Ways to Use AI in Accounting and Finance 

Please note: The “Net 30” podcast is designed for audio consumption. Transcripts are generated using speech recognition software and may contain errors. Please check the corresponding audio before quoting in print.

CHRIS ELMORE  

Hi, my name is Chris Elmore. I am AvidXchange’s Chief Evangelist and I’m the host of the Net 30 podcast. On the Net 30 podcast, we meet with industry leaders to unpack problems and solutions and talk about innovations that are impacting financial professionals. The best part about this? We’re going to do it all within 30 minutes. So let’s get into it. 

0:32  
CHRIS ELMORE 

All right. Welcome, David Tareen. Part three of three. I feel like we’re best friends at this point. So David, you are the I agree. Yeah. We can hang out. So, so David, you’re the senior director of product marketing at AvidXchange. You got a long history with artificial intelligence. I tell you what, this is part three, and we’re going to kind of go deeper into some misconceptions. 

We’re going to talk a little bit about the headline is there’s rise of lawsuits. There’s should I use public-facing artificial intelligence? But. In this episode, we’re going to kind of debunk the misconceptions of artificial intelligence, and we’re going to base it on AvidXchange’s 2024 trend survey that we conducted in September of 2023 where we asked 500 finance executives at middle market companies about their use of AI. 

We’re going to go deeper. I’m going to, I’m going to kind of tell the folks a little bit about the questions that we’re going to ask today. But, let me get to my guest. Kind of tell folks why you’re the person talking about this and what your experience is? 

1:47  
DAVID TAREEN  

Once again, David Tareen. Very happy to be here. I’ve spent a good amount of time in artificial intelligence technologies, working with startups, with more established companies and quite frankly, I think I love talking about AI, but not at the high level. Let’s get really specific into which technology we’re talking about because I think that’s where the interesting conversations happen. We actually have an opportunity to learn some things. So yeah, excited to be here.  

2:15  
CHRIS ELMORE  

Just curious, here we are in the third part and I’m finally asking a curiosity question. But how did you kind of procure your information about artificial intelligence?  

2:25  
DAVID TAREEN  

Yeah, a lot of, on the job – and we’ll actually talk about how the dynamics are changing on how you used to learn about technologies and how people are doing it today. My journey has definitely been non-traditional. I’ve worked for a lot of data with a lot of really smart data scientists and multiple different technologies of AI. We’ll talk a little bit about multimodal, what that means. But yeah, a lot of just practical knowledge from my work experience. 

2:53  
CHRIS ELMORE  

There you go. All right, here’s what we’re going to talk a little bit about today. We’re going to talk about the common misconceptions of artificial intelligence in accounts payable. That’s AI in AP. Lots of initials. And we’re also going to talk about the artificial intelligence revolution as it pertains to the accounts payable process. 

And then, I think this is, we’re going to end big, David, we’re going to end big. I’m going to have you answer the question to help people understand – is artificial intelligence right for their organization? That’s what we’ll land on. So first things first, breaking down some of the common misconceptions. By the way, if you haven’t heard part two and one, you should go back and listen to that. Shouldn’t stop you from listening to this one, but we kind of touched on some of the misconceptions, but what’s out of the gate – when I say misconceptions, what do you think of? 

3:44  
DAVID TAREEN  

The biggest one out there in my mind, and there’s some reality to it as well, is the feeling that artificial intelligence, the algorithm, once I start using it, it learns, it’s going to make decisions, it’s going to take over my job. 

That’s a big one that we see a lot of the times. And there’s so much packed in that one sentiment, in that one sort of generalization that I think it requires us to dig a little bit deeper. There is short-term impact on people’s employments and it is going to have an impact on certain parts of the market.   

Certain parts of roles are going to get affected. However, what we’re also seeing is that there’s new types of roles that are coming up. 10 years ago, data science was like the number one thing to go off and do. That’s right. Now it’s actually something different. So we’ll continue to see this evolution in roles. So yeah, I think that’s the biggest misconception absolutely requires us to unpack it. And get into the details.  

