By Karen Kroll

 

When telecommunications carrier Alltel split in two in 2006, Lynn Smith, staff manager of finance and head of the accounts payable function with the new Alltel Wireless, knew that she would have about 10 employees to handle the approximately 100,000 invoices the new company would receive each year. In a last-minute change, executives with the other half of the spin-off decided to transfer responsibility for processing another 60,000 invoices per year to Alltel Wireless. "We inherited 5,000 invoices per month," Smith says. "We didn't anticipate that."

 

Within a matter of days, Smith needed to let the vendors know of the shift and revise processes and hire temporary workers to handle the invoices. Even so, with a largely manual process and learning curve of several weeks, the department fell behind. Most invoices arrived via the U.S. Postal Service, and those that did come electronically still had to be printed out and scanned to enter into the workflow system.

 

Soon, the daily backlog of invoices was hitting 2,000 for each accounts payable employee. The company was giving up millions of dollars in early pay discounts each year because employees simply couldn't enter the documents into the system speedily enough. And, they had no easy way of identifying the largest invoices, with the most significant discounts, in order to enter those into the system first.

 

In January 2007, Smith and her team members began evaluating different accounts payable systems, deciding a few months later on the A/P Distiller application from Brainware, Inc. Its intelligence engine "reads" the invoices, which come from more than 5,000 vendors in about as many formats, to extract such information as the vendor number and invoice total. It then compares this to the purchase order and transmits it to the ERP system for payment.

 

After several months of testing, training, and customizing, the program went live in September. About three-quarters of Alltel's invoices now are processed electronically, with no human intervention required, Smith says. The upshot? The backlog of invoices is almost eliminated, and Smith no longer needs temporary employees. More important, the company captures about $16 million in early pay discounts each year.

 

While organizations have been streamlining their accounts payable processes for years, most initiatives have stopped short of completely automating the process. Instead, they've focused on such activities as scanning paper invoices so that they could be viewed electronically.

 

Efforts to go further have been stymied by the lack of a standard format for invoices. One might have the vendor name in the upper left-hand corner, and another in the upper right. Even the spellings for the term "purchase order number," can reach into the dozens, when you consider such variations "P.O. number," "PO #," and so on, says Greg Boyd, president of Results Engineering, a Westerville, Ohio-based consulting firm.

 

Now, "things are moving to a new level," says Melissa Webster, program vice president of content and digital media technologies with research firm IDC, Framingham, Mass. Today's systems boast recognition engines that can read and understand even unstructured documents at accuracy levels of around 90 percent, versus about 75 percent just a few years ago. "It's a huge transformation. There's better quality, consistency, and greater visibility," Webster adds.

 

"The engines they use in the software programs are much more capable than in the past," adds Brad Niemiec, a Phoenix-based director of centralized accounts payable with Oakwood Worldwide, Los Angeles, a provider of temporary housing. Oakwood implemented an intelligent data capture application in its accounts payable function several years ago.

 

The languishing economy also is prompting many executives to shift their focus from projects that promise to boost revenue to those that can cut costs. Many document management initiatives, in accounts payable and other departments, fall into that category, says Bob Fresneda, president of ReadSoft North America, a document automation firm.

 

While EDI continues to be a big player in automating the accounts payable process, it doesn't offer the detail that executives need, says Boyd. For instance, an EDI utility invoice probably wouldn't break down the number of kilowatt hours of electricity an organization used in off hours and peak hours or whether the rates charged agree with those negotiated.

 

Of course, any discussion of optical character recognition or intelligent data capture inevitably prompts some skepticism. This is because the accuracy rates of the applications offered even 5 or 10 years ago topped out at about 70 percent. For most organizations, this wasn't enough to justify an investment in this type of system, as too many invoices still required manual processing.

 

Several years ago, Greif, Inc., a packaging manufacturer deployed an OCR application to handle the 180,000 invoices it receives each year, but experienced "a lot of problems with data accuracy," says Barb Myers, manager of central processing. "It didn't work as we had hoped." To be sure, standardized invoices typically went through without a hitch. However, about 25 percent of Greif s invoices come from mom-and-pop shops and are handwritten. "We spent more time making corrections."

 

So, several years after installing the OCR application, Greif turned it off. Myers instead turned to OB10, or Open Business Exchange, to handle the invoices submitted by 275 ofthe company's largest vendors; these account for about 15 percent of all invoices. This "electronic invoice platform" takes the information on the invoices, translates it to a format that works with Greif s systems, and then combines it onto one document that Greif receives electronically into its accounts payable system.

 

Once there, the information is compared to the purchase orders and other information already in the system. If there are any discrepancies, the invoice is kicked out for human review. Greif and its vendors split the cost of the system. The rest of Greif s invoices come on paper and are routed to the appropriate manager for approval after having been scanned into the workflow system.

 

All invoices are entered into the system within about 36 hours of Greif s receiving them. In this way, any questions can be easily researched. And, Greif is less likely to ask a vendor to resubmit an invoice that's been misplaced, which often results in duplicate payments.

 

How Do They Do That?

Today, companies using optical character recognition and intelligence engines typically see significantly higher accuracy rates, experts say. "OCR really does work well," says Alltel's Smith. "In the past couple of years, it's become something you can utilize."

