In this chapter, we focus on how we can create messages effectively in chatbots to build a better customer experience. We aim to develop an understanding of the customer experience, design messages in chatbots with various messaging properties, and see how to create a rich and intuitive customer experience by adding controls and card functionalities in the messages. Let’s being with an introduction to customer experience.
Customer Experience: An Introduction
A couple of decades ago, the software business revolved around software development, implementation, and application management services. Then came the era when many innovative mobile devices were launched; apps were developed; analytics such as sentiment analysis, social media command centers, dashboards, and Power BI were launched; and, finally, infrastructures moved into the cloud and we saw Office 365, Amazon Web Services, and software-as-a-service products. Today, we are living in a world of digital transformation in which product companies launch new technologies to create better customer experiences. The realm of digital transformation includes various technologies such as Azure, AI, chatbots, cognitive services, Pega, Liferay, C4Hana, SAP Hybris, salesforce marketing, commerce clouds, Microsoft CRM Dynamics, Jira, and more. IT companies use these technologies to create a digital customer experience.
Currently, there is paradigm shift in which both product and service companies aim to create a great customer experience. What exactly is a great customer experience? A customer experience is said to be a combination of the reactions, emotions, and perceptions a customer experiences while using different media to interact or collaborate with a business, product, or employee.
What is the customer experience?
In this chapter, we focus on the great customer experience produced by Azure chatbots. However, to produce a great customer experience, we need to design the user experience in the chatbot. Let’s look at how to design a chatbot to create a great user experience.
Chatbots for a Greater Customer Experience
Azure chatbots can be designed using various features such as buttons, images, text, and rich cards displayed in the form of a list or carousel. If the chatbot sits inside channels such as Facebook Messenger, Slack, Skype for Business, and others, then rendering these features depends on the channels. There are cases when features are not supported by channels. In these situations, the channel tries to “down-render” the message contents in the form of text or as a static image, which affects appearance. There are exceptional cases when a channel cannot support a few features at all. For example, GroupMe clients installed on devices cannot display a typing indicator while interacting with a user.
Rich User Controls
Rich user controls are common user interface controls such as buttons, images, carousels, and menus that the bot presents to the user and with which the user engages to communicate choice and intent. A bot can use a collection of user interface controls to mimic an app or can even run embedded within an app. When a bot is embedded within an app or web site, it can represent virtually any user interface control by leveraging the capabilities of the app that is hosting it.
Web site developers have been using modern user interfaces to engage users with mobile apps or web sites to produce a great user experience. The same philosophy can be adopted by chatbots by using modern user controls to understand user expectations. When booking an airline ticket, if the chatbot shows a Book Ticket button at the end, then it is convenient for users to achieve expectations with a single action, rather than forcing users to type “book ticket” so the bot can proceed with the booking. The latter case may contain typos, which results in a delay and confusion for the user and the chatbot. As another example, if a user wants to know nearby restaurants and the chatbot can display the top-five nearby restaurants from a Form button, it is quite convenient for the user to make a choice, which creates a good customer experience.
Cards
Cards allows chatbots to display their message in the form of visual, audio, and/or selectable messages, and assist in conversation flow. If a user is required to select an item from a set list, the chatbot can then display a carousel of cards, each containing an image, text for instructions, and one button for making a choice. If a user wants to select multiple items from a given list, then the chatbot can display a smaller single image and a collection of buttons with various options for the user to make a choice. Cards provide in-depth information using audio or video output, or receipts that detail a shopping experience, for example. There are different kinds of cards available and each depends on the business requirement for which the chatbot was developed. Cards create a seamless engagement with users. Let’s look at some different kinds of cards, the type of action they promote, and their recommended use.
- Adaptive cards: Adaptive cards (Figure 4-2) are typically used for cross-channel card deployment. They can adapt the look and feel of the host channel.
Figure 4-2Sample adaptive card
- Animation cards: Animation cards (Figure 4-3) play GIFs or short videos.
Figure 4-3Sample animation card
- Hero cards: Hero cards (Figure 4-5) typically contain a single picture, with text and multiple buttons. They are used to highlight the user’s choice visually.
Figure 4-5Sample hero card
- Thumbnail cards: Thumbnail cards (Figure 4-6) contain one thumbnail image, text, and multiple buttons. They are used to highlight buttons visually for the user.
