Getting Value Out Of Data Science

A few months ago I stumbled upon a Microsoft commercial where one of my favorite rappers, Common, spoke with vigor about how “AI” is changing the world. Yet the reality that Common may or may not be aware of (but I’m sure Microsoft is) is that a majority of large enterprises have yet to operationalize AI and ML in ways that make them money.

At Relativity6, we emphasize the return-on-investment our AI and ML products bring in an operational environment, and we know AI is not a science project, but we’ve encountered the frustration of business leaders trying to implement new technologies into their workflows, and here are some of the hurdles we’ve seen:

Business units are busy doing business

Most businesses are divided into units, and those units are measured by their profit-and-loss (P&L) statements. More often than not, these units have quarterly P&L performance goals that they must hit. Since it can take months for a data science team to transform data and generate predictions, some business unit leaders are less than thrilled at the prospect of implementing an automated data analytics program that might disrupt their operations. One prominent business unit leader I work with called his large in-house data science team a "glorified science project." Ouch. Don’t worry, we were able to convince him otherwise. When operationalizing AI and ML, less is always more.

Data scientists are busy doing data science

I’ve met with many companies that are stacked to the brim with top level data scientists. These scientists are talented, organized, and extremely dedicated to creating bleeding edge technology. But for all their brilliance, there’s not a good feel for how their business unit counterparts actually operate. It’s rare to encounter a scenario where the data science and business units are co-creating a solution, and even more unique when the technical team has thought through exactly how the automated predictions are going to be implemented and received by their business unit counterparts. Typically, the data scientist/business development relationship is a one way street: request > reply. This is a reactionary way of doing business and it takes away one of the largest benefits of AI and ML programming: proactive insight generation.

In order to see value from AI we have to operationalize it

The good news is that operationalizing AI is achievable. I’ve seen it done incredibly well in a few organizations, but more often than not, without the proper structure and methodology for formulating viable use cases, implementing new workflows, and creating actionable and interpretable results, it can also be a resource drain and cause tension among colleagues.