Magazine Article | August 21, 2008

Save Big With Automated Invoice Processing

Source: Field Technologies Magazine

An OCR (optical character recognition) solution allows Alltel Corporation to automate invoice processing and recoup nearly $735,000 per month in early payment discounts.

Integrated Solutions, September 2008


Lynn Smith, finance manager for Alltel Wireless

Everybody likes a good discount. Discounts, whether a result of clipping coupons or bulk shopping, allow us to stretch our dollars in impressive ways on a personal level. However, in business, discounts can play an even more significant role in a company's bottom line. In fact, most corporate accounts payable (AP) departments establish set discounts with their top vendors for early payment of vendor invoices. These discounts are typically small on a percentage basis, but can result in thousands of dollars in savings per month if the quantities or dollar amounts of the invoices are large enough. Therefore, ensuring your AP department does everything it can to take advantage of these early payment discounts can have a significant impact on the financial performance of your business. Alltel Corporation's AP department recently realized its manual system for processing vendor invoices was causing it to miss out on approximately 60% of its early payment discounts. In response, the company implemented an OCR system that automates invoice data collection and accelerates the payment process.

With more than 13 million customers and $8 billion in annual revenue, Alltel Corporation is the fifth-largest telecommunications provider in the United States. In July 2006, the company decided to split its landline and wireless divisions into two separate companies. This split effectively cut the company's AP team of 45 employees in half. At the same time, it was decided that the landline company would no longer process any network invoices. Therefore, the AP department for Alltel Wireless inherited 5,000 new invoices per month (in addition to the more than 10,000 it already received) that it had to process with half the staff.

At the time of the split, Alltel Wireless' AP department used a manual method of processing vendor invoices. Invoices were received in paper form via the mail and electronic form via an electronic mailbox. This electronic mailbox, however, was only electronic to vendors. Alltel had to print, prep, and stamp all the invoices that it received electronically and image these documents on a Fujitsu fi-5900C scanner in the same fashion as it did the paper invoices. Each image was then routed to a group of data entry employees who were responsible for manually keying important invoice information into the imaging system in order to create an index for that image. Once this index was created, the image was uploaded into Alltel's Open Text electronic workflow system to route for corporate approvals.

The primary bottleneck in this system occurred at the manual data entry phase. "We had three full-time employees and seven temps working 10 hours a day and six days a week trying to manually index our 15,000-plus monthly invoices in a timely manner, but they just couldn't keep up," says Lynn Smith, finance manager for Alltel Wireless.
Smith notes that one of the reasons for these data entry delays was the unreliability of the temp workers. "It would take us three weeks to train new temps on what to do, and, in the end, many of them weren't motivated or able to do the job," she adds. "This resulted in a lot of turnover in that area of our department. In fact, between 30 and 40 people probably filled those 7 temp positions over a year's time. This instability inhibited our invoice processing speed, and our invoice backlog for indexing on any given day soon reached more than 2,000. It wasn't uncommon for invoices to sit in our indexing queue for a number of days or even a week."

This backlog was a big problem, given the importance of image indexing to the overall invoice payment process. For example, even if all the invoices were scanned into the system, nobody could tell what vendor the invoice was for or whether the invoice was for $5 or $5 million until this information was manually entered into an index file. This index bottleneck, and the lack of information that resulted, wreaked havoc with Alltel's ability to receive the early payment discounts it established with vendors.

"We receive a 2% discount off each invoice if we pay the vendor within 10 days," says Smith. "Because of the index backlog, we were missing out on more than 60% of these discounts. Plus, since we didn't have any information on an invoice until it was indexed; there was no way for us to prioritize payment of higher-value invoices in the system. In other words, the 40% we would pay early could mostly be low-value invoices, while multimillion dollar invoices sat in the queue. The net result was we were losing roughly $900,000 per month in early payment discounts."

In October 2006, Alltel's VP of procurement discovered just how much money the company was throwing away on missed early payment discounts and developed a task force aimed at getting the company's AP processes back on track. This task force consisted of Alltel managers in accounting, finance, and procurement. After several meetings, Alltel decided that it needed to deploy an OCR solution that would help automate the invoice indexing process and reduce the company's dependency on temporary workers. Beginning in February 2007, the company evaluated two OCR vendors and ultimately selected Brainware A/P-distiller OCR software in May 2007 (see sidebar below).

"I limited the number of vendors in the evaluation process because it was important for us to get the solution deployed quickly and start recouping more early payment discounts," says Smith. "I didn't want to waste a lot of time interviewing more vendors than I needed to. I found a solution that I thought was a great fit for us in Brainware and initiated the project."

Alltel provided Brainware with a sample set of 100 invoices that were representative of the invoices the company received at the time. Based on these invoices, Brainware's implementation team created a one-to-many learn set in the software that would allow the company to automatically process 180,000 invoices from 5,500 different vendors annually. A one-to-many learn set is a feature in which the software makes current data capture decisions based on the historical data provided by previous documents. For example, the software will scan an entire invoice document until it identifies a specific term, such as 'invoice number.' Then, based on the trends the software recognized by scanning past invoices, it will begin to look to the left, right, top, and bottom of this term until it locates a series of digits resembling an invoice number. It will then extract this information as the invoice number and electronically route it to a human operator for verification. 

Alltel implemented its Brainware solution in September 2007, and the system changed the company's entire front end process. For example, Alltel no longer needs to print and prep the electronic invoices it receives. A/P-distiller integrates with the company's electronic mailbox and retrieves and classifies electronic documents every 10 minutes, seven days a week. A/P-distiller then identifies and extracts pertinent invoice information (e.g. vendor name, vendor number, invoice number, dollar amount) from each electronic invoice and scanned invoice image and automatically creates an index for each file.

More InfoRead another automated invoice processing success story at

Questionable characters are presented to Alltel's data entry employees, who either verify that the data the system captured is correct or manually change data captured incorrectly. Once verified, the invoice images are uploaded to Alltel's electronic workflow system for corporate approvals. Once approved, the invoice data is uploaded to Alltel's JD Edwards system, checks are cut, and the invoices are paid.

Since manual keying is no longer required to index invoices, Alltel was able to eliminate the seven temps and one and a half of the three FTEs (full-time employees) it used to rely on for this function. Now, only one-and-a-half FTEs are required to verify captured data and deal with exceptions (e.g. incorrect totals, missing vendor, or shipping information, etc.).

More importantly, the Brainware solution has allowed Alltel to substantially increase the percentage of early payment discounts it receives. "We were missing about 60% of our early payment discounts, but our new OCR solution has reduced this figure to just 11%," says Smith. "Where we were losing $900,000 per month in early payment discounts, we're now only missing out on $165,000 per month. By automating invoice processes using OCR, we were able to recoup roughly $735,000 per month." Smith is extremely happy with the results of Alltel's new automated invoice processing solution, explaining that the company will never be able to hit 100% of its early payment discounts because there will always be inconsistencies between incoming invoices and purchase orders that will need to be rectified, which impedes the process. However, the solution has removed manual indexing as a bottleneck in the invoice payment process, and the money recouped as a result has already more than paid for the technology investment.