Thought Leaders in Big Data: Interview with Patrick Taylor, CEO of Oversight Systems (Part 2)

Sramana Mitra: Are you positioned today as a fraud detection company, or do you have a different, broader positioning?

Patrick Taylor: Much broader than that. We are a continuous analysis company. Our software can operate on an ongoing basis. I need to be looking for this every day, and the sooner I find the problem, the better. This comes out of our intrusion detection roots. Intrusion detection is also a continuous capability. We are built to operate all the time and are continuously looking for things. When we find stuff, we then try to help people do something with those findings – to make decisions, to resolve problems, to address opportunities, etc.

SM: Let’s pick three of your customer use cases where you have delivered your value.

PT: At the large end of the scale would be a problem we address for the Department of Defense. There are two things we do for them. The first problem we solve is improper payments. The department pays somewhere around $350 billion a year to various vendors. In their case, they have multiple applications – accounts payable systems, if you will, most of which are home grown, and they operate in multiple centers of operation. What we do for them is look at the payment streams, and we start looking for issues like making sure they don’t pay the same bill twice. The simple answer is, “I am never going to pay the same invoice to the same vendor twice.” But it is not that simple. The result is: What happens if I have an invoice to IBM and the other to International Business Machines? I don’t have perfect vendor [match]. So, I need to be able to find those. Then what happens if someone transposes a couple of digits in the invoice number or slips a dash on the invoice number, trying to get it through because there is a rush to pay this vendor, and they don’t realize there is another payment pending. If you start exploring all the different ways that a payment can happen, those are the kinds of things we look for.

Then you start looking for things like, “Is this in line with the pricing terms I have on this particular invoice? Do they match up with the contract?” Let’s say the contract has an increasing discount as volumes escalate. You have to analyze that over time. These are some of the dozens and dozens of indicators we look for that answer the question if this particular invoice I am paying to this vendor is OK. We do that for them to prevent $1.5 billion per year of improper payment.

A second thing is an interesting and challenging problem. The DoD have to reconcile their books. The problem or the challenge they have is they had a handful of accounting systems and a handful of different payment systems. So, they need to reconcile among those systems. Life becomes complicated because one department buys and allocates the expenses to a different department. For example, a barge full of fuel comes into a port, and some of it goes to the Navy and some of it needs to be offloaded to the Marines. All of this has to be reconciled back with the U.S. Treasury. It is a mini-to-mini reconciliation problem. Historically, they had huge adjustments they had to make to the books. What we replace that with an analysis of all of their journal entries every day in all these systems, and we take all the numbers that automatically match from around 60% to 97%. We go through and figure out that “if I pair these three numbers with these two, they reconcile. If I pair these four over here and these two over there, they reconcile, etc.” It is a pairing problem that also deals with 150 terabytes of data. So, as I mentioned, we automatically match 97% and we give a confidence-based suggestion to the remaining 3%: “Here is how we think these other 3% match up.”

One of the things you need to know here is that you can get to a lot of precise answers with analytics. But then a lot of times what you also want to do is bring a human into the decision-making loop. There are pieces of context and information that humans can have that you just can’t see with systems. We believe we produce an insight – an analytic answer. We want a human to be able to easily understand how we got to that conclusion. Therefore, we use things like a plain language description that says: “Here is the reason why this software brings this kind of problem.” We give them easy access reporting data, and it is all along helping a human to make that decision. That is why I described the probability based answers earlier.

Another customer example – and we do this for a fair number of customers, especially in pharmaceutics and energy – is around the Foreign Corrupt Practices Act. This act is the idea of U.S. companies not being able to bribe or unduly influence a foreign official or anyone associated with a foreign government. That gets a little tricky. Like in the Walmart case in Mexico, where the way bribery happened was that Walmart was paying these small law firms to get their permits, and the law firms would bribe the local officials. That can also take the form of, I give my distributor a bigger discount than he normally gets, and then the distributor uses that money to grease the palm of a local official.

There are other places to look, and the things you look for are somewhat nebulous. What we did in those cases was try to apply our analysis to general expenses, the payment side, the sales orders, orders to cash, etc. There we are looking for a variety of factors that make you suspicious about the transaction. The law will tell you that if you do something that is normal and customary, it is OK. If it is something unusual, that is bad. We do a lot of checks to analyze the people and amounts involved and understand that at a localized level and number of people level so we can determine that. For example, people with higher titles might have more money spent on them. You can apply that same logic for discounts. The other indicators we look at are, for example, is this person a politically exposed person? That is going to add to my confidence that I have an issue here. When you start looking through time to identify actors that are repeatedly involved – so that you see patterns of activity – you want to understand if an actor is repeatedly involved in things that look to be suspicious.

This segment is part 2 in the series : Thought Leaders in Big Data: Interview with Patrick Taylor, CEO of Oversight Systems
1 2 3 4 5