"We get the worst of the worst."

As Software Improves in Reading Addresses on Letters, Less and Less Humans Are Required

#Workplace

Tue, May 7th, 2013 10:00 by capnasty NEWS

On the New York Times, a fascinating look at the United States' Postal Service's data conversion operators, a group of postal workers who decipher unreadable addresses; however, as software is getting better at reading the poor handwriting on letters and parcels, "workers have been let go, and after September, the facility here will be the post office’s only center for reading illegible mail." There originally were 55 plants providing this service.

“We get the worst of the worst,” Ms. Batin said. “It used to be that we’d get letters that were somewhat legible but the machines weren’t good enough to read them. Now we get letters and packages with the most awful handwriting you can imagine. Still, it’s our job to make sure it gets to where it’s supposed to go.”

Over the years, the Postal Service has become the world leader in optical character recognition — software capable of reading computer-generated lettering and handwriting — sinking millions of dollars into equipment that can read nearly 98 percent of all hand-addressed mail and 99.5 percent of machine-addressed pieces.

That was not always the case. In the beginning, people sorted mail. As the volume and variety increased, the post office turned to automation. But the machines could read only about 35 percent of the mail at first and had trouble with handwritten addresses. So the Postal Service set up the centers, using people to supplement the scanners. At the height of the program, in 1997, the centers processed 19 billion images annually, about 10 percent of all mail at the time, the post office said.

In the last year, this center, and the one in Wichita, Kan., that will close in September, deciphered just 2.4 billion images, or a mere 1.5 percent of the mail, the post office said.

Speed is important. Each worker in this nearly football-field-length room is expected to process about 1,200 images an hour, and they average three seconds an image.

  697

 

You may also be interested in:

#WorkCanWait, a Response to Microsoft's #GetItDone
Waited tables in a deli/restaurant
Telepresence Robots Are Becoming More Affordable
The Little Mouth of Horrors
Tele-Commuting with a Robot