Virtual Assistant Update

 

We recently published “Virtual Assistant Update.” It’s a broad and not too deep update on virtual assistant technologies, products, suppliers, and markets from the perspective of the five leading suppliers: [24]7, Creative Virtual, IBM, Next IT, and Nuance. These are the leaders because they:

  • Have been in the virtual assistant business for some time (from 16 years for [24]7 via its acquisition of IntelliResponse to four years for IBM).
  • Have attractive and useful virtual assistant technology
  • Offer virtual assistant products that are widely used and well proven.
  • Want to be in the virtual assistant business and have company plans and product plans to continue.

The five suppliers are quite diverse. There’s the public $80 billion IBM and the public $2 billion Nuance. Then there are the private [24]7, a venture backed company big on acquisitions and the more closely held Creative Virtual and Next IT. Despite these big corporate-level differences, the five’s virtual assistant businesses are quite similar. Roughly they’re all about same size and the five compete as equals to acquire and retain virtual assistant business.

By the way, across the past 12 to 24 months, business has been good for all of the five suppliers. Customer growth has been very good across the board. Our suppliers have expanded into new markets and have introduced new and/or improved products.

Natural Language Processing and Machine Learning

Technologies are quite similar, too. All five have built their virtual assistant offerings with the same core technologies: Natural Language Processing (NLP) and machine learning.

Virtual Assistants use NLP to recognize intents of customer requests. NLP implementations usually comprise an engine that processes customer requests using an assortment of algorithms to parse and understand the words and phrases in a customer’s request. An NLP engine’s processing is guided by customizable and/or configurable deployment-specific mechanisms such as language models, grammars, and rules. These mechanisms accommodate the vocabularies of a deployment’s business, products, and customers.

Virtual assistants use machine learning technology to match actual customer requests with anticipated customer requests and then to select the content or execute the logic associated with the anticipated requests. (Machine learning algorithms learn from and then make predictions on data. Algorithms learn from training. Analysts/scientists train them with sample, example, or typical deployment-specific input then with feedback or supervision on correct and incorrect predictions. A trained algorithm is a deployment-specific machine learning model. The accuracy of models can improve with additional and continuing training. Some machine learning implementations are self-learning.)

Complex and Sophisticated Work: Consultant-led or Consultant-assisted

The work to adapt NLP and machine learning technology implementations for virtual assistant deployments is sophisticated and complex. This is work for experts: scientists, analysts, and developers in languages, data, and algorithms. The approach to this is work differentiates virtual assistant suppliers and products. The approach drives virtual assistant product selection. Here’s what we mean.

All the virtual assistant suppliers have built tools and package predefined resources to make the work simpler, faster, and more consistent. Some suppliers have built tools for the experts and these suppliers have also built consulting organizations with the expertise to use their tools. Successful deployments of their virtual assistant offerings are consultant-led. They require the services of the suppliers’ (or the suppliers’ partners’) consulting organizations.

Some suppliers have built tools that further abstract the work and make it possible for analysts, business users, and IT developers to deploy. While these suppliers have also built consulting organization with expertise in virtual assistant technologies and in their tools, successful deployments of their virtual assistant offerings are consultant-assisted and may even approach self-service.

So, a key factor in the selection of a virtual assistant product is deployment approach: consultant-led or consultant-assisted. Creative Virtual, Next IT, and Nuance offer consultant-led virtual assistant deployments. [24]7 and IBM offer consultant-assisted deployments. For example, IBM Watson Virtual Agent includes tools that make it easy to deploy virtual assistants. In the Figure below, we show the workspace wherein analysts specify the virtual assistant’s response to the customer request to make a payment. Note that the possible responses leverage content, tool, and facilities packaged with the product.

ibm watson va illos

© 2017 IBM Corporation

Illustration 7. This Illustration shows the Watson Virtual Agent workspace for specifying responses from the bot/virtual assistant.

 

Which is the better approach? Consultant-assisted is our preference, but we’ve learned over our long years of research and consulting that deployment approach is a function of corporate, style, personality, and culture. Some businesses and organizations give consultants the responsibility for initial and ongoing technology deployments. Some businesses want to do it themselves. For virtual assistant software, corporate style could very well be a key factor in product selection.