Figure 4-6Sample thumbnail card
- Receipt cards: Receipt cards (Figure 4-7) provide comprehensive receipts to users that list the items purchased by the user.
Figure 4-7Sample receipt card
- Sign-in cards:Sign-in cards (Figure 4-8) are used to enable user login. These cards normally contain text with instructions and one or more buttons for users to make a choice.
Figure 4-8Sample sign-in card
- Suggested action cards: Suggested action cards (Figure 4-9) contain multiple buttons linked to actions. When a button is pushed, the card disappears.
Figure 4-9Sample suggested action card
- Video cards: Video cards (Figure 4-10) play videos using a URL or they stream videos.
Figure 4-10Sample video card
- Carousel cards: Carousel cards (Figure 4-11) are presented as a collection that allows users to view, at a glance, a series of possible user choices.
Figure 4-11Sample carousel cards
Cards allow you to design your bot once and have it work across a variety of channels. However, not all card types are fully supported across all available channels.
As stated earlier, an important thing to remember during the design of a chatbot is to use common user interface elements. The principal design objective is to resolve user problems in the most efficient way possible and avoid integration of natural language into the chatbot to make it more complex. Try using the least user interface controls to enable the chatbot to resolve user problems; add other elements if user interface controls cannot help.
Text and Natural Language Understanding
Chris
My name is Chris
Chris Stevens
Mr. Kris Stevens
A specific question was asked, but the answers may be entered in a variety of formats that may confuse the chatbot, and thus the user, and may lead to a bad user experience. Another open-ended question is: How are you feeling today? Again, a different combination of answers can be expected from users. What is the weather outside? This question can trigger multiple combinations of answers.
The important thing to keep in mind is to ask specific questions without having to use a LUIS service; this helps reduce chatbot complexity. A question such as “How are you feeling?” could have the answers Good, OK, or Bad, allowing the user to make an easy choice. There are DevOps chatbots that run on specific commands used to manage VMs. The Start/VM1 command triggers the first VM to start. The command Reboot/VM2 reboots trigger VM2.
In short, avoid using natural language understanding and ask specific questions, making use of cards and user controls to create a great customer experience and reduce complexity.
Knowledge-based chatbots or question-and-answer chatbots can answer user questions based on documents or information loaded in a database. QnA Maker and Azure Search are both technologies designed specifically for this type of scenario. A user may ask a general question. When designing a chatbot that answers questions based on structured or unstructured data from databases, web pages, or pools of documents, consider using technologies designed specifically to address this scenario. Do not attempt to resolve the problem with natural language understanding such as LUIS.
In a scenario in which users want to order a hamburger, they can engage with a chatbot with questions such as: Where is the nearest burger joint? Where is the nearest MacDonald’s within three miles? Show me combo burger offers. In this type of scenario, natural language understanding may be useful, and a LUIS API can extract the main component of words from a user’s question to understand intent and provide the desired answer. When developing chatbots using natural language models, it is dangerous to assume that users will provide all the information needed in their questions. However, design your chatbot in such a way that users do not have to repeat the same question in different ways. Request precise information from the user.
There are chatbots that use cognitive technology, such as speech, for input and output. With these chatbots, speech is the only means of communicating with users.
Choosing among Rich User Controls, Text and Natural Language, and Speech
Naturally, human beings like to communicate with each other using a combination of gestures, voice, and symbols. Similarly, one can design chatbots using rich user controls, text, and speech. These communication methods can be joined together; there is no need to choose one over the other. I illustrate this with a cooking example. This particular cooking chatbot shows users recipes in the form of images or videos. To choose a specific recipe, users simply touch the screen (this chatbot is enabled with rich user controls). In addition, users can input information to search for a specific recipe. While designing, incorporate user experience elements that support different ways in which users can interact with a chatbot.
Key Chatbot Message Concepts
At this point, we have learned about the following elements of chatbots: rich user control cards, natural language understanding, and speech techniques that are used while designing chatbots. Now we’ll focus on creating messages in chatbots and adding media attachments, rich cards, speech messages, and input/suggested messages. However, before delving into these subjects, let’s go through some key concepts used frequently in chatbot development.
Connector
The Bot Frame Connector provides a single REST API that facilitates communication between chatbot and user by transmitting messages from the chatbot using channels such as Skype, e-mail, and Facebook Messenger, and then transmits back from the channels to the chatbot.