4:43  
CHRIS ELMORE  

The headline is artificial intelligence is not human. And kind of the, the subtext, cause we talked about this in the last episode about this job title of the prompt engineer. I got a definition of prompt engineer. It is probably was artificially intelligently generated. But you know what? Popular culture has propped up this notion that robots are taking over the world. We see this in our sci-fi movies all the time, but is that true? 

5:11  
DAVID TAREEN  

Not in the near or the long-term future in my mind for that to happen. And this is David Tareen’s view of the world. Right now, when you look at an algorithm, it’s really good at making decisions. Based on the data that it’s been trained on.  

Humans, people are not like that, right? I can learn how to drive a car and I can probably work my way through driving a motorcycle because I can apply learning from one thing to a completely different thing. That’s creativity intuition. Comes so easy to all of us, but it does not to come to machines and I think for us to get to what’s called general artificial intelligence, which is what sci-fi popularized that it’s going to take so much compute power, so much time. I don’t think it’s going to happen anytime soon. Yeah.  

6:00  
CHRIS ELMORE 

So the prompt engineer, it’s the process where you guide generative artificial intelligence, generative AI solutions to generate desired outputs. 

6:28  
DAVID TAREEN  

I think it’s a repeat of skills that used to exist in previous technologies. So I’m sure a lot of the folks have heard about SQL. That was sort of a language that we used to query databases. Oracle used to use it and a lot of other databases use it as well. But you would have individuals who would specialize in SQL. And what that job really meant is that you are the person who can ask the questions in the right way to this big data set and get the right answer.  

The prompt engineering is, in my mind, a newer type of role that pulls in some of those skills. But the big difference is where in previous roles, those SQL engineers, those were highly technical people. The big shift here is prompt engineers can be domain experts as well. And what that really means is that if I’m a person who knows a business and I can query really large sets of data just by using my natural language. I can speed up my processes. I can get to much better outcomes that are relevant for my business right now. 

In the old world, you needed to query some database. You got to talk to a whole bunch of people, maybe follow a process. And it just took so much time and who knows what kind of result you would get.  

7:42  
CHRIS ELMORE 

So in the survey, we took one step further and we asked those folks in the survey that aren’t using artificial intelligence – why? And 37% said lack of understanding, 33% said they were hesitant because of regulations and safety, and 25% said the cost of implementation, 5% other. So, lack of understanding. By the way, the 33% that says hesitation because of regulations and safety, it kind of makes a lot of sense, it kind of makes a lot of sense. What’s your reaction? 

8:21  
DAVID TAREEN  

In the accounts payable in the finance, financial services space? Absolutely, because it’s so prevalent with regulations and compliance needs. There’s a lot of contracts and we have some of those conversations with our customers. There’s AI algorithms that would be able to absorb contracts and be able to decipher them and understand them using natural language processing. And then from a process standpoint, flag areas when you are at risk for contract breach. So yeah, there’s absolutely hesitation because of regulation, but then I would say there’s a lot of artificial intelligence tools that can help you with regulation as well.  

8:59  
CHRIS ELMORE  

The undercurrent of that is what we found kind of in the survey is that most of the organizations are establishing artificial intelligent technology in house instead of using public-facing AI tools like ChatGPT. 

9:15  
DAVID TAREEN  

Now we’re reaching into the concept of multimodal artificial intelligence, which means, “Hey, don’t just use machine learning for every question that gets thrown at you.” Use a rich set of artificial intelligence technologies because that gets you to a better result. I know we’re going to talk about that later. I’m kind of excited about that topic.  

9:33  
CHRIS ELMORE  

Let’s get into it. So let’s talk about how artificial intelligence revolutionizes accounts payable. So in, in what way to AI tools change the, the AP process? 

9:51  
DAVID TAREEN  

I think from top to bottom, but at the highest level, there is so much potential for process improvements. One of the things that AP professionals go through all the time is they have rules and regulations in terms of approvals and those absolutely have to be followed. From a process improvement standpoint, an algorithm that’s looking across your workflows, looking across – let’s say for in this example how invoices are approved, and it can absolutely make suggestions on how you can accelerate, how you can streamline. 