 

 

 

 

Next IT Alme: Helping Customers Do All Their Work

On September 2, 2004, we published my article, “May I Help You?” It was a true story about my experience as a boy working in my dad’s paint and wallpaper store. The experience taught me all about customer service.

The critical lesson that I learned from my dad and from working in the store was customers want and need your help for every activity that they perform in doing business with you from their first contact with you through their retirement.

That help was answering customers’ questions and solving customers’ problems. That’s the usual way that we think of customer service, helping with exceptions, the times that customers can not do their work. But, that help was also performing “normal” activities on customers’ behalves—providing the right rollers, brushes, and solvents for the type of paint they wanted to use, for example, or collaborating with customers to perform normal activities together—selecting a paint color for trim or a wallpaper pattern.

At Kramer’s Paint, my dad or I delivered all of that help—normal work and exceptions work. In your business, you deliver the help to perform customers’ normal planning, shopping, buying, installing/using, and (account) management activities through the software of self-service web sites and/or mobile apps or through the live interactions of your call center agents, in-store associates, or field reps. And, you deliver the help for customers’ exception activities through customer self-service apps on the web, social networks, or mobile devices or through the live interactions of customer service staff in call centers, stores, and in the field.

Virtual Assistants Crossover to Perform Normal Activities

Recently, in our on customer service research, we’ve begun to see virtual assistant software apps crossover from helping customers not only with the exception activities to performing normal activities on customers’ behalves, activities like taking orders, completing applications, and managing accounts. We wrote about this crossover a bit in our last post about IBM Watson Engagement Advisor’s Dialog facility. And, we provided links to crossover examples of Creative Virtual V-Person at Chase Bank and Nuance Nina Mobile at Domino’s.

Alme, the virtual assistant software app from Spokane, WA based supplier Next IT, can crossover to help customers perform normal, too. In fact, Alme has always performed normal activities for customers. One of our first reports about virtual assistants, a report that we published on March 13, 2008, discussed Jenn, Alaska Airlines’ Alme-based virtual assistant. We asked Jenn to find a flight for us through this request, “BOS to Seattle departing December 24 returning January 1.” Jenn did a lot of work to perform this normal activity. Her response was fast, accurate, and complete. We asked Jenn again in our preparation for this post. “She” prepared the “Available Flights” page for us. Once again, her answer was fast, accurate, and complete. All that’s left to do is select the flights. The illustration below shows our request and Jenn’s response.

alaska airlines blog

Next IT Alme Provides Excellent Support for Normal Activities

Alme provides these excellent facilities for performing normal activities, facilities that are one of its key strengths and competitive differentiators:

  • Support for complex, multi-step interactions
  • Rules-based personalization
  • Integration with external applications
  • Let’s take a closer look at them.

Support for Complex, Multi-Step Interactions

For normal activities, complex, multi-step interactions help virtual assistants collect the information needed to complete an insurance or loan application, order a meal, or configure a mobile device and the telecommunications services to support it, for example. Alme supports complex, multi-step interactions with Directives and Goals.

Directives

Directives are hierarchical dialogs of prompt and response interactions between Alme virtual assistants and customers. They’re stored and managed in Alme’s knowledgebase and Alme provides tools for building and maintaining them. Directive’s dialogs begin when Alme’s processing of a customer’s request matches the request to one of the nodes in a Directive. The node presents its prompt to the customer as a text box into which the customer enters a text response or as a list of links from which the customer makes a selection. Alme then processes the text responses or the link selections. This processing moves the dialog:

  • To another node in the Directive
  • Out of the Directive
  • Into a different Directive.

That customers’ requests can enter, reenter, or leave Directives at any of their nodes is what makes Directives powerful, flexible, and very useful. Alme’s analysis and matching engine processes every customer request and response to Directive prompts the same way. When the request (re)triggers a Directive, Alme automatically (re)establishes the Directive’s context, including all previous text responses and link selections. For example, financial services companies might use Directives to implement retirement planning for their customers. The customer might leave the Directive to gather information from joint accounts at the bank with the customer’s spouse before returning to the Directive to continue the planning, opening, and funding of an Individual Retirement Account (IRA).

