As futuristic as the terms “artificial intelligence” and “machine learning” may sound, they are currently being used in the home and the business world. In the accounting profession, though, they could be better leveraged. Robert Schulte, founder of LumaTax Inc., joins us to talk about the benefits to accounting and finance functions, benefits for CPA firms and clients due to adoption of advanced technology, and areas in which artificial intelligence could be used more.

If you’d like, you can download this episode’s audio file. Additionally, you can follow us on iTunes, Google Play, or subscribe to our RSS feed.

View sponsorship and commercial opportunity details.

By: Bill Hayes, Pennsylvania CPA Journal Managing Editor

Podcast Transcript

Artificial intelligence and machine learning may sound like far-off concepts, but the reality is that they are not only being used inside your homes as we speak, but are also being used within the accounting profession on a daily basis. However, like any other burgeoning technology, there are many ways by which it can be better leveraged. Today, Robert Schulte, founder of LumaTax Inc., joins us to talk artificial intelligence and machine learning, namely the benefits for accounting and finance functions, challenges for CPA firms just embarking on a plan for artificial intelligence usage, and benefits clients experience due to adoption of advanced technology.

Can you give us a bit of context for the uninitiated on what artificial intelligence and machine learning represent, and how they're used in the business world or even in the home?

[Schulte] As we know, we hear these words a lot in today's society. Simplest form, machine learning, that's an application of artificial intelligence. It provides systems the ability to automatically learn and improve from experience without being explicitly programmed. That's a departure from what we're used to. Machine learning focuses on the development of computer programs that access data and large quantities of data, and learn from that data themselves. The machine learns from itself.

The most important piece of this, though, is that the learning begins with observations of data. As we've heard throughout our careers in computers, GIGO – garbage in, garbage out – same thing occurs for artificial intelligence and machine learning. If the data isn't appropriate, it will give you the wrong learnings. The primary aim is to allow computers to learn automatically without human intervention, but oftentimes you have to have supervised machine learning or semi-supervised machine learning, which puts a human in the loop.

It really is powerful because it gives systematic decision-making and learning processes. If we're seeing today some business use cases in accounting specifically, we're seeing the assignment of transactions to specific GL accounts based on prior assignments, meaning they look at a vendor or they look at some sort of information on the payable or the receivable transaction, and they can assign it specifically to a GL.

Sometimes they don't get it correct and they'll ask the human and when the human tells it that this vendor goes to this payable, the system will learn from that. In the use case of the home, we're seeing it ramp up in just light speed because of Alexa. Alexa is learning, voice is a big part of machine artificial intelligence, the smart home system, and travel optimization when you use a Google map. It's a secular growth change. It’s really going to have an impact on the accounting community.

Speaking of the accounting community, how is artificial intelligence and machine learning currently being used specifically in the accounting world, and what are some innovative areas in which it could be used more?

[Schulte] I've been in the accounting community and FinTech specifically – or financial technologies we refer to it – which accounting sits in. The fundamental business problem that AI and machine learning solve is most of those mundane tasks that used to take humans time to process GL transactions, do reconciliations, posts to the payables, post to the receivables, do fraud detection, transactional analysis, all that stuff that used to be done by people. That can really be a misappropriation of people's time because it's such a mundane task. It's going to be done by artificial intelligence.

You've already seen the advent, especially for CPA firms that provide client accounting services, of machine learning and artificial intelligence and new software that helps you categorize transactions, helps you post them to the GL.

But as time goes on, we're going to see more and more robust insights from this data and transactional analysis to where you can actually predict whether transactions compliant with a certain set of rules or laws or regulations before it even occurs, whether there'll be compliant or noncompliant. But the critical factor for an AI platform, and any kind of software you're going to buy cloud-based or not, is for it to perform exceptionally well, it has to have processed tens of millions of transactions to have a high level of certainty and a prediction rate.

So that's one of the things that we're seeing more and more is that companies that are using it in the GL process or the closing process, or even in the tax process like we provide in sales tax, they really have to have a lot of data at the transactional level and learn cross-platform not just from the specific client, but all clients. So that they can see the trends and the analysis from those, develop insights, and return those insights back to their clients. It will give them a better service offering instead of the time they spend doing those data entry points. Now, they're going to be telling the clients, "Hey, this is what you should do going forward."

What would you say are the sort of headline benefits, the main benefits, of artificial intelligence that accounting and finance functions can leverage better?

