A buyer’s guide to enterprise chatbots: The 6 levels of intelligent virtual agents

 In Enterprise Chatbots, IT Service Management Blog

Much has been written about the rapid growth of the chatbot market. Gartner reports that while 38 percent of enterprises are planning or actively experimenting with conversational agents, today only 4 percent have deployed enterprise chatbots in their organizations. Within the customer support sector specifically, the use of chatbots for customer support and assistance is skyrocketing. Gartner forecasts that the use of virtual customer assistants (VCAs) will jump by 1,000 percent by 2020.

And as more players enter the space offering AI chatbots and enterprise chatbot platforms, it’s becoming more difficult for enterprises buyers to differentiate the offerings and separate the truth from the hype.

Natural language processing

Chat Automation by Mat fine from the Noun Project

Natural language processing (NLP) is the underlying technology powering enterprise chatbot development. NLP is a field of artificial intelligence focused on processing human language. Whereas any text processing can be categorized as NLP, NLU entails true “understanding” of language and is a significantly more challenging problem. NLU involves commonsense reasoning, or human-level competency of language.

NLP technology is applied in varying levels of complexity to a whole host of enterprise chatbot applications, spanning customer service management (CSM), employee service (including IT service management, HR case management and facilities management), recruiting and many, many more. Within the employee service sector, these enterprise chatbots are often referred to as virtual agents or virtual support agents.

Key factors to consider when choosing a virtual agent for employee service

Price List by Vectors Market from the Noun Project

Choosing a virtual agent for employee service without a proper assessment of its true capabilities is risky. The wrong virtual agent could end up being costly, time-consuming and ultimately counterproductive to your business goals. These unforeseen costs include, but are not limited to:  

  •      Numerous hours spent manually scripting highly rigid support dialogs
  •      Expensive professional services to train and manage the virtual agent
  •      Sub-par employee experiences due to misaligned expectations with your enterprise chatbot’s performance and conversational ability
  •      Flaws in data security, compliance and controls

Ultimately, virtual agents that do not possess true language processing technologies will have an ongoing, impractical and unscalable need for massive amounts of data to train the virtual agent.

That’s why our Principal AI Scientist, Walid Saba, one of the world’s foremost experts on natural language processing and natural language understanding, has developed a framework to help enterprises cut through the noise. Walid’s levels of intelligent virtual agents clearly describes enterprise chatbot technology and its associated conversational capabilities within six defined levels.

What follows is a sneak peek at the six levels of intelligent virtual agents. Level 0 is the most basic of the six levels, and with each additional level, the conversational intelligence – as well as the underlying technology – becomes increasingly sophisticated. Watch our on-demand webcast to learn more!

The 6 levels of intelligent virtual agents

Level 0 Scripting

Scripting and Pattern Matching Virtual Agent

Level 0 is the most rudimentary of all of the virtual agents. Level 0 virtual agents can perform pattern matching tasks using regular expressions . Many Level 0 virtual agents can be easily built using readily available NLP libraries/frameworks. These libraries, many of which are open-source, provide the chatbot training data needed to create a Level 0 virtual agent.

But be warned, Level 0 virtual agents, are in essence, skeletons. They have the foundation for conversational capabilities, but require someone to create dialogs of phrases, or scripts, to properly function. Building a Level 0 virtual agent takes time and patience. In order for the virtual agent to be useful in answering requests, enterprises have to manually script out countless employee question and answer scenarios.

If you’re using a Level 0 enterprise chatbot platform to scale your service desk, it can match the text of an employee’s request, e.g. “What is the PTO policy” with text in other documents that have the matching phrase, e.g. a knowledge base article entitled “PTO policy.” And while it can match phrases, a Level 0 won’t be able to do much if one of your employees asked “I need to know more about the rules and procedures concerning paid time off.”

 

Level 1 Basic

Basic Search and Routing Virtual Agent

A Level 1 virtual agent builds upon the pattern-matching capabilities of a Level 0 virtual agent with search capabilities. It can take a keyword or phrase, e.g. “PTO policy” and perform a basic search. It can also understand the basic intent of an employee’s query, e.g. a question about “PTO policy” is related to “paid time off.” A Level 1 virtual agent possesses a rich dictionary of relevant words and phrases, and can engage in basic small talk (greetings, starting-over, etc.).

