Controllers Council recently held a panel discussion on Transforming AP Using AI, sponsored by BILL.
BILL is a leader in financial automation software for small and midsize businesses (SMBs). Our expert panelists are Eitan Anzenberg and Joe Fleischer. Eitan Anzenberg is chief data scientist at BILL. He has a PhD in physics from Boston University and a BS in astrophysics from University of California Santa Cruz. Joe Fleischer is a webinar content marketing manager at BILL.
Following are key takeaways to this discussion. If you are interested in learning more, view the full webinar archive video here.
What is Machine Learning? What is Deep Learning?
In the modern age, and now that AI is in the forefront of a lot of people’s minds, it’s good to think about how we would really categorize these types of techniques. That was how I would consider machine learning or really learning from data. So, machine learning is a technique that learns from data.
We generally apply a machine learning algorithm on a dataset that it trains on the dataset. So particularly learns from the data, learns patterns, and generally from large amounts of data. And this is where really the advent of AI has exploded in the last 10 years is in the compute expansion of compute processing, but also in the expansion of the data collection that’s out there. These machine learning techniques are data hungry. So, it’s used to train on the techniques, to make a machine learning model is to apply a data set, large data set and to train on that.
Deep learning is a subset of machine learning. Deep learning particularly refers to neural networks. And these as well have been invented decades ago, I think from the eighties, so the late eighties, early nineties, as neural networks, as algorithms that in a sense very simplistically mimic how a true neuron, so a neuron in person’s brain for example, how it behaves in nature.
What challenges with processing invoices can AI address most effectively?
Whenever you want to release an AI feature, machine learning feature, to your customers on first pass, it might take time for them to build trust in the system. I don’t know if anyone in the audience has a Tesla car or any car with self-driving technology. You may not just flip it on day one and put a blindfold on. You’re likely testing out seeing the capabilities first. And we at BILL have gone through that journey. And we’ve really learned the pain points and we’ve really learned where we can help our end users automate and help them out as much as possible because that’s what we’re in the business for, is simplifying these workflows for them.
So, where we can apply AI specifically. In this automation, we process tons and tons of invoices and receipts monthly in the numbers of over millions, millions of invoices, and receipts per month. And that’s a large volume of different types of invoices. Some are international, some are different languages, some are little like people might hand write a contract and they’ll upload that through the system.
We’ve put in place very secure methods for processing invoices so that there’s no personally identifiable information.
What considerations should finance leaders keep in mind as they determine how to use AI in AP?
First thing, AI is data hungry. There must be a sufficient number, efficient amount of data to apply machine learning and then even more data to apply deep learning.
There’s also a consideration of costs, even with generative AI. That’s very costly. But even doing some smaller models, even the deep learning models we’ve done and BILL, there’s always a cost consideration. Do we want to build this in house? Do we want to work with third party vendors? And then which ones do you choose because so many startups out there that provide these services?
Third one, very much in the forefront for us, security, privacy, ethical reasons as well. When you think about a company like us, we can’t mess around with customer data.
What is the difference between using AI to process invoices and relying exclusively on traditional methods such as optical character recognition?
- Traditional methods don’t generalize well for processing both simple and complex invoices.
- AI is a powerful technique that over time learns to recognize patterns from a large amount of data.
BILL has developed powerful deep learning algorithms that analyze data and images from invoices to help automate AP workflows.
What is the advantage or what are the advantages of employing a deep learning approach to extract information from invoices?
If you have enough data, I can’t stress enough how powerful a technique deep learning is. We’re dealing with the invoice as a high-resolution image and then having the deep learning algorithm tell us what to do with it.
To view the complete webcast, download full webinar here.
ABOUT THE SPONSOR:
BILL (NYSE: BILL) is a leader in financial automation software for small and midsize businesses (SMBs). As a champion of SMBs, we are dedicated to automating the future of finance so businesses can thrive. Hundreds of thousands of businesses trust BILL solutions to manage financial workflows, including payables, receivables, and spend and expense management. With BILL, businesses are connected to a network of millions of members, so they can pay or get paid faster. Through our automated solutions, we help SMBs simplify and control their finances, so they can confidently manage their businesses, and succeed on their terms. BILL is a trusted partner of leading U.S. financial institutions, accounting firms, and accounting software providers. BILL is headquartered in San Jose, California. For more information, visit bill.com.
