IBM Watson Virtual Agent uses the Watson Conversation API to help you build automated, conversational chatbots to provide self-service experiences to customers.

In "Watson Conversation Can Generate Chatbots and Virtual Agents for Your Business, Part I," we reviewed how IBM Watson Conversation can help enterprises build chatbots that can use natural-language text and spoken input from humans to support customer-service applications. IBM Watson Virtual Agent (WVA) carries this concept a step further to provide customized tools that can help you create automated self-service environments for customers, as well as analytics tools for customer engagement.

Virtual Agent and Watson Conversation

WVA is obviously based on Watson Conversation and uses the Watson Conversation API, but all three aren't the same thing although all of them are available as cloud services.

Watson Conversation provides the means for training a Watson iteration to understand natural language expressed in selected national languages (English, French, German, Italian, Portuguese, and Spanish so far), comprehend the meaning of given phrases and sentences, and categorize each phrase and sentence according to a classification hierarchy specified by the developer or user. It also includes software mechanisms for building dialogue streams that a bot or Virtual Agent will use to converse with human beings.

The IBM Watson Conversation API is a specific tool that includes access to Conversation's capabilities to help developers construct conversational applications and deploy them to websites as a means of providing interactive services with which to engage customers. The API originates HTTP requests and returns responses in the JavaScript Object Notation (JSON) data-interchange format.

WVA is an IBM-built application system that already embodies all the capabilities of Watson Conversation, but it's flexible enough to let those deploying it turn on and off elements of its total range of capabilities (and modify it in other ways) to provide a version of WVA that is tailored for a specific business type or even an individual enterprise. It's essentially an app that lets you deploy a Conversation system without having to fully understand what's going on under Conversation's covers.

Intents and Capabilities

WVA has its own nomenclature to understand, as well as comprehending Conversation's own vocabulary. In Watson Conversation, an intent is whatever the human in the transaction is trying to express. In WVA, a capability is WVA's competence at interpreting intent. To interpret that intent, a WVA chatbot iteration will engage in a dialog, which in Conversation parlance is a blueprint of all the pathways a potential interchange between a customer and an app might follow. Dialogs are made up of individual utterances, which are human expressions Conversation can understand as isolated single concepts. Entities are terms or names of things that relate to intents in Conversation - for example, a specific product the customer might be trying to find.

IBM has provided WVA with capability packs, which are clusters of prefabricated capability groupings that contain thousands of example utterances and counterexamples. What's more, there are various versions of WVA that have been optimized for general customer service, energy, banking, retail telecommunications, and insurance (this final version available only in English currently). By selecting a version, equipping it with entity names specific to your business, and perhaps adding some dialogs you know from experience are likely to be expressed by visitors to your website, for example, you can create a Virtual Agent that is capable of handling general customer service inquiries specific to your business.

Each implementation of WVA contains four major components. The Conversation service includes all the artifacts needed to provide a capability - namely, some established intents, entities, and dialogs - as well as processing power to run the chatbot. The second component is the bot itself, already capable of handling basic functions but also including a configuration tool that lets you enter information unique to your enterprise. The third component is an interface to your company website and associated apps and databases needed to underpin the customer services you want to provide. The fourth component is a chat window that is the virtual agent interface that customers will use to communicate, available as a customizable widget. Developers can also build their own widget with a provided SDK.

WVA's Built-in Abilities

To get some idea of how WVA might function in your company's environment, let's look at a breakdown of the capabilities included in the "general customer service" implementation.

Under the general heading of account management, there are numerous basic functions a customer might access. These include the ability to change names associated with an account, what the account's privileges might include, who is authorized to make changes to the account, what the account's billing address is, what the account number is, what loyalty programs the customer has access to and the customer's eligibility, what information about the customer is part of the account, and ways of determining, redeeming, and transferring reward points to which the customer may be entitled. In addition, there are services for opening and closing accounts, making other general inquiries about the account, and changing account passwords and security questions.

Other core capabilities include the bot's ability to tell the customer it doesn't understand a question or can't provide an answer even if it does understand the question (a trait called "None of the Above"), the ability to refer the customer to a human agent (or some other service external to the bot), support for implementing the prefabricated dialogs, and the ability to create custom dialogs to either replace or supplement the prefabricated ones.

The None of the Above trait can't be turned off, by the way. It stands as a standard response for any exchange the bot can't comprehend. What can be altered is the actual dialog or action the bot is permitted to take if a None of the Above condition arises. The customer can be diverted to a special Conversation workspace that tries to match the customer's utterance to an intent the bot can understand, to a custom dialog that tries to elicit enough additional information to make the utterance comprehensible to the bot, or to some other alternate response, such as linking the customer to a human agent.

WVA also includes some of the analytic capabilities one would expect from AI like Watson. There is an engagement metrics dashboard that tracks customer interactions with each implementation of WVA you're using and provides data that can help you enhance its interactions. These metrics are updated every 30 minutes and show such data as how many conversations are taking place, how many new and returning users are engaged, what topics customers are most and least interested in, peak conversation times, interactive diagrams that show how the WVA implementation is interpreting input, what intents and entities are most in use, and what user goals the agent was unable to satisfy. It also creates a database of all interactions for further diagnosis and study.

WVA is available in two editions. Standard includes all the features discussed, while Premium adds secured data isolation features for enterprises needing that service.

WVA Requirements

WVA requires browser software installation on systems using it. These currently consist of IE version 11, the latest version of Chrome for your system, the latest edition of Firefox plus ESR 38, and Safari for Mac.

WVA Benefits

WVA provides a self-service experience that can augment your current customer service regimen and, in some cases, even replace it in time. It can reduce the expenses of providing support to customers with an all-human staff, personalize and increase customer gratification, and resolve many routine customer questions. If customer service is a key component of your business, WVA is worth some serious consideration.

John Ghrist

• Email
• Site

John Ghrist has been a journalist, programmer, and systems manager in the computer industry since 1982. He has covered the market for IBM i servers and their predecessor platforms for more than a quarter century and has attended more than 25 COMMON conferences. A former editor-in-chief with Defense Computing and a senior editor with SystemiNEWS, John has written and edited hundreds of articles and blogs for more than a dozen print and electronic publications. You can reach him at ghrist@comcast.net.