GigaOm Radar for Machine Learning Operationsv2.0

1. Summary

Machine learning operations (MLOps) solutions streamline machine learning (ML) model development, deployment, and governance, enabling production ML at scale. They simplify each phase of an ML project and automate related maintenance and operational procedures.

Customers rely on ML to make sense of massive amounts of data and drive change across their organizations. The goal of MLOps is to make ML a fully integrated part of a business and, by doing so, achieve a faster time to value and reliability. Enterprise MLOps involves operationalizing potentially hundreds of ML models across different teams. This mission usually requires a platform with enterprise-grade capabilities that can automate the ML process, not only through development and deployment but also by addressing post-deployment monitoring and management. Most platforms in this report offer customers end-to-end ML lifecycle coverage, with a specific focus on operationalization.

In the GigaOm report, “Key Criteria for Evaluating Machine Learning Operations Solutions,” we evaluated MLOps features we expect most MLOps solutions to support, which are key criteria to look for while choosing an MLOps solution, and we explored the strategic evaluation metrics used to assess the impact that an MLOps solution might have on an organization. We also briefly discussed the emerging technical capabilities we expect to become relevant over the next couple of years.

In this Radar report, we help buyers become familiar with MLOps solutions and vendor offerings. MLOps solutions covered in this report can help customers avoid production pitfalls by ensuring that ML models make it into production and produce reliable, reasonable, and ethical predictions.

For this analysis, we review pure play and startup vendors that are focused and specialized on MLOps capabilities as well as cloud providers, data platform vendors, and incumbent enterprise software players that offer a broad range of data analytics services and features in addition to MLOps capabilities.

• Pure play and startup vendors provide a valuable solution for enterprises that heavily depend on data science and, specifically, ML use cases.
• Cloud, enterprise software, and data platform vendors offer many additional non-MLOps capabilities and features within their products to provide an optimal solution for large enterprises. But they can also require outsized investments for small-to-medium businesses (SMBs) that may not need such broad functionality.

Note: We have included some products in this report that could strictly be classified as data science rather than MLOps solutions. However, the two categories overlap considerably, and the particular vendors and products we’ve included here all serve broad MLOps use cases in their own right.