A New AI Series on Payables Place
The power of artificial intelligence (AI) is undeniable. Its impact and influence are revolutionizing the Accounts Payable (AP) and Business-to-Business (B2B) payments automation market. According to research from Ardent Partners, AI will soon be ubiquitous in this sector, offering unparalleled efficiencies and capabilities. The transformative possibilities of AI will drive innovations in invoice processing, payment automation, fraud detection, and overall financial management, streamlining enterprise operations and improving business decision-making.
In view of these advancements, Ardent Partners is launching a special “AI in Action” interview series. Our “AI in Action” series represents an opportunity for our Chief Research Officer, Andrew Bartolini to sit down with senior executives at the ePayables solution providers that are focused on AI now and in the near-term and use these discussions as an opportunity to share their company’s focus on, expertise in, and vision for AI within the AP industry.
Today’s AI in Action profile features a conversation with Bill Dorn, SVP Product Operations & Innovation at Corcentric
AB: Good to connect with you again, Bill. While I’ve known you for about 15 years, for our Payables Place readers, tell us a bit about yourself and career.
Bill Dorn: I’m Bill Dorn, SVP Product Operations & Innovation at Corcentric. I’ve had over 25 years of experience in manufacturing, engineering, operations, accounts payable, procurement, product management, information technology, and consulting. I joined Corcentric by way of acquisition, where I was a partner and head of operations of a well-known procurement consulting firm, Source One. I now lead Product Operations and Innovation for Corcentric’s S2P, O2C, and AP products. I have built and sold a procurement consulting firm, managed global SaaS product and engineering teams, authored best practices for IT acquisition for the Department of Defense, advised corporate leaders on AP, procurement, and sourcing, and co-authored two books on managing indirect spend.
AB: You spent many years working with procurement and AP clients, now you’re developing solutions for them. Where are the biggest opportunities for AI to improve accounts payable and B2B payment efficiencies and productivity?
BD: In addition to just freeing up time for people, we see AI as an opportunity to help reduce errors in the workflow and transposition of data, as it can increase compliance while potentially reduce fraud. You can use AI to check itself. We’ve all heard of ChatGPT, Gemini, Copilot, and others. Each one is similar, but has different strengths, weaknesses, and ways to get to answers and solutions. We’re seeing more and more companies that will leverage multiple AIs to check the work of the other AI and determine a confidence level when you see the models agreeing that a proposed answer is accurate. Because of that, you can eliminate a lot of errors and still have high confidence that one algorithm or one language model isn’t hallucinating.
Additionally, pairing the AI automation with existing business processes and systems enables systematic business efficiency increases. For example, after automatically ingesting an invoice and coding it with AI, you can pair the results with your existing business systems and perform two- or three-way match to the PO, and straight-through process the invoice if all business rules and criteria have been met; all without human interference.
AB: Literally everyone’s talking about AI. Many are investigating it – Tell me about how Corcentric’s customers are approaching it?
BD: First, many customers feel overwhelmed with the ins and outs of AI, especially how they should be adopting AI and the steps they must take to implement it. They are eager to implement AI, except their companies haven’t decided what they are trying to solve — which is a critical step. Along with that, security is top of mind for customers. They are concerned that proprietary company info could be exposed through AI usage.
Second, customers want AI adoption to occur in their companies to achieve efficiency, but they don’t have the resources to develop proper rollouts themselves. For example, AI requires clean data, use cases, data scientists, engineers, and existing business rules or systems to pair it with. Overall, most customers feel as though they don’t know where to start or how to begin leveraging AI without the people and resources to properly execute the technology.
AB: In your expert view ,what are the leading AI subsets (e.g., machine learning, natural language processing, deep learning, etc.) with the biggest impact on accounts payable/B2B payment solutions? And why?
BD: GenAI is still a relatively newer technology that’s being adopted. As such, there’s a lot of new use cases coming out as we speak; and it is evolving very rapidly. We’ve already started to implement AI in a couple of different ways specific to the AP workflow. So essentially, one of the most expensive and time-consuming things would be bringing your invoice in and getting the data from an invoice.
Traditionally, that was done by humans. Of course, there’s EDI, APIs, XML, and XML, and a bunch of different formats and technologies that have been around for a while and should help. However, most companies aren’t leveraging them, or they don’t have the IT resources to conduct all of the integrations, especially with their smaller trading partners (buyers and suppliers). As such, north of 80% of all invoices are still coming through as a PDF, email, fax, or snail mail. They get scanned in one way, shape, or form, and then some sort of OCR is involved. But at the end of the day, people are copying and pasting that data over.
Then came robotic process automation; the problem with that is, you’re essentially training a macro and hoping that the invoice doesn’t look different next time it comes around. The challenge there is if that invoice format changes, or you’ve got a new vendor or a new invoice you have never seen before, the RPA breaks and/or needs to be trained. Now, you’ve shifted a person that’s manually keying data into someone that’s managing software and the RPA.
