What is a virtual agent?
Explore IBM's virtual agent solution Subscribe for AI updates
Drawing of a person next to chat message bubbles
What is a virtual agent?

A virtual agent combines natural language processing, intelligent search and robotic process automation (RPA) in a conversational UI–typically a chatbot.

 

What is virtual agent technology?

Virtual agent technology (VAT) is the combination of natural language processing, intelligent search and robotic process automation (RPA) in a single conversational user interface—typically a chatbot—to automate dialogue with end users, provide information, and directly execute actions to meet user requests.

Leading virtual agent solutions represent an evolutionary leap in the utility of chatbots. Recent advancements in conversational AI, applied alongside speech-to-text, optical character recognition (OCR) and sentiment analysis, allow virtual agents to interpret open-ended user input and accurately identify the user’s specific goal, or “intent.” When integrated into relevant backend systems, like a CRM platform or billing infrastructure, virtual agents can often automate actions to achieve that intent without any further human intervention.

While some of the machine learning and automation techniques powering virtual agents have been around for years, it is the assembly of those constituent parts into a single self-sufficient system that drives the versatility and productivity of VAT.

The data store for AI

Discover the power of integrating a data lakehouse strategy into your data architecture, including enhancements to scale AI and cost optimization opportunities.

Related content

Register for the Gartner report

Virtual agent vs. chatbot vs. virtual assistant

There is some overlap between the uses and features of virtual agents, chatbots and virtual assistants, and the lack of formalized definitions for these and similar terms results in them sometimes being used interchangeably.

Despite this ambiguity, there is a general consensus on the technological distinctions that separate each of these related tools.

Chatbot is a catch-all term for a program that simulates real-time human conversation with the user. Chatbots operated through vocal prompts rather than written prompts are known as interactive voice response (IVR) systems. Typically, a chatbot manages customer interactions with a decision tree flow. This does not necessarily involve artificial intelligence; many chatbots rely on a pre-programmed set of inputs that can be recognized, each triggering a corresponding pre-scripted answer. Rudimentary chatbots, unable to parse input that does not exactly match what they have been programmed to recognize, require the user to select from a series of simple pre-written options rather than author (or speak) inputs in their own words.

While most virtual agent technology involves a chatbot that receives and responds to requests, not all chatbots offer true VAT functionality. Many chatbots and IVR systems serve only to provide or gather basic information, like relaying store hours or determining where to route a customer at a call center.

Virtual assistant primarily refers not to software, but to a human being providing remote—that is, virtual—assistance. Somewhat confusingly, “virtual assistant” (or “virtual assistant software”) is also occasionally used as an umbrella term for all virtual products that provide assistance, including services like Apple’s Siri or Amazon’s Alexa (which are also dubbed voice assistants or digital assistants).

Virtual agents, also known as intelligent virtual agents (IVAs) or intelligent virtual assistants (also IVAs), are more than just highly sophisticated chatbots. Virtual agents are defined by not only conversational AI that can identify the intent of freeform text or speech from users, but also the automation of steps to meet that intent—and continuously improve its ability to do both. Whereas a chatbot can only respond, a virtual agent can understand, learn and do.

Voice assistants like Siri or Alexa could be considered virtual agents by this definition, but the term “virtual agent” more commonly refers to organizational use and customized integration with enterprise systems. Put another way, voice assistants usually act as an extension of yourself: they automate actions you would do, like send a text, search for a public info or play a song. Virtual agents are an extension of your business: they automate actions for customers or employees, like paying a bill or updating log-in credentials.

Types of virtual agents

Enterprises looking to optimize their business with virtual agent technology have a wide array of options to choose from. Each offers varying degrees of customization and integration and each requires varying degrees of labor and sophistication to implement and maintain. The best virtual agent solution for your business is a function of the specific needs that the VAT will address and the resources available to obtain and manage it.

