How AI Will Shape Accounts Payable

Accounts payable professionals have all undoubtedly heard of artificial intelligence and how it’s likely to change the business landscape. But how is AI expected to impact AP’s functions? 

In this podcast with David Glass, IOFM discusses the different types of what may be considered AI — from optical character recognition and robotic process automation, all the way to sophisticated machine learning capabilities. Glass discusses how these things differ, from simply reading in scanned fields on an invoice to having the technology interpret what it’s seeing, regardless of document layout; matching invoices to POs; and communicating with vendors. He explains how AI will ultimately simplify the entire end-to-end payment process.  

Listen in to discover how Glass is already using AI, sometimes in unexpected ways, and how to get started on the AI journey, even without having the technology in place yet. 

David O. Glass, MBA

David O. Glass, MBA, is a seasoned Accounts Payable Leader renowned for his expertise in finance, accounting, strategic planning, and information technology.  With a remarkable career spanning various leadership roles, David has excelled in managing day-to-day operations across Accounts Payable and Treasury departments for multiple organizations across North America.  His accomplishments include turning Accounts Payable into a profit center, the implementation of 'touchless' invoice entry systems, cross-functional collaboration resulting in an $80 million positive cash flow, and successful ERP system migrations and implementations.

David's financial acumen extends to cash flow forecasting, long-term borrowing strategies, and relationship management with financial service providers.  He has maintained strong ties with regional banks and minimized foreign exchange exposure, demonstrating a commitment to fiscal responsibility.  As the Director of Corporate Accounts Payable at the West Virginia University Health System, David continues to provide strategic leadership, emphasizing efficiency and compliance.  With a Master of Business Administration from Waynesburg University and a Bachelor of Science in Economics from Slippery Rock University, David combines academic rigor with a proven track record of achieving remarkable results.

David Glass's career has been defined by a commitment to financial excellence and process optimization.  His extensive experience and notable achievements make him an invaluable asset in financial leadership and operational roles, where he consistently delivers tangible results and contributes to the success of the organizations he serves.

The West Virginia University Health System, which operates under the brand name WVU Medicine, is comprised of 22 member, managed, and affiliated hospitals in West Virginia, Maryland, Ohio, and Pennsylvania.  It is West Virginia’s largest healthcare system and largest private employer with more than 2,500 licensed beds, 4,900 providers, approximately 30,000 employees, and more than $5 billion in total operating revenues.

Transcription

Female Voice: Welcome to the IOFM podcast. This is a podcast for accounts payable and accounts receivable professionals who want to stay in the know with current AP and AR trends and ideas. We'll be interviewing professionals in this space on a wide variety of subjects, including automation, artificial intelligence, career growth, compliance, leadership, and much more.

Today we have an interview with David O. Glass, MBA, who is a seasoned accounts payable leader renowned for his expertise in finance, accounting, strategic planning, and information technology. David has had a vast career, spanning various leadership rolls, and has excelled in managing day-to-day operations across accounts payable and treasury departments for multiple organizations across North America. His accomplishments include turning AP into a profit center, the implementation of touchless invoice-entry systems, cross-functional collaboration resulting in an $80 million positive cashflow, and successful ERP system migrations and implementations. 

00:01:08

As the Director of Corporate Accounts Payable of the West Virginia University Health System, David continues to provide strategic leadership, emphasizing efficiency and compliance. The West Virginia University Health System, which operates under the brand name "WVU Medicine," is comprised of 22 member-managed and affiliated hospitals in West Virginia, Maryland, Ohio, and Pennsylvania. It's West Virginia's largest healthcare system and largest private employer. David's career has been defined by a commitment to financial excellence and process optimization. 

He'll be interviewed by IOFM's Managing Editor, Royce Morse. Royce has been working with IOFM for the past eight years, writing and editing content about AP, AR, automation, and industry trends. She has worked on the IOFM Certification Guides and written the associated examinations, edits the Annual 1099 and 1042 Master Guides, conducts podcasts, and manages the IOFM website content. 

Royce Morse: Hi, David, and welcome to the IOFM podcast series. I'm glad to have you here.

David Glass: Thank you. I'm glad to be here.

00:02:13                     

Royce Morse: So today we wanted to talk about artificial intelligence a little bit, and I know you have experience with that. But before we get into the details, I was hoping you could break down for us a little bit about what different techniques people refer to as "AI" and what you consider to be true AI.

David Glass: Looking at AI, I'd really break it into two processes. What most people know is their GPS in their car. You program in your destination and as you go the software is determining if you're on the correct route, and if you have left the route, it's telling you how to return to it.

