Today’s “Expert Tip” is about how to use conversational techniques optimally in the context of (Voice) Bot Conversational Design and thus improve the customer experience.
Questions and Confirmations
Questions and confirmations are essential elements of conversational design. These serve to increase the accuracy of the query requested by the customer, thus narrowing down, or classifying the facts and confirming them in the dialogue. At the same time, however, certain points must be taken into account in the dialogue between bot and human, as we can see from the following example.
Example 1 (Chat Bot):
Bot: „Would you like to expand an existing order or place a new order?“
Here, it is important to never leave out the “I would like something else” so as not to lose the customer because their individual request is not included, and the dialogue then ends in a dead end. In person-to-person dialogue, we always have no problem finding a direct link to a new topic or issue. As shown above, a bot must always make this option available because bot dialogues are designed in conversational flows and link points (out-in) are needed.
Once the user has placed an order, it should also be confirmed in the communication. This can be done in different ways:
Bot: „You have selected three pizzas Napoli and one pineapple. Please confirm this selection.“
In reality, no human being speaks like this, so it makes sense to make the order confirmation and request more in a more human manner:
Bot: “Ok, that will be three pizzas Napoli and one pineapple. Is that all, or would you like to order something else?”
Here you can see again that the option “Something else” is included. Another error that occurs frequently, especially with voice bots, is the confirmation request to users to use certain predefined keywords or word combinations as part of the confirmation.
Bot: “Say ‘Extend existing order’ if you want to extend your order or ‘New order’ if you want to place a new order.”
User: “New order.”
If speech recognition is programmed to use these two phrases, “Existing order” or “New order,” the following phrase provides an equal but more importantly, more human option:
Bot: “Do you want to place a new order or add to your existing order?” User: “I want to place a new order.”
The differences between “explicit” and “implicit” confirmations are important in this context. Explicit confirmations are appropriate, for example, when the confidence level of the customer statement is below 65 percent. Confidence level means that the bot has X percent probability of correctly understanding and categorizing the response.
User: “I would like to reserve a table at Gianni’s.”
Bot: “You would like to reserve a table at the restaurant ‘Da Gianni’?”
Conversational software is less than 65% unsure of what the user wants to know. Although not all software has an analysis or scoring module scores, above 90 percent should be a goal. The more complicated or sensitive the issue, the more often the use of explicit confirmation is advised. Once the bot is trained further downstream, confidence levels can be increased for each response. Likewise, it is important to request explicit confirmation from the user when dealing with sensitive topics.
User: “Please transfer 350 euros to David’s account.”
Bot: “I should transfer 350 euros to David Müller’s account, correct?”
User: “No, to the account of David Meier.”
Bot: “I’ll transfer the money to David Meier’s account, ok?”
Bot: “All right, the transfer of 350 euros to David Meier has been made.”
Analogous to explicit confirmations, implicit confirmations are also an effective tool in conversational design.
User: “I’d like to add a change of address.”
Bot: “OK, I’ll be happy to do that for you. What is the new address?”
User: “Hafenstrasse 5 in Hamburg.”
Bot: “Thank you. Your new address in Hafenstrasse is now stored. Should I also change your landline number?”
User: “Oh yes, it will change as well. The new number is 040-1234567”
Bot: “So, that’s three beers and a pineapple – thank you for your order.”
Bot: “Three beers and a pineapple will be delivered as soon as possible.”
Implied confirmations help speed up the dialogue and can also integrate other queries in parallel, which makes a bot dialog more human. It also helps to vary the confirmation wording at the beginning of a piece of information.
Bot: “Thanks, I’ll book the flight for you?”
Bot: “OK, the cab is ordered.”
Bot: “Great, I have all the necessary data for the booking.”
Bot: “Alright. The estimated waiting time is 3 minutes.”
Implied confirmations are also popular elements for chat. As a “test”, you can look at your own WhatsApp chat history to see how often they are used in human-to-human dialogue.
In general, options are a good tool for streamlining bot dialogues which allows them to run through more quickly. However, there are also sources of errors that must be considered.
Bot: “Are you single, married, in a committed relationship, or a widow or widower?”
Although this query can be answered relatively easily by a human, conversational design offers the separation of the two main elements here.
Bot: “Are you living in a committed partnership or are you married?”
Bot: “Are you a widow, widower or single?”
User: “I am single.”
When voice bots are used, it is advisable to offer a maximum of three to four options, as this is what users feel most comfortable with. You can see this portrayed in the following examples.
Example 12 (without separation):
Bot: “Today, I can book the following events in your area for you: theater, burlesque shows, musicals, standup comedy, ballet, jazz concerts, Cinema, Classical concerts, Clubs, Rock concerts, Restaurants, Bars…”
User: “Um. What was the fourth point again?”
Example 13 (with separation):
Bot: “Today, I can book the following events in your area for you: theater, burlesque shows, musicals or maybe you are interested in something else?” User: “What else is there?”
Bot: “I still have tickets for standup comedy, ballet, or various concerts. Are you interested in any of these options, or should I list other alternatives?” User: “What concerts are available?”
Bot: “Today there are tickets for classical music, jazz, pop or rock concerts available. Are you interested in any of these?”
Other useful elements when using options in conversational design are: Bulleted lists: “First, second, third…”
Iterations: “First, after, next…”; “In addition.”
Prioritizations: “The most important thing is…”; “More important is…” “Urgently needed is…”; “Last but not least…”
Resumes: “In summary…”; “In consequence, this means…”
Conversational and informational content
An important element for the usage rate of a bot is to ensure that the conversations are distributed, and that the bot does not engage in long monologues. As with human-to-human dialogue, this has a friendlier effect and motivates the user to remain in dialogue with the bot.
User: “I would like to order a cab.”
Bot: “I would like to order you a cab. The basic fee is 8 euros and 35 cents per kilometer up to a maximum of 20 kilometers and then 55 cents for each kilometer driven beyond that. The cab currently takes between 25 and 30 minutes to reach your location. Should I book it for you now?”
This statement is very long and overwhelms the user with too much unimportant information. This statement can be clearly focused, reduced and thus generate a better customer experience.
User: “I would like to order a cab”.
Bot: “Gladly. What is the destination?”
User: “Berlin central station”
Bot: “All right. The total cost of the ride will be about 15 euros. Is that OK?” User: “Yes”
Bot: “When should I book the cab for you?”
The last example feels much more pleasant because relevant information is compressed and well presented. Most importantly: splitting up the conversational portions also makes the user feel much more engaged in the conversation.
As in Part 1 of the “Speak Bot” post, introducing the bot is an essential element for increasing usage of this service. At the same time, you should use this introduction in such a way that there is recognition value for the customer even if they use it again, as this is generally perceived more positively. So instead of always using the same greeting “Hello, I am XXX the virtual assistant of junokai…..”, you can work here with variations as in the following examples:
Bot: “Hello XXX, nice to have you back. How can I help you today?”
Bot: “Welcome back XXX. What can I do for you today?”
Bot: “Hello XXX, I see that you have an open order here. Do you have any questions about your last order?”
This can be done using cookies, smart device recognition, or identification/authentication that has already taken place.
The examples make it clear that a good question and confirmation concept in the context of conversational design makes bot use as pleasant as possible for the user, but also as targeted as possible. Likewise, the aspects of conversational shares, information separation, and recognition features play a supporting role.
Carlos Carvalho – Senior Consultant