Why Machine Learning Matters

A conversation with Justin Bass, Crowe chief data science officer

4/18/2019
Justin Bass — Crowe chief data science officer

For non-data scientists, machine learning can seem as unfathomable as rocket science. Justin Bass, the chief data science officer at Crowe, understands the concept on a level few others do. He heads up the Crowe data science group, which includes all machine learning and artificial intelligence development and implementation. Also, one of the five degrees he holds is a master’s from Purdue University in aeronautical and astronautical engineering – that is, actual rocket science.

Machine learning is related to artificial intelligence (AI), he explains. Specifically, machine learning is a more efficient way to create AI systems, though it’s not always necessary. To illustrate the two concepts and how they relate to each other, he offers a comparison between automatic braking systems and self-driving cars.

Both of these things involve AI in that they allow technology, based on data inputs, to make intelligent decisions for us. Automatic braking systems, however, generally don’t need to be built with machine learning because they’re based on a fundamentally simple premise: measuring the distance between and the respective velocities of your vehicle and the one in front of you, and comparing that to the maximum braking power of your car.

With a self-driving car, though, there are a multitude of factors. The system has to figure out how to turn the car, deal with pedestrians, manage heavy traffic, and perform a host of other tasks. According to Bass, humans can’t truly conceptualize all the rules required for those complex environments. But machine learning can take the underlying data – for example, images of surrounding roadways – and develop rules for how that system should act when it’s on those roads.

machine learning new section

What are some of the common types of business problems that are solved by machine learning?

Justin Bass: It really depends on the industry. Here at Crowe we have a lot of focus on healthcare data. It’s actually one of the reasons the machine learning group got started here. Initially, it was very focused on the back-office issues of healthcare, like a hospital’s revenue cycle.

There were cases where these hospitals would have to have people issue refunds because accounts were overpaid, for example, and they needed to figure out who got those refunds and where to send them. That’s a repetitive but important task that’s made much more efficient through machine learning.

Then you have the other side of it, which is when machine learning accomplishes something humans couldn’t really do. People can’t really see all the dimensions of some underlying data set to understand what’s going on in an efficient way.

We generally can visualize two to three dimensions pretty easily. But when we say, “this transaction or contract can be measured across 30 dimensions,” and we’re trying to compare the information in two contracts across those 30 dimensions, as humans, we just can’t conceptualize something of that depth. So we have machine learning focused on anomaly detection to try to look at that kind of information.

We used to say that looking for anomalies or something suspicious was a hindsight type of activity, typically performed when we’re trying to “close our books.” We’d get the monthly and yearly data and do this hindsight analysis, but it would take three months to do. And we’d realize something happened previously that looks strange and we should try to do something about that now.

With machine learning playing a much larger role here, the latency time – or the delay from when we get a transaction to when we do a hindsight analysis – goes to zero. We can translate the findings into continuous monitoring. The transaction comes in, it gets scored, and we know the disposition right away.

machine learning new section

What is the Crowe approach to machine learning, and what are you hoping to gain from the firm’s groundbreaking work in this area?

Justin Bass: Let me start by explaining the three different approaches to machine learning that are general throughout the industry today.

machine learning new section

What are some of the Crowe products and services that have been enhanced by your team’s innovations so far? Specifically, how have they improved as a result?

Justin Bass: The best one to begin with is the healthcare example I talked about earlier. That actually started out as a consulting service. We had consultants create manual rules that answered questions like, “How do I issue this refund to the right person?” They would consult with hospitals on things like that for a fixed engagement fee.

Let’s say a hospital had 1,000 accounts to balance. Previously, our consultants would have enough time to get to maybe 300 or 400 of those. Now that we’ve built machine learning to do that as part of our revenue cycle platform, it can resolve 90 percent of those accounts – all without any human involvement. That scales that process a lot better.

Even within that platform, we have additional machine learning models that now focus on the denials side. So it’s a different issue where, for example, someone at the doctor’s office incorrectly entered the patient’s date of birth. As a result, the insurance company denies coverage and now the revenue cycle back-office staff has to go in and figure out why and fix the issue.

Now we have three different machine learning models that can tell you how to fix an account, how long it’s likely going to take to get paid, and who’s likely to pay. That wasn’t even possible before. These denial models look at millions of different dimensions. Again, we as humans can’t even come close to thinking about how all these things fit together.

machine learning new section

Many people might read your comments and think, “This is all interesting, but my company is too small – or too unsophisticated, or fill-in-the-blank – for things like data science and machine learning to really matter.” What would you say to them?

Justin Bass: The downside of machine learning right now is that it’s very expensive. If someone’s saying, “this is too complex,” or “we can’t afford it,” there is some validity to that because you do need to have a lot of specialties to build your own machine learning capabilities. The good news is that there are tons of programs in academia training the next generation of professionals to have the skill sets needed to do this kind of work.

We believe that everything will be powered by machine learning and AI in the future. What I mean by that is that every human task will be augmented by AI, making people more productive and efficient. Also, they’ll enjoy their work more because AI will take out the repetitive and mundane parts of their job. That said, that doesn’t mean everyone has to create their own AI and machine learning solutions. There will be plenty of opportunities to buy and make use of this stuff.

Few companies are in the area in which we operate – where we’re powering software with machine learning. We see some of the larger tech companies, along with a few smaller startups, coming up with more AI- or machine-learning-as-a-service, where you can throw your data in and get some result out. That also has some significant downsides. If you’re trying to optimize something but aren’t aware of all the underlying issues or which algorithms are appropriate for what you need, you can get a very bad result. You need to know how to create it correctly and understand how to prevent all the biases that likely will show up.

People’s approach to all this should depend on what they’re trying to accomplish. There are a lot of different levels in here. If someone says, “I just need a software product that maps X to Y for some process that I have, and I have some input data,” they might just be able to buy that software. They don’t necessarily have to care about what’s inside of that software. Do you care about how Google works when you type something into the search bar? Maybe, but chances are you just want it to give you the information you’re looking for. You don’t need to know that it’s powered by AI.

Contact

Justin Bass
Justin A. Bass
Chief Data Science Officer