Processing, Machine Learning, Artificial Intelligence etc. chatbots have quickly become the talk of the town in the tech industry.
This rapid development and growth of chatbots led to a new trend in UX – Conversational UX.
What is Conversational User Experience?
If you are reading about chatbots for a while, you must have come across this term at some point or the other. As AI powered chatbots, virtual assistants, natural-language interactions etc. are expected to be the next big thing in the communication world, the need for a seamless, natural and human-like conversation between man and machine is rapidly rising. To simplify, conversation user experience can be defined as the way users experience interacting with a chatbot or an intelligent assistant.
The future of any intelligent assistant/bot depends on how natural it can interact with humans and how swiftly it can understand and address queries. This is where conversational user experience plays an important role.
An excellent conversational UX helps users reach their goal in the shortest time with maximum end-user experience. Also, the user doesn’t have to learn anything new to converse with the bot and gets the required data by asking the bare minimum number of questions.
Let’s take an example of Monica to differentiate between a bad and good conversational UX.
Bad UX
Chatbot: Hello there, please enter your leave requirement.
Monica: I need to apply for a casual leave.
Chatbot: I’m sorry. I did not get that. Please give your instructions in the format TYPE / NUMBER OF DAYS / START DATE / REASON
Monica: Casual 4 6-August-2017 Vacation
Chatbot: I’m sorry. I did not get that. Please give your instructions in the format TYPE / NUMBER OF DAYS / START DATE / REASON
Monica: Casual / 4 / 6-August-2017 / Vacation
Chatbot: Please check your inbox for email confirmation.
Good UX
Chatbot: Hello Monica, how can I help you?
Monica: I need to apply for a casual leave for four days.
Chatbot: Okay Monica, May I know the start date of your leave?
Monica: 6-August-2017
Chatbot: Awesome. What is the reason for your leave request?
Monica: Vacation
Chatbot: Thanks for your input. Casual leave from 6th August is applied. Please check your inbox for email confirmation.
In the first case, Monica has to fail a couple of times before she gets the end result. Whereas in the second case, the flow of the conversation never breaks and she doesn’t have to struggle in learning new commands. A bad conversational user experience results in the end-user getting frustrated and as a result leaving the conversation mid-way.
For a customer-centric chatbot, a bad conversational UX can result in the loss of money and loyal buyers for the company.
For an enterprise bot, a bad conversational UX will lead to lower adoption rates and thereby result in a low ROI.
Now that we have discussed the importance of a good conversational UX, let’s delve into the part where we learn how to build one. Here are 4 tips for a highly effective conversational UX:
1. Build a solid persona
Making a chatbot/virtual assistant more human-like is one of the most important factor for it to be successful. People prefer to converse with an assistant which interacts them like a human.
Needless to say, unlike machines, humans portray a range of emotions. Now, a chatbot cannot (at least for now) express emotions. Nevertheless, you can always build a bot which has a strong and consistent persona with some key personality traits – Humor, wit, sarcasm etc.
Identify and assess each persona (age, demographics etc.) of your potential end-user and build a robust and consistent persona for your bot. The language used by the bot should be consistent and should reflect your brand’s identity. For example, even the slightest humor from a healthcare bot can be considered inappropriate by users.
2. Make the tone sound natural
The tone of the bot’s language should not sound too robotic. It should be natural, engaging and encourage users to continue the conversation. Just like in the example above, the bot should use natural language phrases like “How can we help you” relevant and appropriate adjectives like “awesome”, “great” etc.
Note: The tone of the bot should reflect the bot’s persona as well. Choose the words and phrases according to your bot’s and potential target audience’s persona.
Paraphrasing: While creating utterances for the language model training, the bot should come up with a variety of phrases that describe the same intent or purpose. Apart from different phrases, it’s also important to identify related keywords and add them as utterances to obtain the desired outcome from the trained language model.
3. “Pushing” personalized content
The bot should not wait for the user to start using it. Let’s take the example of Janice who is using an enterprise bot.
Janice is a hard working sales manager in a large enterprise. By adding a bit of personalization, the enterprise bot can send customized and real-time alerts to her whenever necessary. This not only adds to an enhanced user experience but also improves Janice’s productivity. She doesn’t have to ask the bot every day for the latest sales updates.
A bot which can push personalized messages to the user has higher adoption rates and encourages users to interact with it more and more.
4. Use of visual content and rich interactions
A picture is worth a thousand words.
Often, expressing textually can get difficult. A user can request a plethora of data or a complete report at a time. Sending the response in the form of a text in such cases can be a little overwhelming for the user to understand and grasp easily. In such cases, the bot should respond with buttons, images, videos, graphs & charts or any rich media.
Let’s suppose Marshall, who uses an intranet bot in his enterprise, asks the bot for an overview and trends of the company’s financial transactions for the last 3 months. Now instead of sending a response in the form of clunky text, the bot can respond with a graph that shows company financial transactions by month.
5. Level 0 Phrases
Also called the “Golden Phrase,” a level 0 phrase is the most basic sentence that accurately describes what a user’s intent is. These phrases must be used by the bot to present options to the user in case of multiple matching intents.
6. Standard intents
Apart from training the NL model to identify different use case based intents, the bot should also be trained to understand certain standard intents that users typically use when interacting with a bot. Examples include sentences like “help,” “cancel,” “stop” etc.
The same intents can mean different bot responses and different actions based on the context of the conversation. This should not be an after-thought, but a baked in up-front during the NL training.
7. Guided conversations
When users have their first set of conversations with the bot, they look for guidance and help. A good chatbot should be ready to guide users whenever they need help or get stuck. Over time they get used to the interaction and can communicate their needs and asks faster and with fewer sentences.
8. Entity fulfilment
Every bot deals with different kinds and complexities of entities within a user’s sentence, question or request. Often, these entities (or user data inputs) are required to perform an operation.
For example, booking a shuttle cannot be done without the current location, destination and the number of seats required. In case if users fail to specify the required inputs, the bot should be able to get the missing ones from the user.
In addition to all these tips, we also feel a bot with a good conversational UX should
- Send a range of responses, instead of a simple yes or no every time
- Never fallback. It must quickly recover and provide meaningful options and responses to the user when it doesn’t understand his/her intent
- Never commit spelling mistake or grammatical errors
- Be accompanied by the correct UI elements like emojis, cards and all the relevant entities to keep the conversation as seamless as possible
- Maintain the context so that conversations are more natural and ‘human-like’
