Special Report: An Update on Remittance Processing Automation, Part Two: The Promise of AI and RPA

We explored the history of modern cash application in Part 1 of this two-part series. The crux of the story is the expansion of payment technologies and the rise of aggressive AP practices, especially deductions, resulting in an increasingly challenging remittance processing environment. Rather than make remittance processing easier, electronic payment technologies have actually made it more difficult for traditional auto-cash practices to keep up.

Enter both Artificial Intelligence (AI) and Robotic Process Automation (RPA). Before you can understand how these technologies will solve the remittance processing challenge, a deeper understanding of the remittance process is required. That is where this segment begins, followed by an overview of how AI and RPA technologies are driving remittance processing optimization.

Traditional Processes
Up until this decade, expert-designed algorithms and decision tree-based workflows have been the primary drivers of automated remittance processing solutions. Along with these software tools, several means of capturing the underlying data have been used.

Most commonly, a lockbox provider - typically a bank - captures, and often keys in, remittance data. The problem is, the details needed to generated a three-way match (payment, remittance advice, and accounts receivable) bring greater costs.

Optical Character Recognition (OCR) readers, along with user-defined templates, provide an alternative for situations where multiple invoices are being paid by a single remittance. The templates typically require user input on the initial pass for each customer, resulting in increased set-up costs but lower future operating costs.

Additional protocols have also been created to address the large variety of corporate operating environments that impact the way remittance processing is handled. For example, in some industries (such as food and auto parts), customers pay by statements, which can include a large number of invoices. Problems arise if the payment amount does not match the current balance, which can be a big problem for both manual and automated cash posting, especially when there are weekly statements. A system that can recreate the open items on previous statements solves this problem.

Other common issues include truncated invoice or PO numbers being used on remittance advice (or even the original documents) as well as additional characters being appended before or after an invoice number string. Additional algorithms have to be manually created to address the variations that are encountered from customer to customer.

Critical Issues
There are eight critical issues you must address with any remittance processing solution. As the following chart shows, these critical issues present a host of challenges for traditional remittance processing automation tools.

While a company may be able to use, for example, some standard matching algorithms along with the data captured by their lockbox provider, that solution only provides a match for some of the remittances. Increasing the data captured at the lockbox can increase the match rate, but will it be a big enough incremental gain to justify the cost? Likewise, maybe OCR templates could be added to the mix instead of more lockbox data entry, but again does the benefit justify the cost? Traditional remittance processing are only cost-effective to a point. Quite simply, there are technological limitations.

The AI and RPA Fix
AI and RPA can help address these issues and provide solutions that address a greater percentage of transactions at a justifiable return on investment, but neither one is sufficient on its own. "The more complex a business is, the more difficult it becomes to establish effective automation," notes Keith Cowart, Product Marketing Manager at FIS. "A solution needs to have the ability to establish unlimited business rules that govern the system. RPA alone operates under the assumption that processes remain consistent. AI is limited by the exposure to scenarios and feedback on results. To be effective, AI and RPA must be used together within the environment that supports multiple decision points and unlimited business rules."

"The key enabler for the use of RPA has been the centralization of AR systems," explains Sayid Shabeer, Chief Product Officer with HighRadius. The predecessor to RPA, Business Process Automation (BPA), required hard-coded workflows. Now that there are centralized repositories of AR and remittance data, the use of AI-facilitated pattern recognition allows for solutions to move beyond BPA to RPA. This in turn facilitates the recognition of less common matching scenarios, and the subsequent enabling of workflow solutions to handle these very discreet scenarios.

That's the power of AI and RPA working together. Shabeer also relates, "RPA requires less IT support." Instead of hard-coded workflows, for example, user interfaces are available for the creation of automated workflows using drag and drop technology. This in turn contributes to much reduced implementation time-lines.

In terms of remittance capture, AI and RPA-driven process automation can monitor emails and server locations to automatically pickup, digitize and read remittance information to eliminate manual touches. Intelligent Document/Data Recognition (IDDR) technology drives this digitization of both structured and unstructured documents.

