Todd Biske: Outside the Box

The idea of context-driven architecture, as coined by Gartner, has been bouncing around my head since the Gartner AADI and EA Summits I attended in June. It’s a catch phrase that basically means that we need to design our applications to take into account as much as possible of the context in which it is executed. It’s especially true for mobile applications, where the whole notion of location awareness has people thinking about new and exciting things, albeit more so in the consumer space than in the enterprise. While I expect to have a number of additional posts on this subject in the future, a recent discussion with a colleague on Data Warehousing inspired this post.

In the data warehousing/business intelligence space, there is certainly a maturity curve to how well an enterprise can leverage the technology. In its most basic form, there’s a clear separation between the OLTP world and the DW/BI world. OLTP handles the day to day stuff, some ETL (extract-transform-load) job gets run to put it into DW, and then some user leverages the BI tools to do analytics or run reports. These tools enable the user to look for trends, clusters, or other data mining type of activities. Now, think of a company like Amazon. Amazon incorporates these trends and clusters into the recommendations it makes for you when you log in. Does Amazon’s back-end run some sophisticated analytics process every time I log in? I’d be surprised if it did, since performing that data mining is an expensive operation. I’m guessing (and it is a guess, I have no idea on the technical details inside of Amazon) that the analytics go on in the background somewhere. If this is the case, then this means that the incorporation of the results of the analytical processing (not the actual analytics itself) is coded into the OLTP application that we use when we log in.

So what does this have to do with context-driven architecture? Well, what I realized is that it all comes down to figuring out what’s important. We can run a BI tool on the DW, get a visual representation, and quickly spot clusters, etc. Humans are good at that. How do you tell a machine to do that, though? We can’t just expect to hook our customer facing applications up to a DW and expect magic to happen. Odds are, we need to take the first step of having a real person look at the information, decide what is relevant or not with the assistance of analytical tools. Only then can we set up some regular jobs to execute those analytics and store the results somewhere that is easily accessible by an OLTP application. In other words, we need to do analysis to figure out what the “right” context is. Once we’ve established the correlation, now we can begin to leverage that context in a way that’s suitable for the typical OLTP application.

So, if you’re like me, and just starting to noodle on this notion, I’d first take a look at your maturity around your data warehousing and business intelligence tools. If you’re relatively mature in that space, that I expect that a leap toward context-driven applications probably won’t be a big stretch for you. If you’re relatively immature, you may want to focus on building that maturity up, rather than jumping into context-driven applications before you’re ready.

One Response to “A Key Challenge of Context Driven Architecture”