  • End-to-end solutions: full-service offerings replete with professional assistance from providers to manage implementation, maintenance, and integration into relevant applications, systems and workflows.
  • Scalable pro development tools: API-accessible platforms best suited to organizations with dedicated technical resources and developers to directly manage complex implementations.
  • Low code and no code SaaS: flexible VAT solutions designed to be easily built and managed without specialized technical expertise.
  • Integrated solutions: complementary capabilities directly integrated into proprietary enterprise tools, such as a chatbot built into contact center software.
How to create a virtual agent

A thorough understanding of your customer journey is essential to making a virtual agent work effectively. By properly identifying customer intent and the steps necessary to achieve it, you can configure your virtual agent to naturally map to those steps.

1. Determine scope: what problems or opportunities will the virtual agent address?

Focusing on the wrong goals will limit the potential of your virtual agent, as will an overly broad or restrictively narrow scope. Which repetitive issues, questions and tasks consume too much customer support bandwidth? What employee needs aren’t adequately met? Which divisions of your business would most benefit from cost savings or time savings? The FAQ section of your website is often a helpful knowledge base to begin with.

Learn more about identifying opportunities for improvement
2. Determine messaging channels: where will users talk to your virtual agent?

Does the virtual agent address a common issue with your website or app? Is it meant to reduce phone call volume at your contact center? Does it answer questions often asked on Slack? The channels where end users interact with your virtual agent should naturally align with the intents it serves. Messaging channels may also affect how user intents are expressed, how your virtual agent’s conversational AI should interpret and respond to them, and which relevant tools and systems are available for integration.

Learn more about channel planning for virtual agents
3. Train a conversational AI model to intelligently interpret and respond to requests

Your virtual agent must be able to accurately interpret customer queries and recognize user intent. To do so without restricting the user to menu-based selections requires sophisticated natural language understanding: actual human users will rarely articulate their goals in the exact words—or even the exact spelling—that you predict. Leading chatbots then increasingly use generative AI, trained to appropriately represent your brand, to keep the conversation moving.

Read the blog: “How IBM is tailoring generative AI for enterprises”
4. Escalate out-of-scope intents to a live agent

Your virtual agent shouldn’t—and can’t—be made to handle every single request possible. It’s better to provide high-quality solutions to a smaller set of problems than inadequately address a larger list of intents. When a user’s intent is out of scope, connect the user to the appropriate human to provide assistance.

Learn more about "planned" and "fallback" escalations
5. Integrate the systems necessary to address in-scope intents

Each in-scope intent should be mapped to the tools and processes needed to achieve it. For achieving information requests, that means intelligent search connected to any relevant data sources. Action-based intents requiring robotic process automation (RPA) may entail integration with systems like CRM platforms, payments infrastructure, scheduling software, or IT self-service portals.

Explore chatbot integrations and extensions for watsonx Assistant
6. Continuously improve your virtual agent

Once your virtual agent is live and generating result data, continue to refine and improve it. These improvements can be technologically-driven—for example, with machine learning that evolves your AI’s ability to identify intent—or strategically-driven, by evaluating underserved intents, broken flows or opportunities to grow your virtual agent’s scope. 

Read tips for building, deploying and improving your virtual agent
Benefits of virtual agent technology

Successfully implemented VAT has been proven to have a positive impact on your business’s finances, logistics, and employee morale.

  • Increased customer satisfaction: In a survey of 1,005 respondents across 12 industries and 33 countries conducted by the IBM Institute for Business Value (IBV) in cooperation with Oxford Economics, 99% of organizations using AI-based virtual agent technology reported an increase in customer satisfaction. On average, said organizations benefited from an 8 percentage-point improvement in customer satisfaction and 4 point improvement in NPS.
  • Time savings for employees: By diverting the most repetitive, time-consuming tasks to virtual agents, VAT often improves human agent efficiency. The aforementioned IBV/Oxford study found that VAT reduced human agent handle time by an average of 12%.
  • Reduced costs: Reducing the time it takes for human agents to resolve contacts has the added benefit of reducing cost to serve. A recent Forrester Consulting study estimated that a large organization1 could save an average of USD 6.00 per contained conversation using IBM watsonx Assistant™. The same study found that virtual agents correctly routing phone conversations saves USD 7.75 per correctly routed call.
  • Improved employee satisfaction: Customer service agents equipped with proper tools and support are more likely to feel valued by their organization. They’re also more likely to deliver a better experience to customers. Improved employee morale also aids employee retention, which has its own financial impact: Gallup research (link resides outside ibm.com) found that the cost replacing an individual employee can range from 50–200% of the departing employee’s annual salary.
Use cases for virtual agent technology