What AI is now allowing us to do, though, is to skip that RPA stage with high success rates, and be able to extract information, understand what is in that information, and then put it back into the system. It will code your invoice as well as a human can do in many cases, sometimes even better. We’ve rolled the technology out about two years ago now and already rewritten it three different times based off the new capabilities that are surfacing out there.
We’re seeing that every new tool and AI model that we add helps us to extract more quality information and increase our client’s ability to straight-through process. When we started with this technology, we might have been at the 50% success rate. Now, by utilizing machine learning, coupled with different AI extraction models to drive confidence and self-learning language models, we’re seeing 90%+ invoice processing without human involvement. With document recognition AI, we can automatically figure out what is being sent to us. With GenAI, we can “fill in the gaps” of data missing from documents. And now, you completely free that person’s time to be strategic somewhere else.
AB: How are you ensuring the accuracy and reliability of customers’ AI models (AI data output)?
BD: As previously mentioned, you can use different AI models to have them work together and find consensus on their confidence that they answered a question or performed a function correctly. You can adjust the acceptable confidence level based on your tolerance for minor errors as well. But at the end of the day, the AI itself is just a tool to typically perform a basic function. The real key is combining AI with business processes that have already been solved. This means using AI to enable business functions, but also relying on existing workflows, processes, and policies —typically, on existing enterprise systems.
At Corcentric, our AI customers have the ability to adjust their confidence levels, and our existing invoice or AP workflow solutions have all of the business rules, including thresholds, routing, and permissions, already accounted for. We provide exception management technologies that allow the customers to easily train our AI models, as well as managed services, which is a human-powered backstop to conduct manual exception reviews.
AB: How do you plan to support your AP clients if/when an AI skills gap exists?
BD: AI is very easily accessible nowadays, and businesses must be cautious in their approach and how they’re adopting it. The very first thing that everybody should be considering is the security of information. You’ll see a lot of news around businesses turning off access to AI. I was speaking to a business leader at a large chemical company recently, and they’ve completely disabled any sort of AI from the desktop experience for any computer, including people going to the web. They can’t get to these websites and the reason for that is not that they don’t trust the AI or that it will not be productive, it’s because of the lack of controls in place around what people can put into AI. It is a learning model, so if I start putting confidential information into this system, that system is now interpreting that information and other people can potentially learn about your business and get access to information that they should not have.
So, first and foremost, most businesses want to validate that the AI they’re using is in their own contained environment, and that there’s a lot of scrutiny around how the information will be accessed by the person that’s writing the AI itself; but it’s also about the difficulty of implementation. There’s a lot of open-source solutions out there that people can tap into, but it’s not as simple as you would think. It’s almost always better to work with a vendor that’s doing this for a living, so to speak, so that you’re not having to hire engineers, security specialists, data scientists, and similar people to support simple business processes.
And then, lastly with the AI considerations, it’s important to ask where you are going to focus. It’s very easy to run down a slippery slope trying to put it everywhere. Because it can do so many cool things and there are so many new use cases coming out, you can easily get distracted from what you’re looking to achieve. So, start with small things that provide low risk, quick rewards, and tackle those use cases first. We say, “don’t boil the ocean” a lot in our team here at Corcentric. Let’s create an avalanche effect by layering small results on top of each other. So, try and have little wins. A 10%-15% efficiency gain is amazing because a few minutes saved here and there stack up. Focus on the big picture but get there with smaller wins.
AB: Where do you see AI going within the ePayables (AP automation) market over the next five years?
BD: I think the AI evolution is still quite early and moving at a pace that no one could have predicted. At some point, we’ll see a normalization and pullback which will likely come with a slowdown in adoption as companies try to focus on what adoption really should be in their businesses. So, the future is still yet to be seen.
One of the biggest things we’re going to see with artificial intelligence, longer term, is it’s almost going to get to the point where you can take AI andlayer it against the business process, and it’ll figure out how to enable or automate that process. In terms of prescriptive business processes, it’s about AI acting as a consultant to your business where it’s suggesting the policies, behaviors, and workflows that help enable you to be more efficient and operate at a world-class level.
Right now, there’s a substantial amount of training, integration, etc., that you must put into AP systems. For AI to work, you must tell it what you want it to do. I do see a world in the not-too-far future where AI will be learning from the community on what great AP processing looks like — and not just from your business perspective. Hopefully, this will help with predictive suggestions on how you can make your payments better to increase your working capital, telling you where to invest your time, where you might have risks with particular suppliers or industries (e.g., geopolitical to environmental sustainability issues).
I think all these are going to start to come together. So, supplier management, procurement, and AP all start to converge into one solution, versus being a back-end function where procurement does their thing, people cut POs, products come in the door and get consumed, and then someone pays a bill. I think we’re going to see all those things coming together with AI, and it’ll be less labor intensive, more accurate, and certainly faster.
AB: Thanks for your time and insights today Bill!
BD: You’re very welcome!
