Machine learning has evolved into a necessity for organizations that wish to improve customer experience, reduce costs, and build innovative solutions to complex problems, with a 2021 Forbes survey finding that 76% of enterprises prioritize machine learning (ML) over other IT initiatives.

However, even companies with an abundance of data scientists are struggling to deliver on the promises surrounding the buzz of ML for three major reasons:

No method to efficiently monitor and improve production models as they get stale over time.

Data scientists spend more time on maintenance than innovation.

Data product teams start from scratch on new ML applications.

The common threats posed by these three pain points have led to the emergence of a new discipline in data science, machine learning operations (MLOps). MLOps is a set of practices to combat issues with productionizing and maintaining models in machine learning. It borrows the ideologies of version control, automation, and CI/CD from DevOps, but has an added layer of complexity due to the use of data and artificial intelligence. One of the chief issues addressed by MLOps is ensuring that models continue to work as expected in production.

Presentation by: Amanda Aschenbrenner, Matt Pattermann and Cole Harrison from Credera