Still at the Teradata conference today and I watched Wayne Erickson of TDWI present on Analytic Applications and the various available development options for them. His presentation was based on a report that he did that can be downloaded from TDWI's site here.
Anyway, he used the following definition for an analytic application:
An analytic application consists of a series of logically integrated, interactive reports, including dashboards and scorecards, that enable a wide range of users to access, analyze and act on integrated information in the context of the business processes and tasks that they manage in a given domain, such as sales, service, or operations.
Wayne gave a great summary based on this definition but it is with the definition itself I would like to take issue. I believe that EDM offers a way to deliver analytic applications that do not, necessarily contain "reports" at all. Indeed I would offer the following definition of analytic applications instead:
An analytic application enables a wide range of users to act on integrated information in the context of the business processes and tasks that they manage in a given domain, such as sales, service, or operations.
This is clearly a subset of Wayne's definition so let's examine what I removed:
"consists of a series of logically integrated, interactive reports, including dashboards and scorecards"
I removed this becuase I firmly believe that this is only one way in which"integrated information" can be accessed, analyzed and acted on. Let's take an example. If I have the right, integrated data I can predict the likelihood that a customer with 30 days to go on their contract will not renew it. I could do this by giving a suitably knowledgeable user access to the integrated information through an interactive report or dashboard. Alternatively this data could have been used by an analyst to build a predictive model that can be executed in an application to generate a score. Likewise although a user could act on this by taking some kind of manual action, it is also possible to instead define a series of business rules that take that prediction and take certain actions based on the combination of the business rules and the predictive model. Such an application could, perhaps, email an offer under certain circumstances or queue up an outbound call in some others.
Even though such an EDM system, combining business rules and analytics, would not meet Wayne's original definition I still believe it is an analytic application and potentially a very powerful one.
So when would you design an analytic application that met my criteria but not Wayne's? Typically for a high-volume, operational decision where it is much more helpful to have the system take the actions when it can rather than referring it to a person, however well-informed that person might be analytically.
Perhaps I should call them "operational analytic applications"?