Huntington Bancshares Inc. last week became the first U.S. bank to process deposited checks using advanced imaging software that recognizes handwritten characters.
The software enables computers to automatically "read" numbers, letters, and other characters from paper documents and computer images.
Columbus, Ohio-based Huntington's system is designed to recognize the "courtesy" line on checks, where the amount is written numerically.
This function was crucial in making imaging systems for item processing cost effective enough to justify their multimillion-dollar price tags, technology experts said.
Another 10% in Savings Seen
"Installing [character recognition software] is part of a concerted effort by the check-image banks to improve on the 20% to 30% efficiency gains already made," said Stanley E. Miltko, a consultant with Littlewood, Shain & Co., Exton, Pa.
"When correctly implemented, the software can reduce labor [costs] an additional 10%," he said.
Last Wednesday, Huntington officials said they took a giant step toward achieving that goal, processing about 110,000 checks with the amount-recognition software.
Fewer Operators Needed
When equipped with the new software, a check-image system can automatically extract the courtesy amount from a large percentage of checks, thus reducing the number of data entry operators that have to key in that information manually.
Huntington's business case for the multimillion-dollar imaging system from Unisys Corp. is based on the assumption that it will "read" about 45% of the 300,000 checks deposited at the bank daily.
In its first live run, the character recognition module handled only about 33% of the check work, but bank officials said the low rate was expected in the initial stages.
"We are operating the system at very conservative settings at this point," said George Mackinaw, senior vice president at Huntington.
"We'll continue to proceed cautiously for the next couple of weeks, but we should hit 45% by mid-December," he said.
Consultants said caution is needed because anytime a check runs through an imaging system equipped with character recognition software, there are three possible outcomes.
First, the check amount can be read correctly. Second, the system can acknowledge that it will not be able to recognize the character in the amount field and pass the check on for manual entry. Third, a number can be misread.
Costly Situation
The last eventuality, which results in what is termed a "substitution," is one that banks are most eager to avoid. These mistakes result in balancing errors that require extra labor at the end of the process and can sap some of the productivity gains attained in the data entry area.
At the most conservative settings, only about one in 200 checks read by the software resulted in substitutions, which is less than the average error rate of manual data entry clerks of one mistakes per 150 items, Huntington officials said.
When the system is operating at a level that allows 45% of checks to be read by character recognition software, Huntington officials said they expect the rate to rise to one error for every 75 checks.
"The trick is going to be striking a balance between decreasing key entry [staff] and the resulting increases in balancing work," said Mr. Miltko of Littlewood Shain.
In addition to Huntington, several other institutions with Unisys check image processing systems are testing the courtesy amount recognition software. These include Comerica Inc., Detroit; Signet Banking Corp, Richmond; and Barnett Banks Inc., Jacksonville, Fla.
International Business Machines Corp. said its long-awaited check image systems - which come equipped with character recognition capabilities - will be generally available next month.