[Schulte] I can give you a general and then a very specific. Generally, it's those mundane tasks that those of us are accountants, and we grew up in the industry and we remember the 13 column pad for Christ's sake. We have the ability now to have machines just look at millions of transactions in nanoseconds and just start to categorize a sign, put them in general ledger accounts, flag them for further review, ask humans for specific information, and supervise machine learning models. Really, it's taken away the mundane task.

Then it's also that rich data set that we use in the accounting world, especially if you have it at the transaction level, not just the GL balances, but actually the detail of the invoice and the skew number and the ship to, and the ship from. Now with GPS locations and lat longs and all these different things that you can bring disparate pieces of information together to enhance the feature set of these lines, you really are giving yourself the ability to process transactions, deliver the recording of them and the reporting of them in an automated way. Not only in an automated way but in a more accurate and robust way.

Those are the beginning benefits. But it's in that transactional analysis that removes human error and processing from those mundane tasks that eventually allow firms, but also the firms’ clients to drive revenue up and costs down, to increase margin improvement and to have an efficiency and effectiveness of output. You're going to get improvements in that too.

But most importantly, I think what you're going to see in the near term is it detects fraud and it detects misclassification. You have to understand that we've seen these major market meltdowns with the Enron catastrophe and the WorldCom catastrophe over the years where accounting irregularities were hidden or found. The machines won't allow that to occur. The machines, if they look at every single transaction, aren't going to do a population, they're going to do an actual basis examination of the transactions, and they're not subject to greed or fraud like people are unless of course the systems were programmed to do that.

But even that you can build in checks and balances on machine learning to make sure they're following the proper rules and that somebody hasn't overwritten your code. I think that in the early stages of it, the most beneficial part of it is going to be to reduce the man-hours to produce an output. But ultimately, the insights, the derivative data, are going to be your most important outcome from this data.

In the era where we have always told people what they've done in the reporting worldwide, we produce financial statements, and we tell businesses what they've done. We're going to get to a spot now where we're going to tell them what to do. We're going to tell them what to do with a certain level of predictability and accuracy that's going to be far beyond what we could have done if it was just people entering those numbers. So that's where I think the biggest trend and analysis will be as providing insights to the client.

What are some of the challenges CPA firms and business are going to confront when they first start experimenting in artificial intelligence, and how can those challenges be overcome?

[Schulte] I'll take us back to a time when the accounting community, we transitioned from the 13-columnar pad to what was Lotus 1-2-3, or SuperCalc, or the early days of the spreadsheet. There was this sort of hysteria, if you will, around this transition and what's going to happen to my job, and am I going to still be in work? In that era, when the personal computer first came into the business world, we were all worried about that.

But at the end of the day, it created whole new industries. And I see the same thing happening with machine learning and artificial intelligence for those firms that embrace it and get in front of the learning curve instead of reactionary and get behind the curve. There's obviously tons of strengths and weaknesses with using new technology.

Some of the strengths are you get the improved efficiency, the reduction of labor, more insights, but there's some weaknesses too. And some of the weaknesses are born through because of psychological impediments to adoption, just the fear of change or lack of data for sophistication in your machine learning process. As you transition to a new process, if you don't use the right vendor, and that's a big caveat is the right vendor selection, and they don't have enough data to really provide you robust tools, then the systems are going to take time to learn.

If you think about a firm that does mostly tax work, tax works really summarize data. It balances in GL accounts. It's not the detail and you don't really make a lot of decisions effectively if you don't have the details. So I think the challenges for CPA firms will be making sure they have a lot of data, making sure the vendor selection of the companies they work within the cloud computing and artificial intelligence and machine learning, the plug-ins, the business intelligence systems, you vet those very carefully.

Then, the challenges are going to be the integration of the human and the work insight, which is you're going to be creating new systems, intelligence systems, and how do the humans and how do the machines interact? It's kind of an odd way of saying it. So, finding the best way to supplement the interaction between humans and machines, especially in a supervised or semi-supervised machine learning area.

We want the humans to spend more time on the derivative insights and the value-added tasks to deliver the client service rather than the repetitive processing tasks. That'll be an area. Then the final thing, I think the challenge is the learning model, and machine learning is not going to work if you don't have an appropriate learning model.

A learning model requires data. As you get data and the data gets better and better, and the system is told what to do from humans in the loop of the decision-making, the humans can come out a little more as the system learns. But it's vast quantities of data and making sure you've enhanced the data so that the machine can make the best decision based on its rules and its algorithms. And if you don't have that data, it's not going to be an effective first pass at using machine learning.

