3
ANSWER THE RIGHT QUEST IONS
For truly actionable insights, data science needs to know what it’s looking for. According to Ken Schultz, VP of data science at Elevate Credit, “A potential issue with using data science in an organization is solving problems that don’t need to be solved, or solving irrelevant problems.” 27 In other words, to get the best from data science, you need to make it clear what stories you are looking for in the data and why. Again, this comes down to determining the business context and the specific problems that need to be solved. The key is to identify the story you need to know the ending of—and stick to it. Let’s say that two data scientists look at the same set of loan applications. Scientist A finds that appli-
cations peak in the morning; scientist B sees that applicants who meet certain criteria are more likely to default. Both are correct—the data science tools don’t lie—but scientist A’s findings are of no use to the business. Thecompany’s loanprocessingplatformis powerful enough to manage any spikes in volume, whenever they happen; what senior managers needed was a more accurate indication of who will or won’t pay back a loan. The problem is that nobody told scientist A what they were looking for. So, there’s moral to this story. The results of data science will only influence business decisions when they answer the right question. 28 Containerization technologies provide a more modern, workable alternative, allowing you to use native microservice software to meet your changing needs. A microservices architecture also limits any possible technical failures to isolated components, which allows you to take advantage of lightweight, cloud-based applications. Finally, it’s critical that models can scale up easily to the higher volumes of data and performance demands they can face in production. Again, a microservices-based approach can help—by letting you adapt quickly through simple changes to the configuration. 29
4
TAKE CARE OF TECHNOLOGY
A number of operational challenges can slow down the deployment of data science models, but they are easily overcome with the right technology. For example, the model developed by your data scientists may not automatically be compatible with your production environment, so may need to be recoded before IT can deploy it. The best way to avoid this scenario is to use an agnostic scoring engine, which can handle models created in any language and deploy them easily into production. Monolithic modeling platforms can pose another problem and limit the evolution of your models.
© 2019 PROVENIR ALL RIGHTS RESERVED
Made with FlippingBook Digital Publishing Software