Goals

Goals let virtual assistants collect a list of information from customers through prompt and response interactions to help perform and personalize their activities. Virtual assistants store the elements of the list of information that the customer provides within virtual assistant’s session data for use anytime within a customer/virtual assistant session. Alme can also use its integration facilities to store elements of the list persistently in external apps.

Goals have the ability to respond to customers dynamically, based on the information the Goal has collected. For example, if the customer provides all of the Goal’s information in one interaction, then Goal is complete or fulfilled and the Alme virtual assistant can perform the activity that is driven by the information. However, if the customer provides, say, two of four required information items, then the Goal can change its responses and request the missing information, leading the customer through a conversation. Goals are created by authors or analysts who specify a list of variables to store the information to be collected and the actions to be taken when customers do not provide all the information in the list. In addition, Goals can be nested, improving their power and giving them flexibility as well as promoting their reuse.

Healthcare providers (Healthcare is one of Next IT’s target markets.) might use Goals to collect a list of information from patients prior to a first appointment. Retailers might use them to collect a set of preferences for a personal e-shopper virtual assistant.

Rules-Based Personalization

Personalization is essential for any application supporting customers’ normal activities. Why? Because personalization is the use of customer information—profile attributes, demographics, preferences, shopping histories, order histories, service contracts, and account data—to tailor a customer experience for individual customers. Performing activities on customers’ behalves requires some level of personalization.

For example, virtual assistants use a customer’s login credentials to access external apps that manage account or order data and, then, use that order data to help customers process a refund or a return. Or, to complete an auto insurance application, virtual assistants need profile data and demographic data to price a policy.

Alme’s rules-based personalization facilities are Variables, Response Conditions, and AppCalls. They are implemented within the knowledgebase items that contain the responses to customers’ requests.

  • Variables provide personalization and context. They contain profile data, external application data, and session data, for example.
  • Response Conditions are expressions (rules) on Variables. Response Conditions select responses and/or set data values of their Variables.
  • AppCalls (Application Calls) pass parameters to and execute external applications. They use Alme’s integration facilities to access external apps through JavaScript and Web Services APIs. For example, Jenn, Alaska Airlines’ virtual assistant, uses AppCalls to process information extracted from the customer’s question—departure city, arrival city, departure date and return date—and normalizes and formats the information for correct handling by the airlines’ booking engine. This AppCall checks city pairs to ensure the flight is valid and formats and normalizes dates so that the booking engine can display appropriate choices. AppCalls also integrate Alme with backend systems. Ann, Aetna’s virtual assistant, uses AppCalls to collect more than 80 profile variables from Aetna’s backend systems to facilitate performing tasks and to personalize answers for Aetna’s customers after they log in and launch Ann. (See the screen shot of Ann, below.)

Integration with External Applications

The resources that virtual assistant applications “own” are typically a knowledgebase (of answers and solutions to expected customers’ questions and problems) and accounts on Facebook and Twitter to enable members of these social networks to ask questions and report problems. So, to perform normal activities, virtual assistants need to integrate with the external apps that own the data and services that support those activities.

Alme integrates with external customer service applications through JavaScript (front end) and Web Services (back end) interfaces. New in Alme 2,2, the current Alme version, Next IT has introduced a re-architected Alme platform that is more modular and more extensible. The new platform has published JavaScript and Web Services interfaces to all Alme functionality and support for JavaScript and Web Services to external resources.

AppCalls use Alme’s integration facilities. To process an AppCall successfully, developers must have established a connection between Alme and an external application. Jenn integrates Alme with Alaska Airlines booking engine. Ann integrates Alme with Aetna’s backend systems. Here’s a screen shot.

aetna blog

Virtual Assistants Are Doing More of the Work of Live Agents

Next IT Alme was one of the first virtual assistant software products with the capabilities to perform normal activities. Its facilities are powerful and flexible. While integration with external applications will always require programming (and Next IT has simplified that programming), Alme’s facilities for supporting normal activities are built-in and designed for business analysts. They’re reasonably easy to learn, easy to use, and easy to manage.

