Why an analytics culture change needs strong data stewards

Data Scientists have traditionally been the truth bearers in a sea of employees with the need for analysis however changing dynamics and universal access to analytics and data, data scientists need to evolve into the role of a data steward where their role now needs to ensure that the enterprise’ strategic decisions and actions are based on sound analysis that do not misinterpret, mis-analyse or misrepresent the truth. 

As enterprises become more data savvy and arm their employees with analytics tools that democratize access to disparate data that has so far been locked away, they can expect wider adoption of data driven thinking and decision making across their employees. However, this change will create a new challenge for enterprises. 

New challenges from a wider adoption of analytics in the enterprise 

Wider use of data and analytics will create a new set of problems. Aside from the increasing pressure on analytics platforms from a larger group of users, more data sources, higher frequency and granularity of data and the need to get analysis as fast as possible, enterprises should prepare themselves for defining and practicing high quality data analysis procedures. 

Enterprises need to ensure that whenever data analysis is used to push strategic decisions and investments or tactical actions, the analysis/visualizations/interpretations/recommendations:

Evolving role of the data scientist 

History is full of examples where analytics have been purposefully designed to prove a narrative as opposed to analysis defining the narrative. It will be hard for enterprises to remove the tendency of their employees (especially employees used to the old way of doing things) to go by their gut feel, look for analysis that supports their way of thinking or skew analysis to support their opinions. 

This is where the data scientist role has to adapt to become a “data steward” role. The data steward has the responsibility to ensure analytics is used to drive enterprise strategy and actions and that it is done so in a systematic, scientific manner based on sound reasoning and logic. The data steward has to ensure that the incorrect assumptions are not made and that they establish a process through which such assumptions, incorrect analysis or flawed visualizations and interpretations are quickly identified and addressed.  

The area of academic research has addressed this problem really well and offers a great set of best practices to guide the data steward as they establish appropriate data analysis procedures in their organizations or make product and technology investment decisions. 

Data Stewards should actively consider setting up analysis processes and investing and using analytical products that enable the following: 

Double blind analytics 

Analytics products that, through their ease of use and on-boarding, enable analysis similar to “double blind” experiments where a completely detached employee can successfully recreate the analysis and results and that the analysis, regardless of how many times it is recreated, returns the same results if run on the same data set. 

Peer reviews 

Analytics products that, through their collaborative capabilities, enable a group of employees to inspect, correct, interview, question and enhance an analysis. 

References and referential integrity 

Analytics products that, through their interfaces and analytical and reasoning state preservation capabilities, can enable analysis to be built upon past sound analysis or enable the combination of past inferences from multiple analysis done in the past to drive a new conclusion.   

In conclusion, data scientists have to evolve to handle the new demands and pressures of this new world where data and analytics are democratized and accessible by all employees. They have a new role to play where they have to ensure that their organization adapts sound analytical processes and practices and adopts products and platforms that enable such sound analytical practices. After all, the only thing worse than not being data driven for an enterprise is using flawed analysis to drive business strategy that causes often irreversible momentum loss, bad investments and customer dissatisfaction and abandonment.