In-Depth

How Lyza Learns from its Users

The new Lyza learns from how people use it and makes recommendations based on a user's role or interests. The more people who use it, the better, officials say.

• By Stephen Swoyer
• 01/10/2012

Instead of targeting Lyza 3.0 at a select or prescribed class of users, Lyzasoft Inc. is pitching it to everyone. When co-founder and CEO Scott Davis says everyone, he really and truly means it.

"I mean just about everybody. We believe that social is the new search. We think that the conversations and the relations and the recommendations that come from other people [which Lyza 3.0 keeps track of] ... provide us with excellent information for a recommendation engine that's able to say, 'This particular data is likely to be relevant for me because Sam said that it would be relevant for me,'" Davis explains. "The more people like me -- or like Sam -- using it, the better."

Lyzasoft once positioned its flagship business intelligence (BI) offering as a workgroup BI-like tool, designed primarily for power users or business analysts. This positioning evolved over time, however, as Lyza fleshed out its feature set in unpredictable -- some might say unprecedented -- ways.

Lyza 2.0, for example, attempted to marry social media concepts or methods with BI analysis. It marked, arguably, the first or most robust application of social media in a BI or analytic context. It made such a big splash that BI industry watcher and BI This Week contributor Ted Cuzzillo suggested Lyzasoft "stole the show" when it launched Lyza at TDWI's World Conference in Las Vegas almost two years ago.

Actual revolutions are rare, at least when it comes to the upending or rupturing of long-standing paradigms. If Lyza 2.0 comprised a legitimately revolutionary approach to BI and analysis -- and some folks believe that it did -- Lyza 3.0 offers an evolutionary refinement. It brings threaded and contextual discussions, along with recommendations and annotations, to the Lyza Commons environment that first shipped with Lyza 2.0. It introduces an improved and social-savvy emphasis on relationships -- between users, things, or assets.

If the model with Lyza 2.0 was Facebook, the model in Lyza 3.0 is Pandora, the popular commercial music-streaming service. "Our model is really Pandora and not Facebook. I call [Facebook] a first-order service. I type something in, I get what I asked for. There's also a second order thing going on at Pandora: if you do the thumbs up or thumbs down [on a given music selection], as you consume the service, you're actually training the recommendation engine," says Davis.

"Something very similar is happening inside of Lyza 3. As you the casual user are looking at things or talking about things, everything that you do is being harvested into metadata, which is nothing more than a set of descriptions of the things that you're looking at."

The upshot, Davis claims, is that Lyza 3.0's recommendation engine can learn from how someone uses it. Like Pandora, it can make recommendations -- concerning blog postings, spreadsheets, PDF files, and any other conceivable asset -- based on a user's class, responsibilities, or interests.

"In an enterprise [environment], you're going to have hundreds of millions of documents and potential assets, and 99.9 percent of these are going to be completely useless. They're just sitting around," he stresses. "With a social platform like this, we see that those useful aspects start to be discussed. We can see popularity and utility rising to the top, so that the bad stuff can be pruned."

There's a sense in which the idea of pervasive BI sometimes seems like a technology prescription divorced from real-world utility or applicability of any kind. In other words, make BI pervasive to what end? In the Lyza 3.0 usage model, Davis claims, there's a clear logic -- along with a notional utility -- in making BI pervasive.

"If you get all employees into the platform, everything that they say, everything that they do, everything that they look at enriches the metadata engine so that it can make better recommendations," he points out.