By performing normal activities, virtual assistants are doing more of the work that live agents have been doing—quickly, accurately, consistently, and at a lower cost than live agents. That frees live agents to handle the stickiest, most complex customer requests, requests to perform normal activities and requests to answer questions and resolve problems. It’s also a driver for your organization to consider adding virtual assistants to your customer service customer experience portfolio.

The Dialog Feature of IBM Watson Engagement Advisor

We just updated our Product Evaluation Report on IBM Watson Engagement Advisor. It’s an update to our July 10, 2014 report. Both the scientists at IBM Research and the developers in the IBM Watson Group have been busy improving Watson and Watson Engagement Advisor, busy and productive enough to drive us to do the update. Here are the highlights:

  • Dialog is a new feature of Watson Engagement Advisor that provides facilities to support complex interactions between virtual assistants and customers. These interactions include prompt and response conversations as well as business processes, transactions, and supplementary questions. Dialogs can guide customers through the necessary steps to an outcome or help answer customers’ vague and ambiguous questions.
  • Knowledgebase additional input file support. MHT and ZIP files can be ingested into Watson’s knowledgebase, adding to HTML, Microsoft Word, and PDF file formats.
  • Watson Experience Manager is the visual toolset that subject matter experts use to configure, train, test, and administer Watson Engagement Advisor deployments. Improvements include new tools to configure Dialog conversations.
  • The Cognitive Value Assessment (CVA) is a consulting offering designed to help organizations identify use cases and benefits through examination of issues and pain points in their customer and end user business processes.
  • Product positioning. First contact self-service resolution

Dialog is the most important product improvement. Watson Group used technology from its May 2014 acquisition of Chatswood, New South Wales, AU virtual assistant software supplier Cognea to help build Dialog. The feature helps IBM catch up with its virtual assistant software competitors. The leading suppliers—Creative Virtual, IntelliResponse, Next IT, and Nuance—had all been offering Dialog-like capabilities for some time. Prompt and response conversations have become a key customer service requirement for virtual assistants. These conversations been the approach for answering vague or ambiguous customer questions, helping virtual assistants gather information from customers that concretizes or disambiguates questions (or problems) like, “I can’t get my printer to work,” or “What’s my balance?”

And, on that “What’s my balance?” example, prompt and response conversations are also an approach for supporting personalized tasks and transactions, a hot trend and an emerging requirement for virtual assistants to “act” more like live agents within their interactions with customers. Personalized tasks and transactions take a bit of application integration—therefore some programming—to access, retrieve, and use customer profile data and account data, but, once a virtual assistant has that data, it can perform a wide range of activities for customers in addition to delivering answers and solutions. Think of the Nuance Nina virtual assistant taking pizza orders for Domino’s (https://www.youtube.com/watch?v=noVzvBG0GD0) or Sara, a Creative Virtual V-Person virtual assistant helping consumers with online and mobile banking presenting account balances at Commercial Bank of Dubai (https://www.youtube.com/watch?v=rCvMJYQ0OT0).

Dialog is the mechanism that enables Watson Engagement Advisor virtual assistants to act more like live agents and web concierges. With that bit of application integration/ programming, they can perform personalized and/or transactional tasks for customers. Watson Engagement Advisor’s virtual assistants also have the advantages of Watson’s cognitive technology, which lets them have more flexible, more varied conversations. Watson can answer many types of questions, question that have answers like a simple fact, the definition of a term, the description of a topic, yes/no or true/false, and the steps in a procedure, or an approach to trouble shooting a problem. Dialog lets customers interject relevant but out of band questions within prompt and response conversations then, after the virtual assistant delivers the answer, either return to the Dialog or continue out of band interactions, perhaps, entering other Dialog flows. For example, in a customer/Watson Engagement Advisor virtual assistant session about property and casualty insurance, the customer might interject a question about extra coverage for jewelry in the middle of a Dialog implementing the application for a standard property policy. The virtual assistant can answer the jewelry question, answer any additional jewelry coverage questions, return to the policy application conversation, or even complete an application for a jewelry rider. For new customers, the policy application conversation will collect appropriate customer data and pass it to the external app. For existing customers, the virtual assistant will access the appropriate external app for the customer data. Live agents might do that data access manually from their desktops. Virtual assistants must do it with programming.

