< Articles & News

A Quant And A Business Manager Walk Into A Bar

July 31, 2015   |   2:31 PM
Big Data Analytics
Teradata Articles

Did you ever go to a car show and see the concept cars? The question is, do you drive one? Do you even see them on the freeway?

If we’re not careful, the work of data scientists could fall into the same category as concept cars: elegant and interesting but largely unconnected to day-to-day operations. When we look around, we see evidence that some companies are so focused on acquiring data scientists that they overlook the need to connect those data scientists to the rest of the business. It is almost as if having a data scientist or two on staff is the latest fashion statement in the world of analytics.

The point of data science isn’t to just scream “Eureka!” The real purpose is to have that moment and then quickly package the insight and deliver it to the company’s front lines for operationalization.

Being able to collect data at scale is less than half of the data science battle. The other half is taking data science insights and inserting them into tactical systems and business processes. The work of data scientists should show up in decisions made every day across the company, decisions such as price changes, product mix changes, service level tier initiations, customer retention program initiatives, preventive healthcare communications, and many more. Without the plumbing that connects data scientists to operations and a feedback loop that allows operations to inform data science, innovation often fails to get off the drawing board.

bar

Another misconception says if you feed data scientists all the data you can lay your hands on, they will magically elicit models that will revolutionize the business. That is a costly urban myth. Such a uniform diet produces pristine, laboratory-grade models. They look great but fail to account for the less than ideal conditions that exist in day-to-day operations. It’s like a shiny concept car with an engine that is designed to run only on roads that allow for speeds in excess of 100 mph: great for racetracks but of little use elsewhere. A collaborative process that brings data scientists into contact with business analysts, marketing strategists, corporate planners and the people who put models into action makes for smarter data science.

So how do you spot the difference between data science and what I would call operationalized data science? Here’s an example. Data science can create a churn model that indicates four actions that customers typically take before churning. If you wait for those four things to happen, you’ve lost the customer. Operational data science happens when that model triggers an indicator on someone’s desktop telling them the customer has taken two of the four steps and there is time to intervene. That requires connecting insight to business processes and infrastructure that can deliver timely signals and help initiate effective corrective actions.

Operational data science doesn’t stop there. There is attrition in models. Perhaps model attributes gain strength over time or people implementing them notice variables that are absent, making the model less than optimal. There needs to be a mechanism for people on the front lines to give feedback to the data scientist. Perhaps the model is effective in about 75% of the cases. That information can provide the data scientist with the direction needed to refine her segmentation and develop a different approach for the other 25%. Data scientists do not have to be viewed as gods who provide faultless models that work 100% of the time in every conceivable scenario. That is a setup for failure; such expectations actually drive companies to become less data driven and less scientific in their approach.

There is a lot of talk about there not being enough data scientists in the world. While that may be true for the moment, we could more efficiently use the resources that we have today. We can do exactly that by expanding the data science ecosystem—the people, the process, and the infrastructure—and put it all in the context of the business it is meant to serve. The more we implement data science insights in the trenches, the sooner we can all start driving cooler cars.