August 2004

By David R. Butcher, Assistant Editor, Customer Inter@ction Solutions'

A section for FAQs (frequently asked questions) is novel, beneficial and efficacious for Web sites' customer self-service ' if it were still 1993. But at present, we habitually ask questions and expect them to be clearly, distinctively and promptly answered. So why not allow individual site-, application- or company-related questions to be immediately answered on a specific and individual basis?

Conversagent (called Active Buddy until eight months ago), a conversational software solutions provider based in New York City and Sunnyvale, CA, enables companies to deploy natural language applications to solve problems that occur in businesses. Targeting the automated customer service market, it generates resolve as represented in improvement of customer service, rise of employees' productivity and increase of marketing effectiveness.

Employing its new 3.0 version of the Automated Service Agent (ASA) System, Conversagent users type questions in their own words using a text-chat interface and receive immediate and accurate responses to their natural language inquiries. The interactive agents are software applications (often referred to as 'bots' or 'robots') that interact with users. Conversagent's new service uses a single ASA deployed simultaneously over multiple communication channels, including any interface that supports two-way messaging, such as the Web, instant messaging services and wireless. ASA does not require client software to be downloaded for interaction.

Whereas traditional online self-service has companies attempt to discern customers' intentions by analyzing a click stream or other rudimentary measure, companies that adopt Conversagent can capture customers' actual intentions by exercising the ASA. ASA is Web-based and secure reporting and analysis tools allow companies to view customers' most frequently asked questions, among many other items. Companies can also track detailed traffic information.

Coversagent Server will integrate with existing CRM systems, running on Linux (Red Hat version 7.3 or greater) or Windows 2000/XP/Windows Server 2003, and its technology smoothly complements and integrates with 'full-circle' support solutions. ASA systems can be installed on the company's premises or employed as a managed service. In both cases, they can be integrated with back-end systems to access customer-specific information or to resolve more complex issues involving external data. Its scripting language provides means for authenticating customer identities and delivering accessed information in a unified, conversational way to customer-initiated queries.

The genesis of this interactive agent betided as a demo called Smarterchild, which Conversagent still uses as the foundational basis, and was released via Instant Messenger (IM) in 2000, adapting to shorthand text and 'IM language,' by which words are not necessarily spelled out but understood through the use of symbols and cryptic, receded language (e.g., how many emails im allowed 2 keep?). Because Conversagent's origins manifested through IM, it is not built of a lexicon or a painstakingly precise arrangement of words (i.e., a stock of terms particularly used).

This, therefore, allows users to ask questions without having to ask in a specific tone or style, or with a meticulously chosen phrase the company has designated. Whereas FAQs and search techniques typically are predetermined and require customers to describe issues in words familiar to the company, Conversagent allows customers to use vocabulary and phraseology familiar to them. In other words, users ask questions in their own words, for accurate answers translated to their personal speech. It can also acclimate to any Western language, and can be customized to other languages for the company by Conversagent.

Unlike other virtual agent technologies of a similar sort (Ask Jeeves, etc.), Conversagent compasses its mission in responding to questions with specific answers, rather than with links to possible answers. It does so with immediacy, as opposed to with the delay that can come when requesting information through e-mail or possible lengthy holding on the phone. The immediacy of this ASA session provides a level of personalization generally associated with live agent help. It does so at a significantly lower cost, as well.

Often, the user finds irrelevance in the search engine's list of results to a question posed, and the customer then escalates to a live agent customer service channel. As well as invoke frustration on the customers' part, this dilemma also drives up costs to the company. However, because interactive agents are programmed to answer inquiries rapidly and accurately, higher-cost support channels are able to handle more complex transactions, allowing for both financial and time savings for the company that employs the software. Freeing the support professionals so they may focus on higher-value customer cases is one of the attempts to minimize costs, and hitherto, enables Conversagent to cost 20 cents (average) per ASA session versus the four-plus dollars per session for attended service.

The conversational application paid for itself in less than a month at Comcast, the first company to use Conversagent's product, according to Conversagent CEO Stephen Klein, although he was unable to offer numbers to Customer Inter@ction Solutions' about the implementation with Comcast. Comcast, a cable and broadband communications provider, employs Conversagent ASA to apply the software to a tool called Ask Comcast, which allows customers to ask questions in the aforementioned conversational language (although Comcast does still offer a FAQs directory, as well as a search engine). It also provides a link to live chat so that if it fails to answer a question, customers are immediately elevated to live support.

If the customer does require intervention of a live support professional, the ASA can seamlessly escalate the customer, along with detailed information regarding the inquiry. It can also recognize customer statements that indicate frustration, automatically escalating particularly bristly questions. While many FAQ virtual agents attract off-topic questions or abhorrent statements due to user chagrin, Conversagent has the ability to retort with professional diplomacy or humorous reciprocation, depending upon the company's marketing goals and audience. For instance:

Off-topic/frustrated customer question: 'Do you dress in ladies' muumuus and bathe in Spanish rice?'

Possible response 1: 'That's mean. Apologize or your original question will not be answered.' (Referring, of course, to the inquiry prior to the inappropriate inquest.)

Possible response 2: 'I don't understand the question, and you will now be connected to a live agent.'

However, one quirk occurred when I engaged in an arbitrary conversation with 'Chat Bot,' the other side of conversation of the chat demo on Conversagent's Web site. Chat Bot asked what type of food I liked (after I inanely asked it what snack food it liked ' I did say arbitrary), and I replied 'animal crackers' (I prefer frosted over plain or iced). To this, the ASA inquired (and I am not making this up): 'They r people. What are your feelings about white people?'

What?

Klein's response to the anomaly was that it was due to the company's 'processing of the word 'cracker' in response to people 'teaching' Smarterchild that 'cracker' means white person.' When SmarterChild encounters usages that it's unfamiliar with, it often asks users what a word means. Or if SmarterChild has some knowledge about a word or topic, it asks users for further information on the subject (as it did here, asking my opinion about white people because of the word 'cracker'). Conversagent's Web site uses the chat domains from SmarterChild and features much of what SmarterChild has learned from its users.

'We score these 'learnings' very high, as they represent topics that are highly used by SmarterChild's audience, and we like to reveal the technology's brains-that-learn more than its sometimes boring existing knowledge,' Klein continued.

In its IM logs ' handling about 250,000 IM sessions a day ' there occurs objectionable language. SmarterChild has reporting features that alert the company to new knowledge that the program incorporates into its speech. The knowledge about 'crackers' that SmarterChild learned was likely not deemed offensive and was, therefore, permitted to go live in the application.

So should the chat demo that runs on the Conversagent site feature this learned information? Or should it stick to built-in responses only? (The drama here lies in my not answering the question.) Conversagent, for its Web site, opted for the full-learning approach, as it 'makes for a more dynamic product.' However, it is rare, Klein said, when someone using the site ' less engaged users than the IM crowd ' hits one of the learnings.

Despite Conversagent's one kinky conundrum, the ASA's ardor ostensibly substantiates its conjecture ' I found the solution to be serviceably innovative.
The unspecific and impersonal way in which the acronym's customer-serving page (FAQ) has traditionally been presented in other solutions, and the search techniques' often vague labyrinth of elusive ambiguity, both may call to mind a Talking Heads song, which laments: 'Facts are simple and facts are straight/Facts are lazy and facts are late/Facts all come with points of view/Facts don't do what I want them to.'

Facts-searching. Or FAQs. The lyrics are apt in both miens. And now there's a new alternative to make them do what you want them to.

[Return To The August 2004 Table Of Contents]