And so the accounting firms are really going to have to rethink the way they store data, protect data, what systems they use to handle that data, what level of data they're comfortable importing. All those things will be really the challenges of the CPA firms in the future. Probably the best way to address that challenge.

It may be counterintuitive to most firms but you need to hire somebody who is gifted in systems, systems design, artificial intelligence, machine learning. That's a heavy headcount financially, but it's a headcount that you need to have a single port of reference in a CPA firm or multiple people to drive the approval process of vendors that work closely together and whose systems can communicate clearly together and not have a bunch of ad hoc purchases. And then try to stitch it together. That will be critical for the accounting community.

What sort of benefits would be expected for the clients of firms with artificial intelligence experience?

[Schulte] It's fascinating to see right now. We actually use machine learning, or artificial intelligence, to determine whether or not sales tax transactions are going to be compliant with certain laws and regulations. It's fascinating because there's so many differences across industries, and the use cases across industries, and whether it's a use tax audit or a sales tax audit, whether it's the payables or receivable cycle.

What ends up happening is, outside of the deliverable of filing a sales tax return or providing some sort of a reporting mechanism for our clients, we get to see, in basically real-time, transactions that would fail to comply with the state law. It might comply in Michigan, but it wouldn't comply in California. We get to see that instantly, and that gets transmitted back to the client.

The accounting community through this insight of this valuable data, and also taking that data and anonymizing it so that you can compare it to other clients similarly situated in the same industry codes and the same tax and jurisdictions, or the same European Union, or whether it's a 409A risk or an SCC regulation or any of those things, you can program and abstract out these rules. So that the benefits of using this data multiple times … you're going to use the data for accounting and reporting, but you can also use that very same data to give them compliance reports. Or maybe they have to have an audit done by the franchise or so they see that they were paying the right franchises.

All these little things that can be done in real-time now with that same set of data, we can now provide valuable insights to them. So, when you're looking at it like that, especially firms that provide client accounting services because they have that rich data, being able to provide the critical insights to clients based on their own data. Then bouncing that data off, anonymized data of people similarly situated to see what they do and see where the anomalies are really allows the decision-making capability. It puts data, and it puts the insights from that data, at the forefront of the client's decision-making process, which really it's not something they would normally do. They can't compare with other company's data because they don't have access to it. It's an exciting time.

What sort of financial or time commitment is needed for a firm to thrive in this area? Is it something that's manageable for smaller firms?

[Schulte] That's the beauty of cloud computing. Really, if you think back to the days when we had enterprise software, and that was the only thing we could do, and we had these huge Oracle implementations or these huge Microsoft implementations, and it was multi-million-dollar, multi-month projects, that's gone. Today, you can just literally go to a website and you can sign up and start working right away.

In a cloud-computing environment, you just have to be smart about the financial and time commitment that you're going to budget. The time commitment's more important because you want to make sure you do your due diligence on whether they're SOC 2 compliant, whether they have ISO standards they have to reach with their data privacy, where the server's located? Is it the Amazon cloud? Is it the Microsoft cloud? All these things that are important to protect data first and foremost.

But the time commitment really is the most important part of it. The financial commitment you can get subscription services, you can get … most of them have a monthly or an annual fee. It's pretty simple to get in and out of a contract. Where the time comes in is in that single point of reference. The company that's going to make decisions on this has to create an educational effort cross-board for the company, and who's going to use it.

There should be an owner of each product, depending on what kind of service that product delivers. In sales tax, we want it to be someone who's very gifted in the SALT realm and knows indirect taxes. We don't want an income tax guy or gal coming in and managing a sales tax application. That's a critical effort: doing an educational effort across the company, having a plan, having a strategy, having a goal. Outline that goal and explicitly discuss that goal so that you can actually measure whether you're hitting the commitment that you've decided to embark on, whether you're actually hitting those goals.

If you don't do that, if you don't have a goal, you're just floating around in the cloud, hacking away at software. Define the goal, define what you want to see, the metrics that are going to matter most to you on achieving that goal, educate the team, educate the company, and then you have to have a real commitment to adoption. It can't be one of those things where, oh, it's wishy-washy. It has to be a commitment.

If you do that and you have a driver inside the CPA firm, whether it's a big firm or a smaller firm, you will be successful. You will improve your operating capacity. You will improve your advisory services, and you'll give better advice to your clients.