The founders of Oversight Systems conceived of their analytics applications as intrusion detection software for business transactions. The use cases were straightforward: find anomalies in large sets of transaction data and report on unusual events. Things that didn’t make sense could be malevolent—such as instances of fraud—or opportunities for process improvement, such as complying with travel and expense policies and harvesting unused airline credits.
In an interview with Data Informed, Patrick Taylor, CEO of Atlanta-based Oversight, said the use cases have grown as data sets have become more complicated. What started out in 2003 as a set of tools that helped accounts payable managers detect fraud and ensure the integrity of accounts has become a more powerful capability as Oversight’s software, which the company calls continuous analysis applications, examines more data sets. Outputs from internal data warehouses, enterprise resource planning systems, XML data feeds—and more recently, feeds from unstructured data sets on Hadoop databases.
A marquis example Taylor cited was Oversights’ supporting the U.S. Navy in analyzing 150 terabytes for an accounting reconciliation process involving 40 payment and expense systems connected to the Treasury and other federal agencies.
Oversight plans to broaden its offerings at SAPPHIRE, the conference for SAP customers, on May 14 when it announces predictive analytics capabilities for its Oversight CTM (continuous transaction monitoring) in support of SAP’s HANA in-memory database, as well as a new mobile application.
The following are highlights from a recent conversation with Taylor.
On the meaning of continuous analysis in the U.S. Navy use case
“In that case, it’s basically every day,” Taylor said. “We get the day’s transactions from their various systems. So there are 40-odd systems involved in this. Can we then do our analysis at night, and then have the insights available for their people the next morning? Yeah, continuous, in that kind of talk, what is real time, right? It depends on the clock that you’re watching. So in their case, they were just kind of, they live in this daily fast world, so that’s continuous.
“There are other places where we do discuss it literally every five minutes, but that’s kind of on the extreme side. But we’re not, just to be fair, we’re not a flight control system [working in] real time. That is nothing that’s going to be within nanoseconds of the data gathering.”
On how Oversight’s technology works
The analytics provides actionable insight relevant to an organization’s business processes and workflow, Taylor said. He described a three-step step process:
1. Extract the data from ERP systems such as SAP and PeopleSoft, and from Internet-based data sources.
2. Map the data. Oversight’s mapper function puts the data into a data warehouse.
3. Analyze the data. The company’s collaborative reasoning engine, called Core, runs the analytics and finds the insights, he said.
Taylor said the next version of Oversight’s CTM software will use a more robust mapping function to tackle larger and more diverse data sets. “There’s lots of what I’ll call big data, unstructured data technologies out there. And useful information that can be pulled from those systems,” he said. “What Super Mapper lets us do is, based on what we’ve extracted from the ERP systems, or whatever your day-to-day business transactional systems, looking at what came in and driving out searches into an unstructured data source.”
On the indictors that make analytics queries valuable
Taylor said queries the application analyzes different indicators. “Let’s say that I see something happening with a sales order. Let me go see if there have been any relevant email conversations about that customer,” he said. “I need to have a seed for query or searching the data graph to know what I’m looking for. Otherwise, it’s just a big pile of stuff.”
“What we’re doing is, we figure out what the seeds for these queries are, based on the new data that’s coming in. That’s what Super Mapper’s doing. It is saying, ‘OK wait. I just saw something coming in about these 20 customers and these sales orders. Let me go see if there’s any relevant piece of information and how we’re rolling all of that into the overall analysis that we did.’”
Taylor said the analytics applies multiple indicators to examine transactions that it subjects to an integrity check, such as seeking evidence of a duplicate payment. “I can start adding in indicators that are also feedback from [various data sources]. You know, did I see pages and the fact of the presence of an email, anything that I can pull up? What’s interesting about that email is each of these indicators carries a weighting. … Coming out of this, you come up with some kind of a probability that we have found whatever a particular issue is.”
On how Oversight analytics differ from business intelligence applications
“We’re not a data mining tool, we’re an application of data mining capabilities that brings an insight to a worker,” Taylor said. “We can have this core platform and basically you put these configuration qualities, X number of configuration qualities that show our system where to get the data, what analytics to run, how the work flow works. You come up with a bundle of these and you can attack different problems. So we kind of have a platform and then we have application modules. So we’ve come up with a new module called Best Price.
“What [Best Price] is doing is every time that a purchase order is entered in it now someone needs to decide to approve it. How can I make that approval decision as smart as possible? Well, what we’re doing is going through and analyzing all the line items on a purchase order and looking back through history and looking at things like, ‘Where have I ordered a statistically similar quantity and gotten a better price?’ … When I got that better price was it geographically reasonably close to where I’m ordering it today or was it on the other side of the planet? What were the delivery conditions?”
Taylor added: “It’s basically trying to highlight for the person approving the purchase order, this is their opportunity to get a better price in any one of these particular [products], maybe by changing vendors, maybe by negotiating a better price. But giving them that insight to say, ‘Hold on, before I approve this PO I want you to go back and get us a better deal on this particular [product].’”
On SAP’s HANA
HANA speeds up the analytics process by keeping transactional data in memory, Taylor said. “Right now we have this ‘let’s go get data extracted’ kind of role. HANA changed all that. What’s happening there is, I can run the analysis as the transactions are happening.”
“Let’s go to this best priced scenario I just described,” he continued. “I can now pop up that ‘Hey, maybe we should order more’ message to the guy while he’s creating the purchase order in the first place. If someone’s creating a purchase order, putting it in for approval, we would have the ability to do all that analysis. Do all these stats, and pop on his screen ‘Hey, you’ve got to consider ordering more of this because the price has been increasing at an accelerated pace.’”
In such a case, Taylor said, data from email systems relevant to purchase order histories and data relevant to analyzing best prices would be in memory for immediate processing. “What HANA lets you do is really take that timeline way, way, way down. … You can do statistics and things like that right on an as needed basis. Typically, we would confuse statistical baselines as part of this extracting and mapping process and have them there for reference and analytics. With HANA you don’t have to do that step.”