Technological innovation compounds over time, with new advances building on the foundations laid by the inventions of the past. At the Spark East summit this past week, it was fascinating to see this dynamic play out in practice. Many of the higher level services in the spotlight at the conference are evolutionary next steps from lower level services created in the past. This is key to a well-known theory from Ray Kurzweill about how the exponential growth of technology occurs: Combining lower level services into higher order solutions creates true innovative acceleration.
I found this theory highly relevant to one product example from the summit. The product is called Cazena and it allows you to compile a service stack in the cloud to handle a variety of use cases, all while optimizing the TCO of your data environment. Right now, Cazena has stacks available for the data mart, data lake, and data science sandbox as a service. I find the company’s approach fascinating from a productized analytics standpoint. I’ve been chronicling the rise of #ProductizedAnalytics for a while now, (See explanatory graphic.) I think the movement towards these products are indicative of enterprises and tech innovators recognizing that not every analytic solution has to be custom-made – there’s room for Dunkin’ Donuts, just as much as there is for Per SE. (See “Productized Analytics: Why 100 Singles Are Better Than A Grand Slam”)
As I’ve written before, a fully enterprise-grade productized analytic offering allows you to take an analytic stack and apply it to a particular domain. To do this, the product must be able to handle a variety of sources of data, have an ingestion process, a generic data platform, analytic or application-specific data objects, recommended models and analytics, reports and visualizations, and data exports and APIs. My theory is that productized analytics require analytic stacks that offer these capabilities to users – a great example is what Salesforce is doing with artificial intelligence in its productized analytic solution.
What I find clever about this is that it allows companies a place to then plug in their application. In addition to compiling and creating software, the user gets to make choices about the components her or she wants to use. You get a range of offerings, but you can also conduct some experiments to see the best model to hit the sweet spot for your desired price and performance. For instance, if you have a high-performance model in mind, the Cazena tool will choose the components to support it – if you choose a lower cost model, it will choose different components. It’s providing the background infrastructure. And, if you compile it with one type of settings and find you are unsatisfied or want to try something else, you can change the settings later to make adjustments. The application code will still work in the new configuration, although I’m sure getting this type of compatibility perfect in practice will be tricky.
What I find clever about this is that it allows companies to express their applications with a clean separation from the underlying infrastructure. In addition to compiling and creating the platform, Cazena has embedded analytics tools in the service, so you can use methods like R, Python and SQL or connect a visualization, BI or other applications (Qlik, TIBCO Spotfire, Tableau, etc.) It also has built-in data movers for easily ingesting data from a variety of sources. The platform is delivered “as a service” in the cloud, including maintenance, security and operations. Cazena uses a combination of custom, open-source and best of breed software in its platform, but abstracts the technical complexity behind a self-service web interface.
Cazena thus provides a new degree of flexibility to companies interested in the analytic stack. It doesn’t lock users in, as it focused on solving a problem, not about creating the perfect analytic engine for companies. It is designed for those companies who can’t (whether because they lack the technical expertise or because of it being cost prohibitive) build their own stack.