Analytic Personas and Predictive Models

Who are the users for analytic solutions?  In developing solutions, we tend to think of one group of users, those who work with statistical packages like SAS or SPSS.  In reality, every analytical solution has three distinct user communities whose needs must be addressed.

Analytic Power Users are responsible for developing predictive models.  They typically have significant training and experience in analytics, and usually prefer to work from a statistics or data mining client, such as SAS Enterprise Miner, SAS Enterprise Guide, SPSS Modeler, SPSS Statistics or Mathworks MATLAB.  A subgroup of these users may also be familiar with SQL, but in most cases they prefer database interaction to be brokered through the analytics client.  Power users may be attracted to R, primarily by the richness of the constantly growing CRAN library of open-source analytic procedures.

Quality of the predictive models -- measured by statistical precision, lift or related metrics -- is the most important priority for this user group.   Since power users are skilled in the use of their preferred client they tend to be loyal to it and reluctant to switch without a very compelling reason to do so; they may tolerate high costs and performance issues to avoid the costs of learning a new user interface and rebuilding an existing stock of models.

Analytic Developers take predictive models developed by Power Users and embed them in analytic applications for production use.  They are comfortable working at the SQL command line, and in programming languages such as C, Fortran or Java.  If they use SAS, they work in the programming language using advanced capabilities such as macros, arrays, indices and hashing.

The most critical concern for these users is the performance, capability and stability of the analytic procedures, as well as the ability to customize through user-defined functions and stored procedures.   Since developers must integrate analytics into a wide variety of enterprise applications, they value support for a variety of languages and software frameworks, such as C, C++, Fortran, Hadoop, Java and Python.  A key secondary concern for developers is the ability to seamlessly migrate predictive models from the Power Users' development environment to the ultimate production environment.

In most cases, Power Users and Developers are two distinct user communities with different reporting and organization structures.  In startups, though, and in organizations seeking to compete on analytics, there tends to be more overlap between these communities.

The third user group, Analytic Consumers, is the largest user group, the real driver for pervasive enterprise analytics.  Analytic Consumers may not think of themselves as users, and may not know they are consuming analytics: they see a recommendation, a forecast or some interesting content, but the underlying analytics engine is not visible to them.  There is no single client or application that meets the needs of analytic consumers; what they require is that their preferred user interface -- a browser, a spreadsheet, a BI portal, a campaign management application -- has the capability to surface analytics to help them make better decisions and improve a business process.

It may seem obvious that analytical applications must meet the needs of the ultimate end consumers of the analytics, and that analytics must benefit a business process.  Too often, however, solution architects overlook Analytic Consumers, or make assumptions about their requirements, leading to failed applications or costly retrofitting.  That's why effective solution design for analytical applications begins and ends with assessment of a business process.