LAS VEGAS – The size of your data does not matter.
What matters, according to SAS CEO Jim Goodnight, is that you can analyze however much data you have in a manner of seconds. That allows quick decisions, multiple iterations for modeling, the ability to run more complex algorithms and a boost in productivity across the organization.
“With all the non-linear models, you have got to iterate. And that’s where keeping the data in memory really pays off,” Goodnight said. “There are problems… that really don’t have that much input data, but yet it generates so much internal data that if you can keep it in memory you can just make that job fly.”
At its two industry conferences, Analytics 2012 and Premier Business Leadership, SAS showed off its new High Performance Analytic (HPA) server, an in-memory analytical platform that uses massively parallel processing. SAS announced this week that the HPA server now accesses data from with Hadoop and runs on commodity hardware. The server has run with Teradata and Greenplum systems since it was announced in February.
The server is not a database; it doesn’t use SQL, which Goodnight said isn’t capable of doing the advanced analytics enterprises are using today for fraud detection, risk management and predictive modeling. It runs the same advanced statistical jobs that traditional SAS runs in the same coding languages, the latest of which is called DATA Step 2.
Mark Pitts, the director of data science, solutions and strategy for UnitedHealthcare, said the first day eight weeks ago he received his HPA server build on Greenplum hardware, he had a job specially cooked up to test the system. It was running in four hours and 18 minutes on his Unix system; it took 10 seconds on the analytic server.
“That was my a-ha moment,” he said.
Since UnitedHealthcare began using the SAS analytic server, Pitts said he has been able to do several iterations of projects in a single day, testing out different models and algorithms to try to improve customer service responses.
“HPA allows us to accelerate that trial cycle,” Pitts said. “What it allows me to do is do more cycles. Instead of doing one cycle in four hours, I can do maybe 20 or 30, or I can try several more modeling techniques. It improves my productivity, and it allows me to do it with bigger data. Instead of using samples, I can use the whole dataset without breaking anything.”
UnitedHealthcare has 75 million members and hundreds of millions of client interactions each year; Pitts’ team is using predictive modeling to make those customer service interactions better.
SAS is in the process of rewriting most of its algorithms to run on the MPP system, and Goodnight said it will roll out several more of its software packages that are optimized for the analytic server about every six months. Tuesday, SAS announced products for fraud protection and anti-money laundering, marketing optimization and real-time event stream processing. A high performance customer intelligence suite is set to arrive in the first half of next year, Goodnight said.
Goodnight and SAS executives also showed off the company’s new Visual Analytics offering, which is built on top of the analytic server and allows analysts and business users to build charts, graphs and other visualizations. Goodnight said customers had asked for a better user interface, and the front end of Visual Analytics is the new look of SAS.
Randy Guard, the vice president of sales development product management at SAS, said the visual analytics suite allows analysts to do data preparation and some exploration before creating more complex models. The new software’s user interface is approachable enough so business users who can’t code in SAS to begin to explore the data as well to provide hypotheses for the analysts to explore.
“If the data is going to be in this in-memory environment to do analytics, you might as well visualize on it,” Guard said. “Part of [the visualization system] is built for discovery. It’s drag and drop, you see what comes out, and then maybe you want to do some more exploration.”
Guard said that SAS recognizes the importance of publishing the results of the discovery, and the visualization reports are currently available on the web and on iOS devices, and soon to be available on Android.
“The second piece is (when) you’ve done something and you want to publish that to the decision makers,” Guard said. “You make sure you can get it out to the masses so they can see it, and consume it and make their decisions on it.”