James Taylor's Decision Management

Gary Cokins of SAS blogged about Predictive Analytics – Dream or Reality? today. He was responding to Juha Harkonen's post from a CFO conference. Essentially Juha was complaining that while a small number of companies are getting into predictive analytics, many are still struggling with basic performance management and KPIs. I have heard similar comments from folks focused on BI, especially operational BI (for instance in this morning's TDWI webinar).

I am going to challenge the assumption that data quality, data warehousing, data integration, reporting, OLAP and performance management must all come BEFORE predictive analytics and decision automation. In fact, if you look at those industries furthest ahead in the use of predictive analytics such as retail banking, this assumption is clearly false. Many banks have been using predictive analytics in automated decisions for years but are not much further along in the rest of the data/BI competencies than your average large company. Clearly one does not need an enterprise data warehouse strategy nor does one need to have solved all one's problems with BI and data before using predictive analytics. So why the persistent talk of it being "after" all those things?

Well, for one thing it is more complex in its detail so there is an inbuilt assumption that it must come after the others. It is also true that newcomers to predictive analytic usage often have done a lot in terms of data quality, governance, warehousing and reporting. Lastly it seems logical that the group that manages data should also manage predictive analytics.

In fact none of these things is axiomatic. Companies focused on improving operational, high-volume decisions may well find it easier to embed a predictive model than to handle performance management on that process in real time. Many predictive models are built in environments with incomplete data and many predictive modeling techniques are designed to come with these problems. Many organizations have both an analytic or decision sciences group AND a BI Competency Center and they don't overlap at all.

The best way to approach this is to decide which decisions you must improve and what you need to do that. Start with the decision in mind and don't assume you must complete everything at any given "level" before moving on in targeted areas - each decision has its own constraints and needs and should be addressed in that context.

Predictive Analytics does work and has worked for years. Many companies are using it. You don't have to finish BI before starting predictive analytics.