USPS talks up help desk system
- By Jason Miller
- Oct 17, 2002
USPS' old help desk system took about 90 seconds to answer a caller's question. The new system cuts the time to 30 seconds.
When Larry Wills took over the Postal Service's internal help desk organization, he couldn't believe how difficult the automated telephone support system was to use. The manager of distributed computing and IT for USPS said it took 90 seconds to listen to the menu, which included 60 options on three submenus.
'By the time you went through the whole sequence, you couldn't remember what number to hit,' he said. 'It was pitiful. I thought there had to be a better way.'
Wills thought speech recognition software would make the system easier for employees to use and save the agency money.
'Our old system was built up over the years as our network grew and was not designed for what we were using it for,' Wills said. 'The speech application would route users immediately instead of having to listen to all the options and go through the menus.'
The help desk receives 7,000 calls a day, and the old system took 90 seconds to answer the caller's question. The new system cuts the time to 30 seconds, Wills said.
Wills contacted AT&T Corp., which provides long-distance and other phone services to more than 500 postal facilities, to see if the company could integrate the voice technology help desk system into the services they already provide.
AT&T sells the internal help desk service to USPS on a per-minute or per-call basis, Wills said.
The new system, which began operating in June, uses both speech recognition and text-to-speech software from Speechworks International Inc. of Boston. The speech recognition function lets callers 'talk' to the computer. The text-to-speech technology lets the computer respond to the user.
'With speech recognition software, the computer uses phonetic recognition based on how the letter should sound,' said Stuart Patterson, chief executive officer of Speechworks. 'The text-speech software reads instructions that are logged into a large database that usually would be read by a person off their computer screen.'
Both applications were written in C and C++ using voice over Extensible Markup Language, Patterson said. Both programs convert human vocal sounds to .wav files. The program uses algorithms to match the words against a vocabulary database, Patterson said.
Steve Gibson, manager of technical user support services for the Postal Service, ordered 800 additional phone numbers to move part of the help desk system on USPS' network to AT&T's system.
Gibson's office created a list of terms that callers use to describe problems, and AT&T put the words into the system's database.
'We discovered going into this that we were thinking like programmers or designers and not as callers,' he said. 'We would think major system, minor system and then component and callers would go right to the component, like printer or scanner. We were forced to rethink the way we put the system together.'Tested first
Gibson said USPS tested the system by putting 10 percent of all help desk calls through it.
'We didn't get a lot of negative feedback with 10 percent of all calls, but when we increased it to 50 percent, we got a lot of negative feedback,' he said. 'We missed so many things that we shut it down two days later.
'You really need to hit that critical mass to have enough people that are having a problem to understand the issues.'
It took about 10 days to fix the problems. Gibson said he slowly increased the number of calls routed to the new system back to 50 percent.
Now 100 percent of help desk calls run through the new system. The old system will run parallel until users feel comfortable with the speech recognition system, Gibson said.