00:03:03 

The step beyond is the machine learning part of AI that says, I routed them to the destination, but they drove around the block. The machine learning then says, well, if they drove around the block, I should calculate that within the route, and it was because I needed to park on the other side of the street every time I went to that location. 

Royce Morse: So how does that translate then into like accounts payable operations? Can you kind of extrapolate that for us?

David Glass: You're using the AI to make predictions, so the more datapoints that you have — if you're able to within your OCR (optical character recognition) software capture bill to, ship to, routing numbers, account numbers, logos, as much information as you have and you maintain within your ERP, the AI can connect those datapoints.

00:04:06

If you're missing a piece of information, kind of like you're putting together a sudoku puzzle, it fills in and says, well, I have this ship-to account number, which ties to this cost center, and [it] can continue filling in that information. The more time it think its successful, it predicts and captures more of that information at the same time. 

Royce Morse: So would you characterize — there's a lot of confusion, I think, among people about what really is AI and what [it] isn't. Automation and AI are not synonymous, so would you characterize optical character recognition (ORC) as AI?

00:04:51                     

David Glass: No. The OCR is really a method of collecting information, and then how you use that — whether it's a robotic process automation that says, based on this, make a decision — kind of like a train rolling down the tracks, it can do a left or right in those decisions. If you have a purchase order, you could have robotic process automation that says, if it's a one-to-one match to the purchase order, select all of the lines on the purchase order and move on.

Royce Morse: So it's kind of a binary decision-making process, it sounds like.

David Glass: Rule-based. So when you're doing your RPA, it's rules that really drive that in your determining a path to maintain where the AI starts filling in the gaps in the information. If you sat in your car and you started driving — I was on the road once and the location of the road had actually shifted, and the GPS kept saying, "Return to the road. Return to the road." The AI would tell me how to return to the road, and eventually it would learn that the road location had moved.

00:06:17                     

Royce Morse: Okay. I see what you're saying. So RPA (robotic process automation) sounds, from your description, as though it's more of — you said "rules-based." It doesn't really think. It doesn't really extrapolate information and then come to conclusions based on that. It just knows what it knows, and that's all that it knows.

David Glass: Yeah. If you have a calculation that says — the most common is approval processes. You route an invoice to a cost center, and if it is above that manager's threshold, you need a second level of approval. It can take action based on a decision point.

00:07:04                     

Royce Morse: Okay. So if you are in an AP role and you're thinking about automating — let's say you aren't automated, or all you have is some OCR scanning capability — would you recommend somebody go all the way to true AI from there, or would you recommend that they go to RPA first? What's your thinking on that? How would you phase that in?

David Glass: I start by getting clean data. So within WBU Medicine, any time we integrate a new hospital, all of the supplier information is standardized. So we run all of the addresses through the post office standardization tool, so every address is in the same format there, the zip-plus four. Once you have standard data, then you can start thinking about automating and saying, the OCR captured a zip-plus four, and identify the correct supplier.

00:08:14 

I would always start with clean data. Create as many datapoints as you can. And are there things, such as the zip code, routing number/account number combinations, that you can make decision points upon? And once you have those, then you can move to the robotic process automation and make the decision of where something should be charged, and then move to AI, where you're starting to compare two contracts and fill in gaps in processes. 

Royce Morse: So it sounds like what you're saying is: Get your house in order first. Don't try to automate a messed-up process or inconsistent data. And then think about how you want to proceed from there.

David Glass: Yeah. I think it's getting invoices either in EDI or in electronic format that can easily be ingested into a tool, which is the RPA, and then making decision points.

00:09:19 

Recently, we used some logic that we extracted all of our invoices processed for this year and, through two formulas in Excel, we were able to provide our workflow partner decision points for 50% of our invoice volume. It took us about half an hour to extract that and provide that starting data or them. From that, we can get the consistency that makes the AI successful. 

00:09:56 

If we do get an invoice a month from now and we've processed ten invoices for that supplier, and they've all been charged and approved to the same cost center, but we receive a poor-quality document or the purchase order was omitted, the software can go out and say, look at the open purchase orders for that supplier and compare the amount and insert that purchase order, even though the supplier omitted it. 

Royce Morse: That sounds like that would be a great labor-saving device if you could get that working well. What would our advice be to somebody? Let's say I'm a small company I do 10,000 invoices a year or something. I'm thinking, you know, I really could streamline this process. We're spending an awful lot of time doing manual work on this and doing researching stuff and trying to match things up.

00:10:58

Let's say I do have my data standardized and it's in pretty good shape. What would you recommend as a next step for me? 

David Glass: I would look at the initial processes of: How are you ingesting your invoices? Is it someone using a scanner? Is it automated through an email? Or worst case, somebody's printing it from the email and then scanning it. [laughter]

Royce Morse: Oh, my gosh! Okay. That sounds awful.