Activity
ConversationUpdate: The chatbot was added to the conversation and users are added or removed from the conversation.
ContactRelationUpdate: The chatbot was added or removed from the user’s contact list.
Typing: During a conversation, the user or chatbot is compiling a response.
UserDeleteData: The user has requested the bot to delete user data it may have stored.
EndOfConversation: Marks the end of a conversation
Event: A communication was sent to the user by the chatbot that is invisible or not seen by the user during the chat.
Invoke: The user is requesting the bot to perform a specific operation.
MessageReaction: A user is reacting to the message by, for example, clicking the Like button.
Dialog
Dialog is used to model and manage conversations between a chatbot and users. A dialog maintains a stack of dialogs that are active during a conversation. Dialogs are portable across computers, which makes it possible to scale chatbot implementation.
FormFlow
FormFlow is used to collect information from users during conversations with a chatbot. For instance, when a user wants to order a pizza, the chatbot must get information from the user, such as type of crust, type of pizza, toppings, and so on. With basic guidelines, FormFlow can generate dialogs automatically to manage a pizza order conversation, for example.
State
The Bot Builder Framework has the capability of storing and retrieving state information pertaining to the user, the conversation, and user-specific information within the context of a specific conversation. During a conversation, information present in the state can be reused. In short, the state is a kind of database; in the Bot Builder Framework, data storage is possible in SQL Server.
Now let’s focus on message properties that can be used during conversations between users and a chatbot.
Create Messages
During communication with users, chatbots send information to users and receive information from users as well. The information exchange exists in the form of plain text, text to be spoken, suggested actions, media attachments, rich cards, and channel-specific data.
Send a Customized Message
Attachments
The Attachments property of a message activity is used to send and receive media attachments such as images, audio, videos, files, and rich cards.
Entities
The Entities property of a message consists of an array of open-ended schema.org objects that allow for an exchange of common contextual metadata between the channel and the bot—for example, “mention” someone during a conversation, share location-related information using geographic coordinates or a postal address.
Channel Data
Channel data are used to implement channel-specific functionality, such as creating a custom e-mail message, a full-fidelity Slack message, a Facebook notification, a Telegram message, and a line message.
Text-to-Speak
The Speak property of a message can be used to specify the text to be spoken by a bot on a speech-enabled channel.
Suggested Actions
The SuggestedActions property of a message activity can be used to show buttons that the user can use to submit input. The buttons disappear after the user makes a selection.
Add Media Attachments to Messages
An exchange of messages between a user and a chatbot may contain media attachments (e.g., image, video, audio, file). The Attachments property of the Activity object contains a list of Attachment objects that contain media attachments and rich cards used in the message. The connector service transmits data in such a way that channels are able to retain messages. If the attachment is a file, the hyperlink is displayed in the conversation.
Adding a media attachment
Add Rich Card Attachments to Messages
An exchange of a message between a user and a chatbot can contain rich cards in the form of lists or carousels. The Attachments property of the Activity object contains a list of Attachment objects: rich cards and media attachments within the message. Earlier in the chapter, we saw different types of rich cards, such as adaptive, animation, audio, hero, thumbnail, receipt, sign-in, and video.
Response using an adaptive card
Process Events within Rich Cards
Open URL: Try opening the URL in the browser.
IM Back: The chatbot receives the text the of message when the user clicks a button or taps a rich card. This message is visible to all participants in the chatbot.
Post Back: The chatbot receives the text of the message when the user clicks a button or taps a rich card. Some channels may display this text in the message feed and it will be visible to all conversation participants.
Call: Then, Call Destination with phone call: Tel: 999222
Play Audio: Using this URL of the audio to be played
Play Video: Using this URL of the video to be played
Show Image: Using this URL of the image to be displayed
Download File: Using this URL of the file to be downloaded
Sign In: Using this URL of OAuth flow to be initiated
These card actions are used to add different rich cards to the conversation between the chatbot and the user.
Add Speech to Messages
To create a speech-enabled chatbot, you can construct messages—in other words, precise text to be spoken by the chatbot. You can also influence the state of the user’s microphone by specifying an input to indicate whether the chatbot is able to accept, is expecting, or is ignoring the user message. There are multiple ways to specify text to be spoken by a chatbot on a speech-enabled channel. You can use message properties such as Speak, call the IDialogContext.SayAsync() method , and specify prompt options—Speak and retry—in the message prompt.
