Still, while technically sound and powerful, these solutions haven’t generated the expected revenue, which has raised concerns about future growth.
I can understand the pessimism surrounding the space, as I spent the first 20 years of my career effectively building internal MLOps tools at an esteemed investment management firm. More recently, I’ve invested in MLOps startups, but they have been slow to achieve the level of revenue that I would have expected. Based on both my positive and negative experiences with MLOps, I understand why these startups have struggled and why they are now poised for growth.
MLOps tools are critical to companies deploying data-driven models and algorithms. If you develop software, you need tools that allow you to diagnose and anticipate problems with software that could cause you to lose meaningful revenue due to its failure. The same is true for companies that build data-driven solutions. If you don’t have adequate MLOps tools for evaluating models, monitoring data, tracking drift in model parameters and performance, and tracking the predicted vs. actual performance of models, then you probably shouldn’t be using models in production-critical tasks.
However, companies deploying ML-driven solutions without deep knowledge and experience don’t recognize the need for the more sophisticated tools and don’t understand the value of the low-level technical integration. They are more comfortable with tools operating on externalities, even if they are less effective, since they are less intrusive and represent a lower adoption cost and risk if the tools don’t work out.
On the contrary, companies with ML teams who possess deeper knowledge and experience believe they can build these tools in-house and don’t want to adopt third-party solutions. Additionally, the problems that result from MLOps tools’ shortcomings aren’t always easy to identify or diagnose—appearing as modeling versus operations failures. The outcome is that companies deploying ML-based solutions, whether technically sophisticated or inexperienced, have been slow to adopt.
But things are starting to change. Companies are now recognizing the value of sophisticated, deeply integrated MLOps tools. Either they have experienced problems resulting from not having these tools or they have seen competitors suffering from their absence in many high-profile failures, and are now being forced to learn about the more complex MLOps solutions.
Those MLOps companies that have survived the revenue winter so far should see a thawing of the market and a growth in sales opportunities.
Companies selling superficial solutions will start losing business to more integrated solutions that are harder to understand and adopt, but provide more monitoring, debugging, and remediation services for their customers. MLOps software developers should keep the faith that building powerful software that solves problems in a deeper and more thorough way will win out in the long run over simple solutions that give immediate payoffs but don’t solve the full breadth of problems their customers are facing.
The post Why Sophistication Will Win Out In The Machine Learning Ops Sector appeared first on Unite.AI.