David Glass: The approval processes that lean more toward those RPA processes and knowing your workflows, so when an invoice comes in, what are the decision points? How do you determine that it's an invoice that you want to process? Or is it an advertisement that you received in your email account? Can you get rid of that nuisance information first? Then, once you're there, you can create those simple decision points that RPA works well with.

00:12:04

I had worked with a dental group about 15 years ago that was expanding nationally and, at the time, they were purchasing another group that was going to double their size. In understanding their workflow, they only had two levels of approval, so RPA worked well, because once the practice location was identified, all those other decisions were automated and routed to those approvers. 

Royce Morse: So how would you recommend somebody go about finding the right solution for their needs?

David Glass: I spend a lot of time researching different things and looking at the IOFM site for upcoming webinars — at least one per month that I attend, or someone on my team if I think it relates more to their specific area, I have them listen.

00:13:03 

I try to continuously learn and different books and YouTube videos, ask questions. I'll admit here that we recently integrated five hospitals that had different policies for accounts payable, and we used ChatGPT to compare them.

Royce Morse: Oh, ChatGPT. How did that work? Spill the tea here. Tell me a little bit about how that worked.

David Glass: There was a book that I had found on Amazon that told you how to ask ChatGPT a question to illicit better responses, so I used that and said, "I need an accounts payable policy that incorporates all of these things," and it gave me an outline and identified differences between the policies that we were able to provide to our leadership team and say, "Here's these changes that each hospital will see as a result of this."

00:14:10                     

Royce Morse: ChatGPT is kind of a hot topic right now. It's trending, of course, and I've been wondering and thinking about how different organizations could use it. It seems on the surface that it doesn't really apply to a lot of what AP does, but you just named an interesting application for it that I never would have thought of.

David Glass: I've actually used it quite a bit. We've used it for our policies. We used it when we needed to create a visual basic script in Excel or Outlook to accomplish different tasks. We just asked it to create that script for us and it provides the instructions how to implement it. Even saying, "We want to do training videos," it'll draft a script for a training video for you.

00:15:04                     

Royce Morse: Wow. Would you say that that's something that you could implement without necessarily automating your AP process? That's just something you could do separately?

David Glass: You can use it to create more efficiency, so if you're drafting PowerPoints, use ChatGPT to help you draft your outline. Check to see that you're not missing anything. We are looking into using the conversational things of AI [when] we receive and email and having somebody want to know the status of an invoice, have the software be able to extract the info needed to look within the ERP system and respond.

00:15:56                     

Royce Morse: So that would be similar to when you're on Amazon and you get a chatbot and you're saying, "I didn't get my package," or whatever, and then it goes to look it up and tells you what's going on and then resolves it for you without a human being having to get involved, that sort of thing?

David Glass: Correct, yeah.

Royce Morse: I see. That's pretty cool. So if you're going to apply it here to AP and somebody's saying, "When's my invoice going to get paid?" ChatGPT can ask the appropriate questions and then supply an answer looking at the system and the payment schedule and all of that stuff and give people the answer without somebody having to be on the phone with a customer, or the vendor, right?

David Glass: Correct, yeah. We've found a few different tools that are conversational AI that have been very accurate, but it does come down to the training. You want it to use language that's similar to a person. You don't want to have that robot response.

00:16:56                     

Royce Morse: Right. It needs to make people feel as though they're talking to a person, even though they know otherwise; at least it's a more fluid and easily understandable way of communicating with the machine.

David Glass: And really different terminology, some folk will say "statement of accounts." Some will just say "statement." Having a tool that can interpret those different things.

Royce Morse: Yeah, that's a pretty cool application for some AI technology, again, that I hadn't thought of. If I can sum up what you're telling us, it's that AI or automation machine learning comes in different flavors, and in order to find a thing that's right for you, you've got to A) get your house in order, and B) do your research first. Would you agree with that?

00:17:52                     

David Glass: And I would add not to go all out. You don't need to start with AI and machine learning. Build your way up to it and really understand how it will aid in your process.

Royce Morse: Right, so take it incrementally. Don't go all in and then find yourself overwhelmed and having purchased a solution that you really don't know how to use.

David Glass: Correct.

Royce Morse: Awesome. Well, thank you, David. I appreciate your candor. Your conversation is interesting. I think people will come away a little bit more informed about technology and AI, which I know is a bit confusing for people. So thanks so much. I enjoyed speaking with you.

David Glass: Thank you.

00:18:38                     

Female Voice: Thank you so much for listening to the IOFM podcast. Remember to head on over to the Member Forum to discuss today's episode and provide ideas for our next one. And to stay up-to-date on IOFM's current events, both in-person and virtually, head on over to IOFM.com.