To specify text to be spoken by a chatbot, you can use a plain-text string or a string that is formatted in Speech Synthesis Markup Language, which is an XML-based markup language that enables you to control various characteristics of your chatbot, such as voice, rate, volume, pronunciation, pitch, and more.
To influence the user’s microphone to indicate whether the speech-enabled chatbot is accepting, expecting, or ignoring inputs given by user, the message properties should be set to InputHints.AcceptingInput to indicate the chatbot is ready for input and is not waiting for a response from the user. The Message property should be set to InputHints.ExpectingInput indicates the chatbot is waiting for a response from the user. Last, the Message property should be set to InputHints.IgnorningInput indicates the chatbot is unready to accept input from the user.
Add Suggested Actions to Messages
By adding suggested actions to your conversations, you can tap a button to provide input to a chatbot. Also, users do not have to type text and, as a result, there is an improved user experience that is facilitated by providing intuitive buttons for users to answer questions and select responses from multiple choices. In addition, buttons in the form of cards double the user experience because, as soon as they are tapped, they disappear from the screen. As a result, users are prevented from tapping stale buttons during a conversation.
To add suggested actions to a message, set the SuggestedActions property of the activity to a list of CardAction objects that represent the buttons to be displayed to the user during a conversation.
Send and Receive Activities
The Microsoft Bot Framework has a REST API that facilitates chatbot and user communication through multiple channels such as Skype, Slack, e-mail, and others. Communication between a chatbot and the user takes place by transmitting a message from the chatbot to the user and vice versa. To create a reply, the connector should use the message activity object to transmit information between the user and the chatbot. Every activity contains information, such as who created the message, the context of the message, and the recipient of the message. By calling Connect with Message activity “Reply to Activity,” you can create and send a reply message. Furthermore, you can send a nonreplyable message by calling the message property SendtoConversation. The CreateMessage activity method helps to create the new message and sets all property values as needed.
To start a new conversation by a chatbot with one or more users, the message activity Create Direct Conversation is used to create a private conversation. The message activity Create Conversation helps to create new conversations and opens them with one or more users. Last, the Send to Conversation method helps in sending newly created messages.
Implement Global Message Handlers
Message handlers
Intercept Messages
Messages between chatbots and a group of users can be intercepted by middleware functionality in the Microsoft Bot Framework. For each intercepted message, you can save or inspect the message, or create a conversation log. By default, the Bot Framework does not save this information automatically for privacy reasons. It is a best practice to tell users that their conversation with the chatbot is being saved and to explain the specific reason for saving it.
Send Proactive Messages
A chatbot usually sends messages to users that are related directly to the user’s prior input. But, in some cases, it is useful for a message not to be related directly to the last message sent by the user. These messages are called proactive messages . There are various business scenarios in which proactive messages are useful. If the user has set up a reminder in a bot, then the chatbot will send the reminder in the form of a proactive message when the time for the event is near. Furthermore, a chatbot can be asked to keep a close eye on sales offers and then trigger notifications when there is a dip in price. When this is set up and a price decrease occurs, the chatbot delivers an e-mail notification to the user. In another situation, a user may want to make a hotel reservation based on a price cap. When the chatbot finds a reservation that meets the user’s requirement, it sends a notification to the user. Proactive messages are also used to inform users that their input is getting processed or the chatbot is waiting for user input to continue the conversation.
Unexpected proactive message
In the example in Figure 4-14, we can infer that the user has set up a reminder for hotel prices to drop by certain limit. However, while booking a reservation for a different trip, a notification pops up and the user gets confused. This situation can be handled in a better way, such as displaying reminder setups in a different color, sending notifications after a booking is completed, canceling the current booking and delivering a notification, or interrupting the current booking until the user responds, then switching back to the current booking.
Ad hoc private messages are the simplest form of proactive messages, and chatbots inject them into messages whenever triggered, regardless of whether a user is engaged in a separate conversation with a chatbot. Chatbots can also send a dialog-based proactive message to a user. In this case, the chatbot must first collect and save information from the current conversation.
Summary
With this, we come to the end of this chapter. During our journey in this chapter, we enhanced our familiarity with the customer experience and then examined how we can make conversations between chatbots and users intuitively strong by using rich controls and cards. Last, we looked at connectors and different message properties to help us build conversations in the most efficient and intuitive ways.