Ten years ago, the greatest achievement of a speech application was that it recognized words at all. Today, businesses that deploy speech applications, and consumers who use them, take it for granted that speech applications can distinguish among lots of different words. Companies in various industries have discovered ways to offer speech applications that enable their customers to automate tasks like refilling prescriptions, tracking packages or gathering information about flights.

Numerous developers of call monitoring systems employ speech recognition to allow call centers to classify conversations that contain certain words or phrases. This broad type of categorization can reduce the time it takes for those who evaluate agents' calls to flag conversations for specific, unambiguous issues, such as when customers request refunds or when agents deviate from scripts.

But there are limits to what mere recognition of words can accomplish. The presence of a particular word during a call isn't always sufficient to reveal why, for instance, customers close their accounts and whether customers' interactions with agents have any influence on these decisions. Nor does recognition occur in a vacuum. Because words have different meanings in different situations, speech applications only function properly when they distinguish among contexts, and not only among sounds.

To develop more advanced applications that involve one type of technology, it's helpful to borrow concepts from other types of technologies. By definition, knowledge management requires judgment about how and with whom you share information concerning your company among customers, colleagues or partners. Knowledge management complements speech recognition because it entails a cumulative, self-correcting approach to classification. Knowledge management also complements quality assurance and training if it enables call centers to categorize not only calls, but also resources for coaching and training agents on how to handle these calls better. Because knowledge management also requires collaboration, it provides an excellent framework for evaluation.

Like knowledge management, evaluation relies on gathering information from different constituents. Knowledge management and evaluation are also both closed-loop processes that reflect experience and feedback. You can easily tell how often customers and colleagues view specific pieces of information in your knowledge base. And, if you want further details about why they choose (or don't choose) to consult certain items in your knowledge base, you can enable them to rank and comment on these items. Subsequent contributions to your knowledge base can factor in the expertise of those who provide feedback and your organization's capacity to act on it.

Unlike contributions to knowledge bases, which change over time because of an organization's ability to elicit feedback, methods of evaluation change over time according to the needs of those who provide feedback, whether they're customers who complete surveys or quality assurance staff who listen to agents' calls. As with contributions to knowledge bases, evaluations have to weigh certain kinds of feedback more heavily than others.

For instance, a customer who asks a call center agent questions about life insurance policies or mortgages may be able to evaluate the agent's ability to communicate. But unless customers already have expertise in these areas, the agent's colleagues and supervisors historically have been better able than customers to gauge an agent's performance in his or her job.

With that said, the Internet's role in disseminating information has dramatically redistributed expertise among the general public. The Internet fosters a knowledge-sharing culture that, as my colleague Keith Dawson has noted, gives customers more knowledge and more power. Just as it's no longer enough for speech applications only to recognize certain words, it's no longer enough for evaluations only to reflect whether agents say certain words. The practice of quality assurance, and applications of speech technology, must expand their purview from the words people say to the context in which call center agents communicate with customers, and influence their willingness to remain customers.