But what really separates a Level 0 from a Level 1 virtual agent? A Level 0 virtual agent is tied to a script. It matches keywords. A virtual agent that is Level 1 Basic, on the other hand, uses some language processing to identify more than just matching keywords. It can perform very basic thesaurus-driven matching. For example, it can relate terms like “computer” and laptop” and it can process search queries using plain natural language rather than the canned keyword phrases a Level 0 needs to function.

 

Level 2 Contextual

Contextual Understanding and Domain Independence Virtual Agent

A Level 2 virtual agent should be able to understand the subject-matter or domain of a request, and can beyond the very basic Level 1 ontological matching, to be able to understand that multiple employee requests about PTO policy can be grouped together based on the vocabulary and sentence structure employees use. A Level 2 can understand relevance.  It knows how to differentiate between questions it can and can’t answer, and can escalate a query if it knows it cannot resolve it.

For example, a Level 2 HR support virtual agent knows it cannot answer a question about the weather forecast for next week. It doesn’t misinterpret a phrase like “paid time off” to mean that an employee paying something or someone called “time.” It also has the ability to learn new domains quickly. A Level 2 virtual agent that is used for HR case management resolution, for example, can also be used for IT service management without first having it train on a huge amount of IT data. Finally, a Level 2 has the ability to conceptually link topics that are thematically related. As a milestone, therefore, Level 2 moves from keyword and basic sentence structure matching capabilities to more advanced semantic and conceptual understanding. It knows, for example, that knowledge base articles about “sick leave” are related to “PTO policy.”

 

Level 3 Adept

Advanced Dialogue Virtual Agent

A Level 3 virtual agent can perform all of the tasks a Level 0-2 virtual agent can perform, but it also possesses the ability to engage in a natural dialogue. A Level 3 can ask clarifying questions. Clarification requires deep semantic understanding, It requires resolving references previously mentioned in the discourse and filling in the gap based on the previous part of the conversation. In short, a Level 3 can do what computational linguists call deep discourse analysis.

A Level 3 virtual agent can understand the larger context of a request including whether the virtual agent has engaged with the person in the past, and if it has, can connect words or phrases that refer to past conversations, e.g. an employee responding to a virtual agent’s request for an employee ID number by replying “I have it right here.” Finally, a Level 3 virtual agent knows when there are equally plausible ways to solve the resolution. A Level 3 may find out there seems to be more than one answer to a question, that each of the options lead to equally plausible resolutions. It may, for example, clarify an employee’s PTO policy question by asking whether they are referring to sick leave, paid leave, or personal leave. This kind of intelligent clarification with the employee reduces mean time to resolution.

 

Level 4 Resolution

Learning, Personalization, and Problem Resolution Virtual Agent

A Level 4 virtual agent embodies all of the intelligence of a Level 3 but it can also learn over time and can proactively engage with employees in a personalized way to understand, analyze and actually resolve issues. Compared to previous levels, a Level 4 doesn’t just point an employee to a knowledge base article to solve an issue. It is an intelligent virtual agent that actually solves the issue, issues that may have otherwise required high-touch involvement from someone within employee support. Level 4 virtual agents learn from past conversations and use that knowledge to develop shortcuts to efficiently resolve issues in a personalized way. A Level 4 Resolution virtual agent, for example, would not simply point an employee to a PTO form to fill out. It would be able to solve the problem on the spot without requiring the employee to take that extra step.   

Level 5 Fluent

Human-level Conversational Virtual Agent

While Level 4 virtual agents have ability to resolve issues, it cannot yet resolve linguistic issues that require high-level, or human-level reasoning. A Level 5 virtual agent, on the other hand, can engage in human-like conversations. It is now an agent that fully understands and comprehends what an employee is asking. It learns an employee’s specific situation (using its understanding of previous interactions with that employee or employee group), and can resolve issues or escalate and route to the proper channel. This level of understanding requires very advanced semantic and pragmatic capabilities. A Level 5 virtual agent can understand compound nominals, metaphors, idioms, scope resolution; metonymy, co-predication, anaphor resolution. It reasons through requests using common sense.

This is just a preview of the content in our on-demand webcast: The 6 levels of intelligent virtual agents: a buyer’s guide to enterprise chatbots.  Watch the webcast to hear Walid Saba, Naghi Prasad and Astound’s Senior Director of Product Marketing, Rob Young, share in-depth examples of each level, as well as our predictions for the levels that haven’t yet been achieved by virtual agents on the market today.

Watch on-demand now!

 

Recent Posts
Contact Us

Please send us an email and we'll get back to you, asap.

Not readable? Change text. captcha txt

Start typing and press Enter to search