Moving on to cash application, AI and RPA can automatically identify deductions and assign reason codes. Likewise, configurable rules can be established to automatically handle short payments and disputes either writing them off if under thresholds or initiating a resolution workflow process that includes escalation protocols when issues are not being resolved within parameters.

The following chart illustrates the differences between a manual remittance process and an automated process driven by AI and RPA. While similar, it is interesting to note that the automated process involves more discreet steps than manual processes. This is because a human operator can process a lot of information, very quickly recognizing important patterns and associations.

Manual and Automated Remittance Processing Compared

For a machine-learning environment to succeed, it is necessary to identify every elemental activity that goes into the process so the resulting AI and RPA tools can be deployed appropriately. In other words, for a successful machine-learning environment, visibility of every activity in the process is required.

Is AI-Driven Remittance Processing Right for Everyone?
Remittance processing challenges vary greatly from company to company. Some have high deduction volumes. Others must deal with multiple invoices on every payment. Multiple payment locations and channels challenge others. One of the key strengths of AI is that it can be applied to all these situations, even when low volumes have precluded other automation solutions.

"When it comes to AI adoption, there's no hard and fast revenue or payment volume constraint for a given client. However, in aggregate, organizations like HighRadius need to see a critical volume of data -- that's high counts of payment transactions across clients in varying verticals -- to build robust AI and machine learning models," says Shabeer.

The key to productivity is the AI system's ability to learn given the transaction volume and complexity it must handle. Notes Cowart, "The beauty of RPA and AI is that there is no minimum level of transactions to make it effective. However, the AI engine will take longer to learn with fewer examples."

"As with any solution," Cowart continues, "there is a cost involved to take advantage of the benefits. Each business will need to review the return on investment (ROI) that they can achieve. This should include both the reduction of expenses as well as the improvement of cash flow. Businesses that have a relatively low volume of transactions will find the ROI curve flatter than those who have more volume."

For example, a business with a single cash application clerk who is able to apply all payments by the end of each day has little incentive to speed up that process. If however, the business is growing or if accuracy is a concern, reviewing possible automation solutions before the business has to start adding resources would be prudent.

The delivery platform for the technology is also important. Rob Sherman, Chief Revenue Officer of VWi argues, "It's not just about transactions, as many tools are available which reduce the total cost of ownership. Gaining access to the tool set in an outsourced model is the most cost effective as these tools are mature, scalable and do not require the major investments to build in-house or buy."

Building your own AI and RPA solution and integrating it with your ERP is probably the least feasible way to precede. Doing that requires a huge IT investment and has you essentially reinventing the wheel.

The issue then comes down to buy or outsource. Both have their advantages. With cloud based SaaS (software-as-a-service) deployments, you are effectively relegating all IT issues -- including upgrades -- to the vendor, as you would in an outsourced environment. The key differences then are who does the manual entry of any unmatched items and how do you pay the vendor (upfront costs verses ongoing costs). The operating scenario of your remittance processing should clue you in as to whether it is best to keep your staff in-house or let an outsourcing partner handle the entire process.

The Benefits of Remittance Automation
Automating remittance processing has always provided a strong return on investment. To a large extent, this is due to the labor savings benefits. If you've had three AR staff members handling cash application, and automation enables you to handle that task with just one person, it's not difficult to assess a monetary benefit.

The goal is to maximize your benefits by driving the match rate as high as possible. This is where AI and RPA are providing the boost that is missing in the current "fragmented payments environment."

Whereas a traditional auto-cash solution might provide a 60 percent match rate, AI and RPA-driven automated remittance processing should help boost your match rate to 75 or 80 percent, and maybe even higher than that. By the same token, if your traditional solution is at that 75-80 percent level, AI and RPA may be able to get you well above a 90 percent match rate.

Keep in mind, it's those incremental improvements in match rate that return the most labor savings, not to mention processing throughput. AI and RPA help clear the 'hard to match' items that are not cost-effective for traditional auto-cash. That's the beauty of AI-driven machine learning.

Interested in learning more about Digital Finance?

Receive access to proven accounts receivable automation research, expert advice, benchmarks, case studies, and more by filling out the form here.