How Organizations Struggle with Data Fluency - Data Fluency: Empowering Your Organization with Effective Data Communication - page 38

coordinate with the human resources (HR) department to connect information
on teachers with their students. Unfortunately, HR had no systematic way of
generating electronic records on teachers. In fact, most of the records were
kept as hard copies in file drawers. When HR began entering the information by
hand into spreadsheets, it found that teacher identification numbers changed
annually and didn't match state identification numbers. Years of student data
could not be connected back to teachers charged with these students.
DATA E LITI SM
Working with data can require a lot of technical skill. And data can tell stories
and reveal truths that an organization may not want to share broadly. Why
not centralize your efforts and limit access to data to the highly trained few
who can be trusted to bring order to chaos? This is the viewpoint of Tom
Davenport in his topic, Competing on Analytics . In his view, the best analytical
organization is one that has centralized control:
Stage 5 organizations develop a robust information management environment
that provides an enterprise wide set of systems, applications, and governance
processes. They begin by eliminating legacy systems and old spaghetti code and
press forward to eliminate silos of information like data marts and spreadsheet
marts. They hunt for pockets of standalone analytic applications and either
migrate them to centralized analytic applications or shut them down.
(Harvard Business Review Press, 2007)
Like an over-eager police force hunting down deviants, this IT-led vision of
business intelligence focuses on control, consistency, and data management.
An extreme approach, however, comes at the expense of the individuals who
use the data. Distancing analysis from the people who must use it results in
data producers and their products that are disconnected from the decision-
making process. Data products aren't trusted and they often aren't useful. All
the problems of a command and control economy emerge.
We've seen this happen in credit card organizations where data scientists built
models to predict customer behaviors that were not in alignment with the
regulatory environment. The knowledge of what decisions could be made
given the regulatory environment was in the hands of the data end users.
Getting the data scientists to understand the real business environment took
extra time and effort and the initial set of models were useless.