Those are really big questions while staying within those regulations and compliance pieces that we talked about, we have some conversations with our customers on those. And then more on the tactical side, a big cause for inefficiency in AP automation is am I finding the right expense code? 

10:30  
DAVID TAREEN  

Am I coding everything to the right GL level? Am I doing the right PO match? So artificial intelligence is huge there. You no longer have to memorize 300 or however many different expense codes you have because the tool can take a look at the expenses that are coming in and then make a recommendation that will help you make a better decision. 

I think this belongs to, for example, if you’re providing yard services for a large community of homes, um, the artificial intelligence tool can tell you, I think this belongs in this expense code and now you just have to approve it. Versus having to actually think of it from scratch  

10:57  
CHRIS ELMORE 

When you’re pushed a little bit on specific tasks within AP departments, what’s kind of your laundry list of specific tasks? 

10:58  
DAVID TAREEN  

Yeah, I think PO matching is a huge one, especially in some industries where you have a single invoice and that has multiple different areas that are covered for a tool to go in and be able to make recommendations. And like we said coding, your accounting codes, expense codes, GL codes, all of that. I think those are tactically a big cost for inefficiencies and artificial intelligence is being used. A lot of our customers use it right now for those tasks.  

11:33  
CHRIS ELMORE 

All right. So topic number three is AI right for your organization? I wasn’t wanting to sit on this one for a little bit because I think this is the part that rubber meets the road – use whatever kind of saying you want here, but people are still trying to figure out how to use this stuff. 

So specifically what features should companies consider? When evaluating AI tools? 

12:00
DAVID TAREEN  

I think it’s going to be different for a lot of different organizations, but there’s going to be some commonalities that thread each one. Foremost in my mind, data privacy concerns are going to be are going to be top of mind. The fact that artificial intelligence can consume data so fast, so broadly, and then make that data available to so many different places. You’re no longer confident that, okay, I put all of my assets into that one office and I locked it up and it’s done. The internet’s connected to everything. 

So I think personally identifiable information within your data has huge risks for privacy concerns and even regulation as well as how your brand is perceived. So data privacy and putting some really tight controls around that, in my mind is absolutely the right question to ask when you think of is AI right for my organization, it should be at the top of your list. 

12:55 
CHRIS ELMORE 

What’s the ramifications there?  

12:56 
DAVID TAREEN  

Some best practices in my mind is you think about artificial intelligence. It also has to learn just like we people need to learn an algorithm needs to learn as well. So how does that happen? But you think about services like ChatGPT -one of the ways that they continuously educate themselves and make their models better is every time a query comes in, they’ll use that data to train the model. So if I am asking questions, I am probably not going to mention our company’s name or any company’s name because all of that information is being taken to make these algorithms better and that’s in my mind, a cause for concern from a privacy and intellectual property protection standpoint. 

13:41  
CHRIS ELMORE  

Do you think people will have any recourse maybe now or in the future where they load in their code and then the artificial intelligence because it’s public domain has the code and then it gives the code to somebody else like a competitor. Do you think there’s going to be any recourse for that?  

14:59  
DAVID TAREEN  

It’s just so hard to A, detect that that has happened and B, prove how it happened and then three – discover the intent, which are three really important things for recourse. So I think it can be done, but it’s just practically so difficult to do. I think where this evolves into is when you think about how we started trusting applications for anything. I use Excel at work all the time. Excel has some really confidential information, but I’m okay with that because all those Excel instances live behind my firewall at work So it’s protected. It’s not like I’m putting those out in the public domain.  

I trust applications like PowerPoint because there’s a lot of intellectual property there, but that’s okay because it’s behind my firewall. The way artificial intelligence evolves and becomes more trusted. Is as those tools are baked into the applications that you use every day – Microsoft, for example, if you’re on the Microsoft ecosystem, they’re coming out with the artificial intelligence, they call it, the branding is Co-Pilot, that’s going to be built into Excel and into PowerPoint and into Outlook. So now suddenly it’s going to become part of my workflow because all of that is going to live inside my firewall. 

