-->

Human Resources: Human Resources Recognizes Speech

Every fall corporate human resource executives throughout the nation brace themselves for a trying time. With open enrollment for health care benefit plans come the seemingly endless editing and off line processes that go with steering employees through this period.

In 1997 that story changed for MedWorks, a large US based pharmaceutical company which used speech recognition to facilitate the process.

The pharmaceutical company had two objectives: first, to automate the existing open enrollment process using an IVR solution, second, to eliminate redundant and expensive off line processes that caused unnecessary delays and frustrations for employees and the human resource group.

The Problem

Historically, IVR enrollment solutions struggled when employees' needed to add a dependent to their record and select a physician PCP (primary care provider). Traditionally, these routines were time consuming, error prone and cumbersome.

When adding a dependent, employees often waited weeks to receive and complete a dependent add form only to have it returned due to an error. Likewise, alphanumeric PCP's caused data entry errors and incorrect physician assignment, taking weeks or months to resolve. These problems not only added expense and delayed distribution of final enrollment materials, but also caused coverage problems in the interim.

After analyzing the enrollment process with the client and looking for the greatest impact to the workload, Lyrix recommended an IVR solution that included voice recognition to select the physician PCP and to add and select new dependents.

The Solution

The challenge was considerable. Because the pharmaceutical company had employees throughout the US, they had to deal with well over 250,000 PCP physician codes of variable lengths.

Lyrix solved this problem by recognizing the employee speech against a more defined PCP database to increase online editing and reduce post enrollment error correction. For dependent additions the problem was more difficult. There was no predefined name vocabulary to recognize against and the spelling often varied for the same name i.e. John and Jon.

Andrew Lazar and Dr. ViJay Thukral, both specializing in speech recognition at Lyrix, had considerable experience in speech recognition name conventions having worked with large directories, some as large as 100,000 names. As a result, they developed an advanced, phonetically spelled name vocabulary, as a part of Lyrix's open enrollment IVR solution, MedWorks.

Using speech recognition routines within MedWorks, the employee was asked to speak and then spell the dependent's name. MedWorks then initiated recognition subroutines against both the name and the spelling; cross-referenced the results and presented the most logical name for the employee to verify. This process was repeated for the first, middle and last names. When recognition failed; the name and spelling were captured for same day transcription.

The Results

By utilizing these techniques Med-Works recognized approximately 90% of the names spoken for dependent selection and new dependent additions plus 92% of the PCP selections during the call. This enabled MedWorks to add the PCP code and the dependent name, along with other time-critical data, to the employee record for immediate access and updates. The editing that resulted from having these additional pieces of information further helped eliminate off line processes and reduce overall costs.

As a result, the pharmaceutical company lowered costs and significantly reduced the open enrollment period vs. prior enrollment periods.

Lyrix is a Boston based provider of advanced CTI solutions to large, distributed companies. They use speech recognition within several of their solutions including auto-attendant, follow-me and others to increase the value and to help better navigate the solution.

Doug Burke is VP of Sales and Marketing at Lyrix, 836 North Street, Tewksbury, MA 01876 .

SpeechTek Covers
Free
for qualified subscribers
Subscribe Now Current Issue Past Issues