In India, predictive analysis, also known as predictive analytics, can be easily compared to the class system. While some companies have flourishing information intelligence systems, others have meager adoption levels. However, to establish an organization’s reputation in sales innovation and operational efficiency, predictive analysis is important.
The adoption of predictive analysis depends on the processes in an organization and the industry it belongs to. In the banking and finance industry, the adoption is high owing to the regulatory compliance. On the other hand, in industries like auto and retail, the adoption of predictive analysis is in the early stages as the texture of data is not yet mature.
This three-step exercise will help you launch the predictive analysis program in your organization with ease.
1) Examine prerequisites for a predictive analysis program
Before finalizing the predictive analysis program, check for the following prerequisites:
- Healthy data: Initiating a predictive analysis program may require an incredibly hygienic set of data for a minimum of three years to build data models upon. For example, if you are trying to judge the propensity of a customer buying an insurance product, you will need to have information of at least three years describing his buying behavior.
- Technological maturity: The level of technological maturity in an organization should be assessed. For instance, to capture customer behavior in a bank, it is necessary to gather information across the various sales and managing systems, along with systems that capture risk management data about defaulters and discrepancies. For a predictive analysis program to work efficiently, the applications that capture all this data need to be mature.
- Executive sponsorship: Executive sponsorship can make predictive analysis an easy game if the business itself demands for it. If this is not the case, one way to pitch would be to identify a business process and leverage the vendor’s slice of the tool before buying it. After generating some results, evangelize that to create a business case.
In essence, you should have four key roles before starting the project: executive evangelist, data analysts, statisticians, and actual practitioner of the field.
2) Evaluate predictive analysis tools
Use the following criteria to select an appropriate predictive analysis tool:
- Pricing: Thorough examination of the pricing model is necessary before finalizing the deal with the predictive analysis tool vendor. Ensure you exactly know what the vendor is offering you: is it a ‘fixed price’ or ‘fixed price with an annual recurring cost’ or ‘pay per use’ model? Ascertain the benefits and risks associated with each model.
- Flexibility: Consider if the predictive analysis tool would need a hard core statistician to model it or if it could be configured using some visualization tools available in your existing business intelligence (BI) package. If the predictive analysis tool lacks the flexibility to utilize features of the existing systems, a lot of labor will go into configuring it, thus increasing costs.
- Deployment measures: Consider how well can the predictive analysis tool be deployed to an operational process. For example, see how a cross-sell model can be created and how it would be integrated with the call center application, which, in turn, would help in intercepting the conversation with the customer and would use some triggers to sell the products (to the customers).
- Integration: Checking for integration would entail inquiring about the various adapter options the vendor is offering. The predictive analysis can take place only through integrated BI systems.
- Acceptance of flat-file data: A BI system creates an interface in the form of a flat file, which can be pulled into an SPSS environment to model. There is a workaround for tools from different vendors; but the question that needs to be kept in mind is, how effective is it?
3) Make predictive analysis pervasive
It is necessary to scale up the predictive analysis program and make it enterprise-wide to maximize the return on investment (RoI). The following guidelines will help you in making your predictive analysis program pervasive.
- Map processes for optimization: The first stage would be to map the process that is generating the data and identify its owner. Processes mostly work sub-optimally owing to newer acquisitions, which may not generate clean data. The owner has to be made liable for optimization of the process and the data it uses to pull him onto the predictive analysis bandwagon.
- Create targeted models: The next stage would be to create targeted models. Over a period of time, 30% to 40% of the processes in an organization would have been touched by the analytical model. This would grow with maturity.
- Express output as a business rule: Thirdly, the output of the predictive analysis program could be expressed in the form of a business rule. Then it is a question of time and effort to incorporate that business rule in the existing applications.
Lastly, evolve all these processes slowly and steadily to achieve an effective cross-sell and measure the ROI by tracking the key performance indicators.
About the Author: Derick is the vice president - advanced analytics & research at MindTree’s data analytics solutions group. He has 20 years of experience in telecom, CPG, retail, and banking industries. In his current role, he is responsible for creating industry-specific offerings using advanced analytical constructs to transform the efficiency of the business processes. He can be reached at firstname.lastname@example.org
(As told to Sharon D'Souza)