≡ Menu

Some thoughts on advanced analytics


As part of the build up to today’s tweet jam on advanced analytics, Jim Kobelius discussed some of the questions they are planning to use in a blog post – Advance your analytics strategy. There’s a lot of good stuff in the article but I do have to take issue with a few things.

First, the definition of advanced analytics. There’s the usual problem whether to consider advanced analytics part of BI or not. I prefer not to as I find talking about BI pushes people to immediately think about reporting and dashboards. Instead I talk about a sliding scale from BI to descriptive and predictive analytics – from descriptive knowledge about the past to increasingly prescriptive predictions about the future. I would also take issue with Jim on social media analytics – there is nothing particular about analyzing social media that makes it “advanced”. Social media is a source of data (and an under-exploited one for sure) not a style of analytics. Social network analysis is a kind of advanced analytics, however, though it often does not use social media data (see this post on social networks in telco). CEP (Complex Event Processing) is also not a form of analytics but a kind of system that uses analytics.

While I agree with Jim that next best action is a classic use of advanced analytics I am not sure I am willing to accept that it covers all uses of advanced analytics and Jim seems to imply it does. Have to think about this one more.

I am not as sure as Jim that there is a lot of overlap between advanced analytics and BI/DW/Data Governance. Sometimes there is but often the focus on reporting inherent in existing BI programs means that the infrastructure and processes don’t lend themselves to advanced analytics. I also thing there is a fundamental problem with saying there is a need for all data being used for advanced analytics to be perfect. This is simply not true – there can be no absolute measure of quality as it depends on what you are trying to do. Data quality is about being good enough to improve the quality of decision-making, nothing more.

I was also frankly irritated by this question:

Is advanced analytics ready to roll out to all information workers, or is it still the province of a priesthood of data mining specialists?

This is the wrong question. The systems that use advanced analytics are often used by people with far less skills than “information workers” and that makes them easier to push out to more people than BI systems. This idea that, because the models are built by experts that advanced analytics only impacts a small number of people is just wrong. Instead of trying to push BI to more people companies should focus on pushing decision-making systems out to them and embedding analytics into these systems. Call center reps, store clerks and other front-line folks have neither the time, the skills or the inclination to use BI tools and we should focus on helping them with decision management applications. In contrast to Jim I have seen dozens of cases where many non-information workers are using systems that EMBED advanced analytics. They don’t build the analytic models but they sure benefit from them. The systems in which they are embedded are much more broadly used than any BI tool and that’s the point.


Comments on this entry are closed.

  • Mark Eastwood December 16, 2010, 7:48 am

    I completely agree with you on the topic of rolling out advanced analytics. While information workers are needed to develop the analytics the value an organization derives from using them comes from pushing good decisions all the way out to the edges. The call center and the web site and any other customer point of interaction. These are clearly ont the same information workers that developed the models yet they use them, knowingly or unknowingly inside the systems they use to make product recommendations, cross-sell suggestions, loan-work out options, product qualification decisions, pricing decisions and more.

    I’m not ready to accept there’s much overlap between predictive analytics and BI much less data governance. BI tends to be reporting on what happend to provide insight into taking future action. Predictive analytics, as the name implies, is about predicting the likelihood of some event some distance into the future. This prediction is then used to make decisions/take actions. BI tells me how many units of some widget I sold last month and if data exists how much market share my wdiget has against the competition and alternatives. Pehaps I use that to make decisions about actions I need to take, but that’s low-volume/high-value manual decsions. Predictive analytics helps me make good high-volume decisions which individually might have relatively small financial impact, but taken in aggregate represent significant value/revenue. BI simply isn’t in the same operational process; they each have their place.

    As for data governance, I can imagine a role for BI and/or analytics as it relates to data quality but see that as a totally different space than making an operational decisoin while my customer is on the phone with my call-center agent. Some recent work for a large automobile manufacturer concerns leveraging analytics and data governance comcepts to help maintain a high-quality database of vehicle owners. The difference to me is that I’m not predicting anything.


    • James Taylor December 17, 2010, 5:45 pm

      Like you I worry about considering data governance, quality and integration across both BI and more advanced analytics. Not sure there is as much common ground as people sometimes imply. The debate about including predictive analytics within BI continues to rage and we will have to see where it ends. I think BI has a genuine focus on delivering information to PEOPLE and advanced analytics can do that and deliver information to systems/processes. Moving the definition of BI to include people and systems strikes me as broadening it without necessarily improving it. Second I think, in a purely pragmatic sense, the current definition of BI means reporting/dashboards to most people and this means they don’t “get” advanced analytics when they get included in BI.

  • Riz Mithani December 16, 2010, 10:19 pm

    James and Mark are spot on as usual!

    The most extreme example of advanced analytics that requires tremendous expertise to create but can be used by anyone on the street is the FICO score. Just because it is a simple number, it does not imply its analytics are simple. It is our job as anayltics experts to operationalize the results of advanced analytics so that they can be ubiquitously used by people who need to make decisions at the individual transaction level.