 

How do these applications work their magic? Kicking things off usually is some sort of document capture device, such as a scanner. This gets the paper documents into the system, says Sean Baird, senior manager for product marketing in EMC's transactional content management group. Documents that are received electronically may need a capture solution as well, depending on the format they're in, he adds. Emails and PDF documents, for instance, need to be converted to digital re images so that the information contained within them can be analyzed.

 

An optical character recognition (OCR) application also is needed to identify the characters contained within the document. Today's OCR applications are much improved from those of even a few years ago, Webster notes.

 

Then, the solution needs an intelligence engine to make sense ofthe words and numbers. "It's the secret sauce," says Webster. Given that every vendor's invoice is different, the application needs to first figure out that the document is an invoice. Then, it needs to find such information as the vendor name and amount owed, so that it can process or route the document accordingly.

 

For standard forms, such as tax returns, the software can use templates to extract and analyze the information. So, the system knows that the first five boxes of every Form 1040 always contain the individual's filing status.

 

When documents' structures vary, as they do with invoices, templates aren't of much use. Because the location of information changes, the solutions need to rely on business rules to read them. For example, to find the invoice number, it may be necessary to tell the application to look for the word "invoice," and then to look to the right of or under that for a string of numbers that could be assumed to be the invoice number. As the variations in document structures proliferate, however, continuing to add more rules to work with them becomes unwieldy.

 

Today's intelligent data capture applications use such tools as pattern recognition and mathematical algorithms to identify data. For example, upon encountering a table of figures, it might conclude that one column contains the quantity ordered, while another contains the total.

 

To aid their learning process, the applications need to review several documents of one type. Alltel, for instance, gave the Brain ware application about 100 invoices from different vendors to study.

 

Having extracted the relevant information, such as the invoice number and total, the engine compares this to the information contained in the accounts payable database to determine whether the two documents match. Assuming that they do, the invoice is electronically routed to the next step, such as to a manager for approval or to an ERP system for payment.

 

Utility invoices can be some of the most difficult to process, as they typically extend for several double-sided pages; some even contain advertisements. Even so, at least a few OCR systems are able to automatically process them. Oakwood Worldwide receives about 80,000 utility invoices each month, Niemiec says. Prior to 2006, the company relied on several dozen staffers and temporary workers to key the information into a homegrown accounts payable system. They averaged about 125 invoices per day, as they had to look up the apartment address that correlated with the bill and code each invoice with a general ledger account number.

 

What's more, if a bill didn't get into the system quickly enough, the service could be disconnected. Customers at Oakwood's temporary residences, needless to say, didn't appreciate this.

 

Niemiec and his colleagues decided to change from Oakwood's time-consuming, paper-intensive, manual system to an electronic process using Kofax Intelligent Capture and Exchange. The Kofax system electronically searches for information, such as the vendor number, from each invoice, so that it can be compared to the data in the accounts payable system. Assuming that the information matches, a check is cut to cover the invoice and the information flows to the accounting system.

 

Because they don't need to enter the data but instead can focus on reviewing it and researching discrepancies, Oakwood's employees each process about 300 invoices daily. Most checks are cut within about four days from the time the invoice is received. As a result, disconnects have been cut from 0.05 to 0.025 percent of services provided. "Service recovery is costly, especially if food is damaged," Niemiec says. In addition, the size of his department dropped from 24 to eight. Between the two changes, Oakwood recouped its investment within about 18 months, he adds.

 

While OCR and intelligent data capture applications have advanced tremendously in the past few years, the technology still is in its adolescence, says Henry Ijams, managing director with Paystream Advisors, Charlotte, N.C. That is, an organization can't simply install it and forget about it. The need to continue monitoring and working with the software has prompted some firms to take advantage of these solutions by outsourcing them, he adds.

 

The cost to implement an intelligent accounts payable system can vary dramatically depending on its size and complexity. Brainware's software licensing fee, for instance, can run between six and eight figures, says James Zubok, chief financial officer and cofounder. Ongoing maintenance costs are in addition to this.

 

Even so, the magnitude of efficiency gains that result from automating accounts payable has many companies reporting payback periods of less than one year. Of course, like EPA mileage ratings, discussions of returns on investment need to be taken with a grain of salt, says Don Post, partner with IMERGE Consulting, Rockford, 111. Companies with lower volumes or widely differing forms are likely to have longer payback periods than companies processing large volumes of similar documents.

 

The job of the accounts payable employee changes once most invoices no longer require a human review. "They don't have to be data entry clerks with their heads down," Webster says. Instead, they can manage exceptions and focus on optimizing the use of early-pay discounts and monitoring vendor compliance.

 

At this point, only about 40 percent of companies have implemented imaging solutions on the front ends of their accounts payable processes, estimates Ijams. The investment required to change processes holds the rest back, he says. And, many executives assume that the benefits lie just in reducing labor costs, while overlooking the greater visibility to cash flow and more efficient procurement processes. "When you're competing on a global basis in real time, anything that you can do to take from costs will go to the bottom line," says Andrew Pery, chief marketing officer with Kofax, Inc.