UT Southwestern Medical Center Routes Calls Using Speech
The University of Texas Southwestern Medical Center at Dallas is part of The University of Texas System and is governed by the UT Board of Regents. The medical center's mission is to educate future health professionals and scientists; to remain at the forefront of biomedical research; to provide compassionate, scientifically-based care for the sick and preventive care for the well; and to provide a continuum of medical education for practicing physicians and medical scientists. As part of this mission, UT Southwestern is continually seeking new technologies to simplify daily activities for its employees, students and patients.
UT Southwestern provides inpatient care to nearly 89,000 people and manages approximately 2.1 million outpatient visits annually. Its physicians provide care to patients at UT Southwestern University Hospitals, Parkland Health & Hospital System, Children's Medical Center Dallas, the VA North Texas Health Care System, the James W. Aston Ambulatory Care Center, the Simmons Comprehensive Cancer Center in Dallas, and other affiliated hospitals and clinics in North Texas.
Serving as an academic medical center, UT Southwestern is a patient-care provider and research institution of more than 4,000 medical, graduate and allied health students, residents and postdoctoral fellows each year and more than 9,300 employees. UT Southwestern also conducts more than 2,500 research projects annually totaling more than $340 million. Deploying Speech
Applications for
Call Routing "Deploying Speech Applications: Call Routing" is an upcoming topic at SpeechTEK in New York City Wednesday, August 9, 2006. Michael Ahnemann of Intervoice, John Kirst of TuVox and Kirshna Govindarajan of Nuance will be presenting during this session and have provided some of the information below on natural language call routing. Natural language call routing, also referred to as "Call steering," "How may I help you?" and "Say anything," enables users to utter a few words about what they want rather than transverse a hierarchy of menus answering questions that eventually route the user to the desired target. Natural language call routing uses a Statistical Language Model (SLM) to enable users to answer questions like "How may I help you?" with open-ended responses to be routed to the appropriate area/agent. When used properly, the SLM can reduce the amount of thought required to navigate a menu tree. Ways to improve the effectiveness of an SLM system: · Have a directed dialog back-up for the SLM grammar to help direct callers who do not want to speak with an automated system. · Design the automated system to reassure a caller that he/she is on track to resolving his/her issue as well as to ask effective follow-up questions to help route the caller to the appropriate place. · Perform careful analysis of caller goals and gain an understanding of the company's call center and how it is organized. |
UT Southwestern first learned about speech during a road show hosted by SpeechWorks, now Nuance. The company was displaying its new product - an auto attendant called SpeechSite. The medical center first implemented the SpeechSite application in September 2001 to answer external calls to the medical center's main listed number as well as internal calls from employees who would dial zero to speak to the operator. The following year, UT Southwestern expanded its use of SpeechSite to two other affiliated hospitals: St. Paul and Zale Lipshy University Hospitals. To date, the auto attendants have saved more than $2 million in operator salaries as well as reduced hold times and eliminated the need for paper or electronic internal directories. Looking to expand its use of speech to the outpatient billing system, UT Southwestern implemented a solution using VoiceObjects X5 Voice Application Management System in January 2006.
The primary goal for the first phase of the expansion was to implement the system in a short time frame. Other goals, which apply to future phases, include reducing the number of calls to live agents, reducing agent "talk time," and reducing hold times for callers. UT Southwestern also wanted the ability to move to another Interactive Voice Response (IVR) platform in the future without modifying its existing applications. With these future phases, UT Southwestern hopes to automate calls related to report balances, payment details, and credit card payments. The biggest challenges that the medical center faced in realizing these goals were defining the existing processes and developing and structuring the databases to meet the requirements of the applications. Before moving forward with the first phase, UT Southwestern performed usability testing with internal employees, who are often patients of the medical center's physicians. The results of these tests demonstrated a need to tweak the wording of some prompts and add synonyms and/or shorten responses to the grammars. The medical center continued to monitor calls after the system went live and found additional prompts that warranted this same type of fine-tuning. Future phases will likely involve the use of focus groups to test the functionality before those phases are brought into production.
UT Southwestern used VoiceObjects X5 to design a call routing application to direct callers with patient billing inquiries to the appropriate areas of its account services department. The solution needed to work in conjunction with the medical center's existing and future IVR platform, so UT Southwestern decided to migrate all of its voice-driven applications to VoiceObjects X5 with its media platform driver concept. The call routing feature is just the first phase of the application. Future phases will provide the caller with his/her current balance, recent payment and payee information as well as the ability to make a credit card payment using speech technologies. For calls that the system cannot handle or that zero-out, the initial information gathered during the interaction with the automated system will then be transferred to the live agent via screen-pops. This is designed to reduce the agent's total talk time with each call. The medical center also prototyped, created, and deployed a disaster response system for its employees. System administrators dial a number and use speech recognition to record a disaster-specific message and select an option to have it play to a published employee disaster number. Administrators also have the ability to record and play an "All is normal" message for employees who call into the disaster response number after a disaster has been addressed and mitigated or during "normal" times. UT Southwestern developed and deployed both of these applications in less than one month.
From the initial deployment of speech through the latest use in the disaster response system and call routing application, internal call volumes have increased by more than 70 percent with completion rates for all of the speech-enabled applications averaging approximately 90 percent. A recent platform change and VoiceXML deployment for the medical center's auto attendant brought call completion rates down initially to 72 percent. With prompt changes and increased tuning, after one month, the rate is now 84 percent and expected to climb back above 90 percent within the next month. While the original auto attendant eliminated the need for 12 full-time operators, the medical center has no plans to measure the new applications until the next phases of the patient billing application are rolled out, and will not have results for the impact on the number of agents and agent answer times before then. According to Elwyn Hull, UT Southwestern's telecommunications director, "Speech technologies have enabled us to present a more standard and professional image for our institution as well as provide a more user friendly interface. We have already saved millions using speech technologies and foresee additional savings as we move into future phases of the patient billing application."
As with most technologies, speech can have drawbacks. In the medical center's case, large recognition grammars, especially those in the auto attendant application, have a slight tendency to cause difficulties with name recognition. Also particularly difficult for the applications to recognize are names that are short or have very "soft" sounding consonants. The speech systems' abilities to recognize words, phrases and sentences are often complicated with thick or foreign accents. Outside factors such as speakerphones and some cell phone transmissions have also skewed the recognition. However, these minor nuisances have not prevented UT Southwestern from expanding its use of speech technologies.
In addition to the upcoming phases of the patient billing application, UT Southwestern began developing in-house speech-enabled user surveys for auxiliary departments that want to survey the relevant university community regarding the level of support provided by that particular auxiliary department. The medical center is also in the process of developing speech-enabled, "mini-auto attendants" for some of its larger departments. The "mini-auto attendants" will direct callers within each department. UT Southwestern plans to expand the use of speech in several other areas as well as replace an old IVR system. In the next four months, the medical center is considering the implementation of speech applications for guest and patient subscription in its in-house data network as well as for the medical center's help desk to automate functions such as password reset, status check for open issues, and call routing. Within the next two to three years, UT Southwestern plans to implement new speech recognition front-ends for all of its clinics, replacing the existing Dual Tone Multi-Frequency (DTMF) front-ends (touchtone system); an application to handle continuing education enrollment, scheduling and payment; and voice-enabled patient surveys.
Stephanie Owens is the associate editor for Speech Technology Magazine. She can be reached at stephanie@amcommpublications.com.