“Starting in the 1950s, AI has a long history of being the next big thing,” said Megha Sinha, vice president of Digital – Data Science, AI, ML at Genpact in her talk titled “MLOps – The Strategic move to actuate Data Insights”. She discussed MLOps at length during her session at The Rising. Today, every enterprise aspires to be data-driven, she said.
While it has become imperative for business leaders to actuate data insights for business growth, studies show only half of AI proof of concepts are scaled to production. MLOps holds the key to create production-ready, scalable AI solutions. “MLOps is a set of practices that defines how to deliver solutions and build pipelines,” said Megha.
Watch all the recorded sessions of Rising 2022 here>>
AI for business
AI has gone through a series of ups and downs, including two extended periods of doldrums, dubbed AI Winters. Businesses started taking an interest in AI through the 1980s. However, the excitement petered for the lack of workable AI-based solutions in the market owing to a dearth in computing power and data. “Today, 85% of AI and machine learning projects fail to deliver. Only half of our projects reached from prototype to production stage. Now, while the latest findings show no increase in AI adoption, many companies are beginning to incorporate AI and ML in different functions slowly and steadily. But the question now arises, are we staring at yet another AI winter? Megha asked.
The need of the hour is integrated custom design applications with ML-powered software and pipelines that can deliver repeatable experiences to clients across the breadth and depth of AI and ML. The companies are investing heavily to improve their product portfolio, capture wider markets, and in the process is keeping the AI winter at bay, she said.
MLOps strategy
Megha then discussed the three fundamental blocks of AI for businesses:
- Data
- ML model.
- The code to integrate all components
“While it sounds very easy to do, changes in one part of the pipeline can trigger a change throughout the system. This proves the principle of changing something, changes everything,” said Megha. MLOps helps streamline this complicated process. It helps businesses identify best practices, methodologies, tools, and technologies to unify the entire pipeline.
Additionally, it is important to understand business analytics to answer questions such as where exactly do we need to invest, how to build a change management strategy etc. The answers will differ from organisation to organisation.
Megha spoke about five focus areas for AI in business:
- Define team structures
- Unified ML platforms to build, deploy and monitor.
- Continuous model monitoring and automated deployment to reduce the release cycle and failure rates.
- IT team for infrastructure and asset provisioning.
- Responsible AI practices at all stages.
After studying the key areas, businesses need to create an MLOps strategy. First, they need to identify the objectives and understand the need for MLOps in the first place. Then, explore questions like is your system big enough, do you intend to scale it, etc. Follow it up with assessing the company’s tangibles- the number of models, infrastructure and the team size. Then, evaluate the intangibles by designing the culture. Follow this by defining the scale and designating change management. Lastly, build the metrics and then actuate.
Megha also discussed an interesting case study from Genpact for an insurance client looking to streamline the underwriting process for a global life reinsurer. The company created two models where one predicted the likelihood of acceptance concerning underwriting appetite as an initial triage to filter out potential declines and the likelihood to prioritise further the accepted risks for underwriting action to achieve the target of responding to the best set of 30-35% risks. The result achieved a top of line 12-15% improved risk selection, enhanced customer experience and more strategic power to the underwriter.