Google’s doing the same exact thing. They’re baking Bard and some enterprise implementations of that into their ecosystem for the back office so that anyone who’s using those email services and those Excel spreadsheets or the Google flavors of that are automatically going to have these AI capabilities. 

So I think that’s how the evolution starts to happen.  

15:30  
CHRIS ELMORE  

There we go. All right. So. AI implementation, best practices, what’s at the top of your list?  

15:39 
DAVID TAREEN  

I think in episode two, we talked about the biases that exist in data and the black box nature of artificial intelligence. I think those have to be top of mind as well, but that absolutely does not mean you should stop. What that does mean is in episode two, we talked about it’s a black box. You can get the good decision, but you can’t explain why that’s a good decision. What that means, I think for a lot of customers is, “Hey, let’s not make decisions on who we’re going to go off and acquire next to an AI algorithm.” 

But if we’re doing some internal analysis, yeah, let’s try it because it’s low risk. It’s data that we can verify and we can weigh in from other decision points as well. So that’s the journey that I think you have to start on. But bias in the data and the black box nature of artificial intelligence has to be top of mind as well. 

16:33  
CHRIS ELMORE  

So best practices on getting this stuff implemented. I have a list and hopefully you can react to it. The first one. I think is the most interesting. Position AI as a partnership. And then the second one is upskill, reskill, embrace multi modal AI. Because you’ve been desperate to talk about this. And provide bias training. So those, those are kind of the four of implementations. What do you want to talk about on these four? 

17:03  
DAVID TAREEN  

The one that’s missing in my mind is strong leadership. That absolutely has to be an ingredient because oftentimes what we see is artificial intelligence projects come up and you’d never find that champion who’s going to push it through, who’s going to align the skills, the data privacy, cleaning up the data, finding the algorithm, finding the problem and moving the team forward. That requires a champion pretty high up in the organization. And I feel like that’s such an important ingredient. But multimodal AI is a close second. It’s my favorite. This word is – it’s a big word. 

Some people call it multi-modal, some call it multi-disciplinary artificial intelligence. What it really means is that, “Hey, one technology is not going to solve your problems.” You have to apply a multitude of technology. So think about a person. The way we think is when we’re driving a car, we’re watching the road with our eyes. That’s sort of computer vision. We’re listening for any noises that should drive our attention. That’s another sense that we’re using. We’ve got passengers in the car. If I’ve got my parents in the car, I’m probably going to be driving a little bit slower because I want their comfort, right? So there’s machine learning in there. So I’m applying all these different technologies to make decisions on how I’m going to drive a car.  

And oftentimes when people latch on to, let’s say, machine learning, they say, I want to apply machine learning to this problem. You have to apply machine learning, and deep learning, and computer vision, and natural language processing. You just get better decisions. So absolutely bring that in.  

Think about it as a multi-modal, multi-disciplinary type of approach. It’s just going to get you better results.  

18:50  
CHRIS ELMORE 

It’s fascinating, David. I feel like this is the third of a three-part series, and I feel like we’ve just scratched the surface. I agree with you. But I want to say thank you David Tareen for taking time out and giving us some wonderful information about artificial intelligence, tons of takeaways. I’m sure folks are getting a bunch out of this. So thanks a lot for your time. 

19:04  
DAVID TAREEN  

No, thank you for yours. I love these conversations, love when we talk to customers about all the possibilities and hope that goes on. Thank you.  

19:10  
CHRIS ELMORE  

Thanks for listening to the Net 30 podcast presented by AvidXchange. I desperately need to know what you think about this. Leaving a five-star review would be fantastic. You can subscribe to the channel too. And, oh, by the way, while you’re waiting for the next episode, head over to avidxchange.com to learn more.  

One more thing. If this conversation has somehow piqued your interest, which I really hope it has, about accounts payable automation from AvidXchange, there is an additional link to get a demonstration of our solution. I always say it shows way better than it tells, so click on that link, fill out the information, and someone will be in contact with you.  

So thanks again for listening to the Net 30 podcast, and we’ll see you really soon.