Introduction

The research report published in March 2022 by NASSCOM in collaboration with Genpact and EY can be viewed as a compendium of MLOps implementation framework and industry best practices.

As an increasing number of organizations are focusing on adopting AI/ML practices to scale up operations and generate meaningful business insights, what has been observed is that many of them has been unable to scale up their ML models at an enterprise level either due to inadequate model training or monitoring. Studies show that only 27% of pilot projects have successfully gone into production. Realizing the need to address this gap, NASSCOM published this compendium to provide a framework and methodology for enterprises planning to adopt MLOps and bring to forefront the best practices to guide the organizations in their journey.

The playbook is aptly structured with adequate sections on a) understanding MLOps and its benefits, b) end-to-end implementation framework starting from design, pre-processing, model development, its deployment and management, c) specific industry use cases spread across multiple sectors such as banking, automobile to food and beverage, d) business and operation pillars that talks about transformation planning, areas of investment and change management and most importantly e) control and data governance.

The report also features an additional section on the future of MLOps where it states that many organizations are looking forward to having centralized ML Operations despite it being an unchartered territory. Much of the insights of the report is based on understandings from interviews with experts from the field and secondary research on global practices, challenges in the field and the various ways organizations are overcoming them to adopt MLOps successfully.

Relevance of the Report

The report focuses on two important aspects of MLOps i.e., implementation pillars and business & operations pillars. The report will not only act as a guide for organizations for effectively setting-up and scaling MLOps but also help the stakeholders gain a clear understanding of it and be well-prepared to face any challenges that might come on the way.

Key Takeaways

• Though many organizations have embarked on the MLOps journey, a few managed to scale it up successfully
• The low rate of success can be mostly attributed to operations engineering practices such as validations of the model output, and monitoring for the integrated systems
• Deploying MLOps successfully requires clear design, organized data pre-processing, accurate model development consistent monitoring to figure out quality issues and other conflicts
• Organizations need to properly plan transformation to MLOps and invest accordingly
• Certain regulations are driving MLOps governance such as the EU General Data Protection Regulation (GDPR). In the Indian context, the government has started making laws and regulating in this regard and has formed multiple committees and taskforce to look into it