Innovative Research in the Labs, Part VII: SpeechCycle

With SpeechCycle, formerly Tell-Eureka, we have come full circle in looking at innovations within companies and research labs. In past columns we have looked first at individual innovative company announcements, then at how speech R&D is being done in large corporations and universities. At SpeechCycle we find both—innovative research in speech and knowledge management being productized to improve business. However, even without the rich, decades-long history of R&D that some companies have, SpeechCycle's team of speech scientists and subject matter experts in technical support have decades of experience in speech and are providing true, cutting-edge call center automation products that deliver outstanding value in the virtual agent space at a much lower cost. With SpeechCycle's agents able to perform complex tasks such as technical support, Speech- Cycle is offering alternatives to offshoring or nearshoring agents. They call this speech sourcing.

Some companies market virtual agents as part of their applications and promote adaptive interaction to improve the customer experience. Typically these virtual CSRs handle more complex transactions, such as hotel bookings or flight information. SpeechCycle goes several steps further, by enabling its CSRs to perform unstructured problem solving, using the same tools, data repositories, and call center applications that live agents use, so that they can do complex, multilevel problem solving in technical support and customer care.

One of SpeechCycle's prime markets is cable providers that offer triple-play technical support (cable TV, broadband Internet, and VoIP digital phone). Using statistical natural language understanding, combined with triple-play domain expertise and a vast corpus of data, two things happen. First, the virtual agent can use preemptive knowledge to more accurately hone in on the caller's problem before the interaction starts. Using data, such as the type of service and equipment, the agent can better troubleshoot and, at the same time, diagnose operations directly on the caller's devices. Second, the virtual agent has the skill set to understand the problem, and can dynamically troubleshoot with the caller, by having the caller verify PC settings, test basic network connectivity, or power cycle equipment.

SpeechCycle says that rather than deploy ship-and-forget applications, which get tuned up front and abandoned, it uses adaptive resolution management to continuously tune applications through the wealth of data that flows in and gets logged through its LevelOne software. For example, SpeechCycle can build multiple ways of real-time problem solving into an application, watch which way is working best, and dynamically shift traffic to that one. This vast repository of data provides a valuable source of business intelligence for further tuning the applications to best serve customers' business objectives.

SpeechCycle's LevelOne CSR platform scales to support high call volumes and call times. It's comprised of a speech application server that manages speech, call flow, and integration with call center systems; a business intelligence portal for customer information gathering; and a call traffic controller for routing traffic and performing load balancing.  The traffic controller routes traffic and provides differential learning, for example, to evaluate the effectiveness of different prompts in real time to further tune the application. SpeechCycle delivers its products using a hosted, Software as a Service model.

Further R&D

What further advances are forthcoming? SpeechCycle's customer interactions can run 10 to 20 minutes for complex issues, involving dozens of turns between agent and caller, which is much longer than most current virtual agents' calls. To keep caller involvement high, SpeechCycle is researching how to maintain user engagement and context across the length of these calls.

The industry has made a lot of progress in deploying applications with virtual agents, but not to the level of SpeechCycle. For example, the ability to tune applications in real time, based on information from thousands of calls coming in, is not something easily done with live agents. As SpeechCycle works with its data stores and improves upon caller interaction, these agents can only get better.

Nancy Jamison is the principal analyst at Jamison Consulting. She can be reached at nsj@jamisons.com.