Virtual agents offer a variety of opportunities to improve customer experience, market-facing business operations and internal productivity and coordination.

  • Customer service: Virtual agents, in the form of text-driven chatbots or call-based interactive voice response (IVR) systems, are often employed as automated customer service representatives across a variety of channels, from owned-and-operated websites to social media platforms to messaging platforms like Slack and WhatsApp.
  • E-commerce and sales: Virtual agents can be deployed to enhance sales funnels and lead generation by qualifying leads and completing transactions in diverse retail environments.
  • Workforce productivity: Virtual agents can help increase employee productivity by automating rote tasks and inquiries, freeing employee time for more complex tasks. They can also aid in streamlining collaboration, workflow and project management through scheduling automation, managing and transcribing stand-up meanings, and enhancing the capabilities of workplace communication tools like Slack and Microsoft Teams.
Key performance metrics for virtual agents

While a truly comprehensive evaluation of how effectively virtual agent technology has been implemented will depend on the specific challenges and objectives of your business, here are three key measures of how well VAT is meeting performance expectations.

  • Intent recognition: How accurately does your virtual agent interpret user intent? Users often articulate their needs in unique ways, from word choice to syntax to spelling. A customer might ask, “how do I settle my account balance?” when your virtual agent is programmed for “pay my bill.” Properly handling natural variance in the way users express intent is a critical component of successful implementation, and is often a function of natural language processing (NLP) capabilities.
  • In-scope segment: What percentage of incoming user requests—assuming their intent has been accurately identified—match the intents your virtual agent is programmed for? If most requests align with what your VAT was programmed to handle, it’s well calibrated to your users’ needs. If most requests are falling out of scope, you may need to re-evaluate your VAT strategy. The IBV/Oxford Economics survey found that, across all respondents, the average proportion of inbound contacts that fall within the scope of the VAT was 63%.
  • Containment: How often is the VAT able to successfully resolve a given case without any escalation to—or involvement from—a human agent? This is a nuanced metric: the denominator may or may not include requests with out-of-scope intents; the numerator may or may not count cases with multiple intents in which some but not all intents are successfully contained. The IBV/Oxford Economics study, which defined containment as “the portion of total contacts the VAT has been trained to handle” that are resolved without escalation, calculated average containment across relevant respondents to be 64%. A difference of 38% separated the highest and lowest reported containment figures.
Related solutions
IBM watsonx Assistant

Deliver consistent and intelligent customer care across all channels and touchpoints with conversational AI.

Explore IBM watsonx Assistant

IBM Watson Discovery

Find critical answers and insights from your business data using AI-powered enterprise search technology.

 

Explore IBM Watson Discovery

AI chatbots

Meet a natural language AI chatbot that understands human conversation and improves the customer experience.

Explore AI chatbots
Resources Unlock AI success

Harness pretrained models, embrace omnichannel operations and avoid common chatbot pitfalls for effective virtual assistants.

Take the next step

IBM watsonx Assistant helps organizations provide better customer experiences with an AI chatbot that understands the language of the business, connects to existing customer care systems, and deploys anywhere with enterprise security and scalability. watsonx Assistant automates repetitive tasks and uses machine learning to resolve customer support issues quickly and efficiently.

Explore watsonx Assistant Book a live demo
Footnotes

1 For the purposes of the study, Forrester aggregated the interviewees’ and survey respondents’ experiences and combined the results into a single composite organization that is a financial and insurance services company with revenue of USD 7 billion per year.