Analysis: Next-Generation Imaging Beats Data Entry

Document imaging has witnessed some fairly sweeping advances in recent years.
For decades, businesses have been struggling to eliminate paperwork and associated manual data entry--not just to cut costs but also to reduce errors and improve service. They've made some headway with electronic data interchange (EDI), e-forms and Web-based self-service applications, yet they've barely made a dent in the huge volumes of paperwork still involved in day-to-day business. Paper-based transactions are particularly intractable in billing, shipping and other scenarios in which organizations communicate with scores if not hundreds or thousands of customers or suppliers.

Document imaging has long been the only choice for gaining at least some efficiency when paperwork can't be eliminated, yet many think of it as a mature technology that offers only incremental improvements. What these skeptics may not know is that imaging has witnessed some fairly sweeping advances in recent years. For one thing, scanners have plummeted in cost while improving in both paper handling and image quality. Optical character recognition (OCR) engines have continued to advance and now deliver accuracy in the high 90-percent range. Finally, intelligent classification, extraction and rules technologies have reduced the need for pre-scan document sorting and preparation while making capture systems far more flexible and accurate in recognizing and extracting data without human assistance.

Having cut two days and considerable labor out of its imaging-based processes, the example of Trendset Inc. is typical of the productivity gains companies are realizing with the latest generation of capture software. A freight payment services provider, Trendset helps companies including Maybelline (cosmetics), Electrolux (vacuums) and Ryobi (power tools) audit shipping charges and then pay on the most favorable terms. Customers spend anywhere from $1 million to $250 million each year on shipping, and between spotting billing errors and maximizing discounts, Trendset can save them millions.

Large shippers working under long-term contracts tend to use EDI, yet 25 percent of Trendset's workload still involves paper invoices. To handle a massive volume of 120,000 to 150,000 pages per week, the company employed five full-time operators just to open and sort by shipper and customer, insert barcoded separator sheets between individual invoices and then scan all the documents. Because the documents are so variable, Trendset realized few benefits from its conventional capture system, which, like most template-based systems, was designed to look for data in consistent locations on the page. Thus, 25 full-time data-entry clerks were required to key in billing numbers, names, addresses, fuel and mileage charges, and other key data. Despite this platoon of workers, the paper-based process took as long as seven days.

Capture vendors including Abbyy, AnyDoc, Datacap, EMC/Captiva, Kofax, Oce, Readsoft and Top Image Systems have all been working on so-called Intelligent Document Recognition (IDR) systems designed to classify and extract data from variable documents such as invoices, orders and explanation-of-benefit forms. After considering a number of these systems, Trendset implemented the Flexicapture Studio from Abbyy. Now eight months into deployment, the company has realized impressive savings. Because the software can classify invoices and recognize multipage documents automatically, presorting and separator sheets are no longer required. Thus, the company is saving about $100,000 per year on prep labor (eliminating four of the five scanner operator positions) and about $60,000 on separator sheets. And now that software can spot and validate up to 18 key data points that Trendset pulls from each invoice, the company has reassigned five of its key-entry clerks (saving $150,000 to $175,000) while cutting processing times by two days.

"We expect to cut five more data-entry positions within about two months as we add tables and line items to the list of data we extract automatically," says Jay Johnson, Trendset's director of IT. "Right now we handle what we call the header-level information with Flexicapture, but the data-entry staff still has to fill in the line items." The remaining 15 data-entry clerks will handle data validation and correction of low-confidence OCR reads as well the manual data entry required for the 20 percent of documents that have handwriting or poor image quality.

By breaking out detailed information and posting it on a secure Web site, Trendset lets its customers drill down on cost details. "We capture a lot of data, and that's been a differentiator for us. We offer Cognos Cubes and analytics tools on the site so they can sort and break out costs for mileage, fuel surcharges and other aspects of their freight cost," Johnson says.