In the current Watson Engagement Advisor release, Dialog functionality and the tools are essentially what Cognea had built and offered. Watson Group’s developers are working hard to integrate the functionality more seamlessly within Watson Engagement Advisor and to integrate and improve the tools on the Watson Experience Manager toolset. We don’t think that Dialog is part of any of the five or so live Watson Engagement Advisor deployments, but going forward, we think that it will become part of most deployments. In fact, every virtual assistant should provide the capabilities to perform actions on behalf of customers.

Oracle Service Cloud Virtual Assistant

Oracle Service Cloud Virtual Assistant is a relatively new brand (November 2013) that dates back to 2001 when it was known as Q-go, a product with the same name as its privately held, Amsterdam, NE-based and supplier. Q-go, the product, was first commercially deployed in 2001.

Neither RightNow, which acquired Q-go in 2011, nor Oracle, which acquired RightNow in 2013, has done very much to enhance the old Q-go other than to rebrand it. (RightNow branded it RightNow Intent Guide and Natural Language Search at the time of the 2011 Q-go acquisition. As we mentioned just above, Oracle branded it Virtual Assistant in November 2013.) Today, Oracle Service Cloud Virtual Assistant doesn’t support voice. It deploys on web browsers in HTML. Its deployments are language-specific and only Western European languages are supported.

However even after all this time, the product technology remains reasonably attractive and quite useful. The product’s core NLP technology, the technology developed by Q-go that analyzes customers’ questions and matches them with knowledgebase answers remains fresh and innovative. This is the technology that makes virtual assistants that are deployed on Virtual Assistant powerful and flexible.

Briefly, here’s how it (still) works. Customers enter questions as natural language sentences or phrases or as individual keywords or short “telegram-style” phrases. Virtual Assistant uses NLP technology to analyze customers’ questions and to match them with Questions in its knowledgebase.

  • If Virtual Assistant finds a Question that is a best match with a customer’s question, then it presents the knowledgebase Answer that is linked to the Question.
  • If Virtual Assistant cannot find a best match, it presents a short list of Questions that are likely matches to customer’s question. From this list, the customer selects the Question that best represent her/his question. Virtual Assistant delivers the Answer associated with the Question selected by the customer.
  • Alternatively, the matching Question may trigger a Prompt and Response Dialog with the customer to arrive at an Answer through conditional sequence of steps.
  • Virtual Assistant presents the knowledgebase Answers that are linked to the Questions that best match the customers’ questions.

Here’s an example from klm.com. KLM Royal Dutch Airlines had deployed Q-go. The Customer Support tab on klm.com continues to use the NLP technology to answer customers’ questions.

I asked the question, “golf clubs,” two keywords, certainly not a full sentence or even a good phrase. Virtual Assistant did not find a single best match. It found five likely matches and presented them as links as shown in the screen shot below.

blog klm 1

Question 3, “Can I take a golf bag with me,” represents our question. I clicked it and klm.com presented the Answer shown in the screen shot below.

blog klm 2

That’s exactly the information I was looking for, a “Very useful” Answer. I then clicked the “Read more about taking a golf bag” link to get additional information. See the screen shot below for that information.

blog klm 3

Pretty good, no? Flexible and powerful. When Virtual Assistant can’t find a best match, finding and presenting a short list of likely, possible matches is very useful. It’s a reasonable and fast extra step to take to get to that best match. Note though that this example was not implemented by a completely automatic process. No surprise there. Analysts and administrators had some work to do to specify and manage the keywords that customers would likely use and to associate those keywords with knowledgebase Questions, like those in the first screen shot,  to help Virtual Assistant find likely matches. They also had to specify and manage the Questions.

Oracle certainly has some work to do to bring Virtual Assistant up to its competition, but the work builds on a very good foundation. Read our Product Evaluation Report to get the details.