Although many enterprises are beginning to heavily invest in data science activities, only a handful of them are seeing the desired ROI. That’s because it’s hard to do. It’s difficult to shift to a culture of building and scaling data-driven services and products. It’s a challenge to operationalize data science processes and integrate data science into business practices. It’s an uphill battle to put an enterprise’s data “house” in order, to get a complete view of what data exists and eliminate data silos and connect disjointed analytics teams. All of these challenges are among the reasons why most enterprises aren’t getting the return they expected from their data science investments.
However, it is possible to do, with the right approach. If your data science practice isn’t delivering the results you expected, or you are in the process of exploring how a data science capability might fit into your organization, here are a few considerations to take into account to improve your success.
When enterprises begin to dabble in data science, most want to start small, with a proof of concept (POC). This makes sense. However, many data science POCs never make it into production. With the pressure to find more value in Big Data, many companies rush into doing a POC without first putting strategy, controls or processes in place that will make it easier to build and deploy a successful POC at scale if desired. Without these prerequisites in place, many essential components necessary for scaling the project are not documented as data scientists do their exploration and model building work. While POC projects are experimental, it is also essential to develop a well-defined scope, plan and evaluation metrics at the start with an eye on future production deployments.