Password Reset Use Case

Some might ask: If a chatbot resembles a mobile app or web site, why not just build the web site or mobile app instead of a chatbot? Building a mobile app or web site instead of a chatbot is a choice made by an organization. But, if you look at the situation from a design perspective and consider the amount of time and money required to develop the three options, chatbots are much cheaper and may not require a complex design. Furthermore, chatbots can be made available in a web site or mobile app to create a conversational experience with the user community. As mentioned previously, you can do this using the Microsoft Bot Framework direct-line API or web chat control. Thus, chatbots are a simple way to resolve complex problems, such as resolve consumer queries in utility business; manage tasks efficiently in a 24/7, 365-day support services environment, help people file insurance claims with supporting information, get real-time details of flights during travel, schedule doctor appointments, answer health-related queries, reset passwords, and protect against fraud in financial services.

Embed a Chatbot in a Web Site

In general, chatbots reside outside web sites or mobile apps, but there are several examples when chatbots are embedded in web sites or mobile apps so that users can seek information in the most efficient way. Sometimes, as a result of the complex structure of a web site, users are unable to find the information they need. Using a chatbot, however, they can engage in a useful conversation to get information. As we have seen, chatbots can resolve simple issues and hand off more complex issues to human agents. In this section, we look at the integration between chatbots and web sites, and the process of using back-channel mechanisms to facilitate communication between chatbots and web pages. Microsoft provides two ways of integrating chatbots with web sites: Skype web control and an open source web control.

Embed a Chatbot Inside a Mobile App

IoT App

An IoT app can communicate with the Microsoft Bot Framework using a direct-line API. It can also use Microsoft Cognitive Services to enable capabilities such as image recognition and speech.

Design Knowledge Chatbots

Knowledge chatbots are chatbots that provide knowledge-based responses to user queries. What will be the weather in the city next week? Which is the nearest coffee shop to my house? Which hit film is available in the nearest movie theater? Does my credit card offer access to the lounge facility at the airport? If a chatbot is designed to answer these questions, it is a knowledge chatbot. Knowledge chatbots are smarter and more powerful compared to FAQ-based chatbots, which are designed to answer with predefined replies. Regardless of their design, chatbots pull information from various relational databases and present it to users in an understandable format.

Search Using Chatbots

The search functionality is a valuable asset for chatbots. A standard search provides accurate results based on the input string entered by a user. For example, if a user enters “India” in the search results, entries with a precise match and close match to India appear. If a user asks a knowledgeable chatbot for information on music by Impala, the chatbot responds with relevant information, as shown in Figure 7-8.

If the chatbot is unable to provide a precise match, it can present a result with the comment: “Here is the event that best matches your search” (Figure 7-9). Based on the accuracy of the search result, the chatbot can tailor communication while presenting results to the user.

If there is no accurate or precise match, but a close match for a search string such as “confidence is low”, then the chatbot may respond with the search result: Hmm . . . were you looking for any of these events? (Figure 7-10).

Using Search to Guide a Conversation

If there is a motivation to build chatbots to enable basic search engine functionality only, then there is no requirement for a chatbot. A conversational interface offers multiple benefits that users cannot get from a typical search engine using a web browser.

Knowledge chatbots are most effective when they are designed to guide the conversation with a user. Conversation consists of an exchange between a user and a chatbot. During the conversation, the chatbot has the opportunity to ask questions, present options for selection, and validate outcomes, which a standard or basic search is incapable of doing. In Figures 7-11 though 7-14, the chatbot converses with a user using facets and filters until it locates the information requested by the user.

By processing the user’s input and presenting relevant options during the conversation, the chatbot connects the user to the information being sought. After the chatbot delivers this information, it can also provide guidance on more efficient ways to find similar information in the future. Check out Figure 7-15.

Azure Search

Azure chatbots can be enabled with Azure Search to create an efficient search index that a chatbot can easily search, facet, and filter. Figure 7-16 shows an example search index created using Azure Portal.

To access all properties of the data store, set the property as “retrievable.” You expect to find musicians by name, so set the Name property as “searchable.” Last, you want to facet and filter over musicians’ eras, so mark the Eras property as both “facetable” and “filterable.” Faceting determines the values that exist in the data store for a given property. Figure 7-17 shows there are five distinct eras in the data store. Filtering returns only specified instances of a certain property. For example, you can filter the result set to contain only items where era is equal to “Romantic.”

Triage

In this situation, the chatbot provides a response to the user queries and later collects user data such as name, e-mail address, and any pertinent information related to the queries, then stores it in a database to transition conversation control to a human agent. Using a chatbot to triage information helps human agents work on the complex task to be addressed.

Escalation

Typically, in a help desk scenario, a chatbot may be able to answer basic questions and tend to trivial issues while performing simple operations such as resetting a user’s password. However, in a situation when the user issue appears to be complex and the chatbot is unable to manage it, a human must intervene. In these situations, the chatbot must understand the set of issues it can fix and those it cannot fix and thus requires a human agent. There are multiple ways chatbots determine that they need to transfer control of the conversation to a human.

User-driven Menus

One of the simplest ways for chatbots to handle user dilemmas is to present users with a menu of options from which they can select. Tasks that the chatbot can handle independently appear in the menu, along with a link labeled “Chat with an agent.” This implementation does not require advanced machine learning or natural language for understanding. The chatbot simply transfers control of the conversation to a human agent as soon as the user selects the “Chat with an agent” option.

Scenario Driven

The bot may decide to transfer control based on whether it determines it is capable of handling the scenario. The bot collects some information about the user’s request and then queries its internal list of capabilities to determine whether it is capable of addressing that request. If the bot determines that it is capable of addressing the request, it does so; if the bot determines the request is beyond the scope of issues it can resolve, it transfers control of the conversation to a human agent.

Supervision

In some scenarios, human agents prefer to monitor the conversation between the user and the chatbot instead of taking control. In the situation of a help desk, where a chatbot is communicating with a user to diagnose a computer problem, a machine learning model helps the chatbot determine the most probable cause of the problem. However, before advising the user to take a specific course of action, the chatbot can privately confirm the identified diagnosis and solution with the human agent, and then proceed with authorization. When the human agent authorizes the solution, the chatbot presents the solution to the user. The chatbot is still performing the majority of the work, but the human agent retains control over the final decision.