Conversational agent

From Wikipedia, the free encyclopedia

Conversational agents (CAs) are communication technologies that utilize natural language and computational linguistic techniques to engage users in human-like, Web-based “dialogs.” They can support a broad range of applications in business enterprises, education, government, healthcare, and entertainment.^[1] For example:

• Customer self-service: Responding to customers' questions about products and services via a company’s Website or Intranet portal
• Customer service agent knowledge base: Allows agents to type in a customer’s question and guide them with a response
• Guided selling: Facilitating transactions by providing answers and guidance in the sales process, particularly for complex products being sold to novice customers
• Help desk: Responding to internal employee questions, e.g., responding to HR questions
• Website navigation: Guiding customers to relevant portions of complex websites --a Website concierge
• Technical support: Responding to technical problems, such as diagnosing a problem with a product or device
• Personalized service: Conversational agents can leverage internal and external databases to personalize interactions, such as answering questions about account balances, providing portfolio information, delivering frequent flier or membership information, for example
• Training or education: They can provide problem-solving advice while the user learns

Contents 1 Importance and Impact 2 Technical Challenges and Requirements 3 Example Conversation 4 Notes and references

[edit] Importance and Impact

Conversational agents are playing an increasingly important and prominent role in enterprise software, particularly around customer service and support. Two key areas are improving customer satisfaction and experience, and reducing costs. Conversational agents offer a solution to the cost versus effectiveness tradeoff.

Customer Experience, Satisfaction, and Retention
Improving customer service and support is essential to many companies because the cost of failure is high: loss of customers and loss of revenue. Yet today, while customer expectations are higher than ever, the costs of delivering service are high and the quality is low.

Two-way communication is critical for acquiring, servicing and retaining customers. Not only must companies educate their customers about their products and services, but also they must gain a clear understanding of their customers’ needs in order to satisfy and retain them. Deployed properly, a conversational agent can help to provide companies these capabilities.

Moreover, customers seek answers to their inquiries that are accurate and timely. Unfortunately, with many human-assisted and self-service interactions, this isn’t a reality. Customers and users are frustrated by fruitless searches through websites, long hold times to speak with customer service representatives, and delays of several days for email responses. Their experience gets worse with poor self-service implementations. These experiences often drive customers back to the contact center for “live” assistance, or drive them to a competitor. This can prove much more costly than companies realize.^[2]

Cost Reduction
While quality human-assisted support is highly effective, it is also the most expensive. Plagued by mounting labor and training costs, the customer service function is a key area where companies are seeking to reduce costs and improve effectiveness. Automated, self-service solutions have the potential to significantly reduce these costs.

Consequently, many companies have responded to this need by encouraging customers’ use of interactive voice response (IVR) systems and Web-based search and FAQs. While these solutions have become “mature” technologies, studies indicate that the poor service experience resulting from these solutions’ rather limited capabilities has had a negative impact on overall customer satisfaction.^[3]

By capitalizing on enabling technologies of the Web and computational linguistics to assist customers/users through automated dialogs, customer-facing conversational agents offer companies the ability to provide customer service much more economically than with traditional, “live agent” models, while delivering a much more positive user experience.

Moreover, internal-facing conversational agents can act as a knowledge base and enable contact center/customer service agents to efficiently and accurately answering and addressing customer questions.^[4]

Because conversations and dialogs are ingrained in our behavior, a conversational agent can be user-friendly compared to technologies that cannot understand words and phrases. The benefits of good usability can affect the top line and the bottom line of the business value equation in terms of increasing revenue, decreasing costs and maximizing earnings.^[5]

[edit] Technical Challenges and Requirements

Lester, Mott and Branting, several of the key researchers and creators behind the patent-pending RealDialog^{TM [6]} Conversational agent technology, outlined the following technology challenges and requirements for conversational agents:

Conversational agents must satisfy two sets of requirements. First, they must provide sufficient language processing capabilities that they can engage in productive conversations with users. They must be able to understand users' questions and statements, employ effective dialog management techniques, and accurately respond at each conversational turn. Without a robust language processing facility, agents cannot achieve accuracy rates necessary to meet the business objectives of an organization.

Second, they must operate effectively in the enterprise or they cannot be used in large deployments. They must be scalable and reliable, and they must integrate cleanly into existing business processes and enterprise infrastructure.

Natural Language Requirements
Accurate and efficient natural language processing is essential for an effective conversational agent. CAs that leverage Conversation Natural Language Processing (CNLP) can effectively emulate the rich give-and-take of human conversations. Unlike pattern matching technologies that just look for keywords, CNLP performs an actual “linguistic breakdown” of the information entered by the user.

Each piece of a customer utterance, which can be a statement or question, is broken apart into its components and identified based on its syntactic and semantic qualities. This allows not only the words themselves to be understood, but also the intent of those words to be realized.

By taking advantage of highly effective parsing, semantic analysis, and dialog management technologies, conversational agents can clearly communicate with users to provide timely information that helps them solve their problems. While a given agent cannot hold conversations about arbitrary subjects, it can nevertheless engage in productive dialogs about a specific company's products and services.

Enterprise Deliverability Requirements
Conversational agents can be introduced into the enterprise provided they meet the needs of a large organization. To do so, they must be able to enter into dialogs with thousands of customers/users on a large scale. They must be scalable, provide high throughput, and guarantee reliability.

They must also offer levels of security commensurate with the conversational subject matter, integrate well with the existing enterprise infrastructure, and provide a suite of content creation and maintenance tools that enable the enterprise to efficiently author and maintain the domain knowledge, and support a broad range of analytics with third-party business intelligence and reporting tools.^[7]

[edit] Example Conversation

Here is an example of such a “conversation” between a conversational agent and an end user. For illustration purposes, a graphical depiction of this scenario can be found on the RealDialog product's conversational agent Web page^[8].

The conversation begins with the user utterance:

“I would like to buy it now”

The agent must first determine the literal meaning of the utterance: the user wants to purchase something--probably something mentioned earlier in the conversation. In addition, the agent must infer the goals or intent of the user. Although the user's utterance is in the form of a statement, it was probably intended to express a request to complete a purchase.

Once the agent has interpreted the statement, it must determine how to act. The appropriate actions depend on the current goal of the agent (selling products or handling complaints, for example), the dialog history (the previous statements made by the agent and user), and information in databases accessible to the agent, such as data about particular customers or products.

For example, if the agent is ‘designed’ for selling products, the previous discussion identified a particular consumer item for sale at the agent's website, and the product catalog shows the item to be in stock, the appropriate, the agent might take action by presenting an order form and ask the user to complete it.

If instead the previous discussion hadn't clearly identified an item, the appropriate action might be to elicit a description of a specific item from the user. Similarly, if the item were unavailable, the appropriate action might be to offer the user a different choice. Finally, the agent must respond with appropriate actions, which might include making a statement, presenting information in multimedia formats, or logging information to a database.

For example, if the appropriate action were to present an order form to the user and ask the user to complete it, the agent would need to retrieve or create a statement such as “Great! Please fill out the form below to complete your purchase”, then create or retrieve a suitable Web page, display the text and Web page on the user's browser, and log the information.

[edit] Notes and references