Carlos Carvalho, Senior Berater junokai
Expert Tip: Conversational Design – Do You Speak Bot? Part 1
In the new "Expert Tip" section, the colleagues from our consulting firm junokai share their knowledge and experience from their day-to-day consulting work in the field of customer experience and reveal their best tips and tricks.
In today’s Expert Tip, I would like to introduce you to some basic rules and pitfalls for creating conversational design dialogues.
Conversational design for (voice) bots includes knowing how to analyze and interpret data and information using a structured approach, as well as how to structure the output of a bot the so that the communication is as natural and emphatic as possible. This means that both worlds (bot and human) must be correctly understood, interpreted, and linked in the symbiosis of psychology, technology and conversational copywriting.
Basic understanding of a bot
Basically, bots and humans don’t understand each other at first, because human communication logic and understanding is based on different prerequisites and many years of experience. While certain phrases, gestures, and conversational logic are natural to us from birth through constant “training,” they are new to the bot or contradict its basic understanding.
Human communication is also defined by empathy, while bot communication is primarily defined by the analysis of structured data and the associated interpretation of the query.
Here are two examples of “miscommunication” between user and bot:
User: “Do you know what time it is?”
User: “What time is it?”
Bot: “In New York it is 7:31 a.m. and in London it is 12:31 a.m.” (but the user is currently in Berlin).
From the bot’s point of view, a correct answer was given in each of these two cases. However, humans evaluate the answer as wrong because humans expect a completely different answer.
Both answers contain two different types of interpretation errors. In the first case, the supposed error is that the typical human phrase was not interpreted correctly. In the second case, the bot needs more information about the user’s current location to give a correct answer.
Tip 1: When creating dialogues, make certain that typical-human questions in the spectrum are interpreted correctly and classified as such. Also, ensure that additional important information for the answer, such as location in the second example, is available or can be queried. Even if this sounds trivial, this is where the breaking point between human vs. bot communication lies. This is when the errors in dialogue design arise, and a user breaks off the communication, in order to talk to a real human. Trying the bot again is then much less likely – You never get a second chance to make a first impression.
At the beginning of a dialogue during a “bot introduction”, typical mistakes are made again and again, which affect the customers’ usage or actually prevent them from using the bot altogether.
Bot: “Hello, I am XXX, the virtual assistant of junokai. I’m learning every day and I try to help answer questions. To improve my skills, please rate your experience with me at the end of our conversation. What can I help you with?”
This introduction text contains a typical mistake. The focus is primarily on the bot, its needs, and not the customer’s. (Just count how many times the words “I”, “my” and “me” are used in this introduction text).
This needs perspective often arises when bot dialogues are designed and implemented solely from a technical/logical perspective without involvement of customer service or dialogue managers.
A possible alternative is Example 4, which shows a clear focus on the customer while narrowing the topic spectrum of the bot to customer service and making appointments.
Bot: “Hello, I’m XXX, junokai’s virtual assistant, and I can provide you with information on many customer service topics or make an individual appointment for you with one of our consultants. What can I do for you?
Tip 2: Put the customer’s concern at the center of the dialogue right from the start and simplify the presentation from the customer’s point of view. Involve experienced customer service staff in the dialogue creation, as they have good expertise and experience in customer dialogues.
A typical bot introduction mistake is to introduce a bot too arrogantly or with exaggerated competencies.
Bot: “Hello, I’m XXX, junokai’s intelligent assistant, and I can help you improve your customer service in the long term or set up a custom appointment for you with one of our expert consultants.”
The more neutral and focused you present your bot, the more usage rates will increase.
Cornerstones in bot response design
Bot information, also called prompts or utters, have five cornerstones to consider when creating dialogue.
Generally, only true information is conveyed and if no answer exists from the bot’s point of view, this is conveyed and ideally an alternative (e.g., forwarding to an employee or scheduling a callback) is offered.
A challenge for a bot is quantitative queries or responses that a bot must provide.
Bot: “How many rooms does your apartment have?”
User: “We have two bedrooms, one living room, one nursery…”
A less than customer-service friendly answer would be, “Please tell me a number” or if it only concerns chat, e.g., use a choice of synonyms. A good example:
Bot: “Excuse me. How many rooms is that in total?”
This dialogue is perceived to be much more customer-friendly and less demanding by the user than “Tell me a number, please.”
Bot: “You can pay with MasterCard, Visa or PayPal. Do you want to complete the checkout process?”
User: “I’ll pay with Visa.”
In this example, the bot’s question is closed, there is only the yes/no answer. However, as humans, we are used to answering such questions in the above manner. Therefore, in Conversational Design, do not always expect a human to give a yes/no answer to a closed question.
Communication of information or queries is reduced to only the most important and essential. No more, no less.
Bot: “Do they have all the required documents to complete their travel booking?”
User: “What are those exactly?”
Bot: “They are a valid visa, a current passport, and a Visa credit card or PayPal account if you want to pay with one of these methods. If you have these documents ready, say “OK” and we’ll be happy to get started.”
Certainly, other information might be useful for this trip (e.g., vaccination against tropical diseases or travel cancellation insurance), but these are not relevant for this particular booking process. At the same time, it does not make sense to omit any of the above items or to ask for them later, otherwise the booking cannot be completed.
Bot, “Your transaction has been successfully completed. A confirmation will be sent by mail.” Of course, this bot answer is correct from a purely formal point of view, but in reality, no human would communicate like this. Therefore, responses should also be made as human-like as possible. In this example, one option could be:
Bot: “OK, your order/booking is now completed, and you will receive a confirmation from us by mail. Thank you.”
A common mistake in Conversational Design is artificial empathy to the bot in return as in the following example:
User: “I can no longer make my connecting flight. What are my options?”
Bot: “Oh I can understand how inconvenient this must be for you. I’m looking into options for rebooking.”
Users are well aware that they are communicating with a machine, and especially in ambiguous or emotionally charged situations, artificial empathy tends to have a counterproductive effect. In addition, in the example, there may also be the option that the flight should perhaps not be taken because of the user’s private reasons for doing so. This makes the answer inappropriate from the user’s point of view and is interpreted negatively by the user.
Empathy in conversational bot design must be aimed solely at identifying the problem and having this confirmed by the customer.
User: “I can no longer make my connecting flight. What are my options?”
Bot: “I see – Is this about your Lufthansa flight 123 today at 22:30?
User: “Yes exactly, that’s the flight.”
Bot: “All right. I’m looking into the possibility of rebooking you on another flight. Do you absolutely need a connecting flight on the same day, or can I also consider options for the following day?”
Tip 3: When creating (voice) bot dialogues, check whether the essential elements of quantity, quality, relevance, form, and empathy are taken into account in the bot’s responses in the sense of the examples described here. If necessary, have non-involved members of the Conversational Design Team run through them in a dialogue reality check. It is here that errors in dialogue design become apparent and these weaknesses can be remedied.
Even if the creation of (voice) bot dialogues may seem trivial or intuitive at first, it follows clear rules, as we have shown. Following these is also recommended to simplify the communication between bot and human as well as to increase the bot usage rate.
It is important to understand that bots and humans have different modes of communication as well as focal points. Adopting these directly is not optimal and can even create the opposite effect. As in the examples shown, it is necessary to combine the respective characteristics of both sides as much as possible.
junokai is a consulting company based in Berlin that supports clients from different industries in all areas of customer service. junokai was founded in 2013 by experienced managers with extensive professional expertise in sales and marketing, and customer service. The company’s strategic pillar is its operational experience and focus on the areas of customer experience (CX), customer service and sales. For more information on junokai, please visit: www.junokai.de