It is no secret that the billion-dollar interactive voice response (IVR) business was built on cost savings strategies resulting from “call diversion” and “agent attrition.” Investment in voice response units (VRUs) was justified by savings resulting from reductions in force in the call center. The math was simple: A call handled “in the IVR” costs less than a dollar, while live agent handling costs in excess of $4.00.

The governing metrics monitored by contact center managers were rudimentary. The heavy hitters were “Agent Work Time,” which was to be kept low, and “Capture [or Automation] Rates,” which were to be kept high. Of course, several other metrics worked their way into the mix, such as “Queue Times” (the length of time on hold) and “Abandon Rates.” Last century, the high cost of inbound transport made it important to try to keep Queue Times short. Now, with the cost of minutes on a network measured in pennies, the length of time in queue has become of secondary importance, although there is a direct correlation between queue time and abandon rates.

Early CAT Deployments Followed Big Iron
Because higher volumes lead to bigger savings, early IVR implementations were large-scale (more than 1,000 ports) implementations at airlines, catalogue retailers and government agencies. Even today, most deployments of CAT are relegated to large applications in contact centers (meaning automated speech recognition, text-to-speech and application logic). A study conducted by The Kelsey Group in 2002 to determine the ROI of speech recognition by a group of well-managed ScanSoft customers reflected this.

All respondents to the 2002 ROI study “met or exceeded” their self-defined ROI goals. The payback period for the average implementation was around nine months. Finally, as an indication of the scale of their deployments, the average annual savings generated by these implementations exceeded $1.1 million. A deeper dive into the data provided in the early study showed that, much as it was with early IVR implementations, control of operating costs — primarily head count — was the primary objective of speech implementations. The fundamental rules applied: higher automation rates and lower agent work time.

That Was Then
For IVR and early CAT deployments, the math was easy, but not necessarily customer friendly. A study that Opus Research fielded in the second quarter of 2005 reveals major changes in the factors that motivate companies to add voice processing systems to their self-service strategies and metrics by which they measure the success of both new installations and system upgrades.

Of course, ROI remains the most important metric from top management’s point of view. In this respect, the results remain encouraging. On average, nearly eight out of ten respondents reported that their ASR deployments “met or exceeded” their financial objectives. Payback periods are still within acceptable parameters. The systems paid for themselves in less than a year, and dollar savings ranged from $300,000 to more than $1.5 billion. Yet, many of the traditional metrics surrounding agent work time are turned upside down.

Three years ago, the second most important factor was to “increase successful call resolution rates”; meaning, to capture more calls in the automated system. In 2005, the second most important factor was to “increase customer satisfaction.” What impressed Opus Research was that customer satisfaction is a factor that came in fourth in 2002. So, in addition to increasing capture rates, it also outranks “increase revenue per agent interaction.”

From an operational metrics point of view, the introduction of CAT into a contact center coincides with a number of changes that reflect the realities of phone based self-service. Average Handling Time (AHT) — or the total duration from the start of a phone call — has gone up even though CAT systems have been implemented. While it seems counter-intuitive to many contact center managers, AHT grows when a speech-enabled system replaces a live-agent or IVR-based service. The predominant reason for this is that speech applications, due to their “directed-dialogue” nature, tend to be more wordy than live-agent interactions. However, Opus Research recommends confirming this after each new CAT implementation by actively monitoring live calls.

While calls are longer, the impact on contact center operating expenses is minimal, adding mere seconds to the duration of a call in an era when network time is very inexpensive. Respondents reported that automated services extended the length of a call by six seconds in the case of new applications and two seconds for applications that are essentially DTMF upgrades.

Surprising Reduction in Agent Talk Time
In a particularly surprising finding, Agent Talk Time (ATT) fell 3.92% for new implementations and 17.28% for DTMF upgrades. This appears to be a testimonial to the power of CAT to continue to deflect calls from live agents. Opus Research characterizes it as “surprising” in the context of the current study because so many of the new ASR platforms have been engineered to handle the most frequent and least complex inquiries — leaving more time-consuming and less routine operations for live agents.

The drop in ATT shows that the automated systems, which routinely handle simple requests like balance inquiries, dealer locators and order status, are expanding their repertoire to include the resolution of ever more complex customer service roles, including a growth in sales-oriented applications. In the next few years, Opus Research expects that, thanks to the expansion of Web-based standards for voice processing (VoiceXML) and telephony (CCXML), enterprise implementers will find it routine to leverage the business logic and databases underlying existing self-service Web sites through speech applications.

Extending Web-based logic and applications through voice platforms is at the heart of plans that will allow enterprises to realize higher ROI by minimizing the staffing costs and capital expenses associated with programming, maintaining, monitoring and tuning multiple, redundant systems.

The lower ATT statistics show that the overall plan is working. Speech as a mechanism for extending Web-based logic and reusable code is ready for “prime time,” real-world applications. By utilizing speech applications as a viable replacement for both contact center agents and knowledge workers, enterprises are seeing a downward trend in ATT and the associated ROI implications.

Life after DTMF
More than half of our sample (55%) were “migrating” or “upgrading” existing DTMF-based voice response applications to incorporate automated speech. The remainder treated automated speech as a “greenfield” technology. This has given us fertile ground for contrasting the financial and operational metrics surrounding the move to automated speech.

Join our Webinar — November 17th at 2PM Eastern (11AM Pacific)
Opus Research analysts will be online and on the phone on November 17 to describe the study’s findings in greater depth and discuss the implications for companies contemplating implementations of automated speech for self-service. Please join Dan Miller and Avery Glasser, along with Carter Gentry of Nuance by registering at http://www.nuance.com/news/webinars/cc_opus/.

Categories: Articles