In the News: Enterprises Redoing the Sins of CloudOps in MLOps

Last decade as cloud infrastructure was starting to emerge, many enterprises thought they could do it better if they built their own. Instead what they found out is that it took longer to build, was much more expensive to build and set up, and once set up was less agile to respond to market or tech changes compared to the public cloud.

We are seeing enterprises reenact the same play from the last decade, but this time by operationalizing AI and Machine Learning. In this article, our VP of Operations, Aaron Friedman, talks about how the focus on building in-house MLOps solutions (particularly with deploying, managing, and observing a model) has led to the poor return most enterprises get.

According to a joint study from MIT Sloane and BCG, only 10% of organizations achieve significant financial value from AI. Too often the answer is to just hire more data scientists and ML engineers, but Aaron argues that the right approach is to move from DIY deployment solutions and instead use platforms that 1) make it easy and fast to deploy ML models no matter what frameworks or tools were used to build it, 2) are purpose-built to run complex ML models efficiently on less compute, and 3) simplify model observability and testing to ensure you have the most accurate models feeding the business.

Read the entire article MLOps | Is the Enterprise Repeating the Same DIY Mistakes?