This article is part of a special issue of VB. Read the full series here: The Quest for Nirvana: Applying AI at Scale.
To say that it is difficult to implement AI at scale across the enterprise would be an understatement.
Approximately 54% For 90% machine learning (ML) models don’t go into production in early pilots for reasons ranging from data and algorithm issues to defining the business case, getting buy-in from leadership and the challenges of managing change.
In fact, pushing an ML model into production is a significant achievement for even the most advanced company that employs ML and artificial intelligence (AI) specialists and data scientists.
The company’s DevOps and IT teams have tried modifying legacy IT tools and workflows to increase the chances of a model being promoted to production, but have had limited success. One of the main challenges is that ML developers need new process workflows and tools that better fit their iterative approach to coding, testing, and re-running models.
The power of MLOps
It’s there that MLOps comes into play: the strategy emerged as a set of best practices less than ten years ago to respond to one of the main roadblocks preventing the business from putting AI into action – the transition from development and training to production environments.
Gartner defines MLOps as a comprehensive process that “aims to streamline the development, testing, validation, deployment, operationalization, and end-to-end instantiation of ML models. It supports publishing, activation, monitoring, experience and performance tracking, management, reuse, updating, maintenance, version control, risk and compliance management and governance of ML models.
Manage models just to gain scale
Verta AI Co-founder and CEO Manasi Vartak, an MIT graduate who led undergraduate mechanical engineering students at MIT CSAIL to build ModelDB, co-founded her company to simplify the delivery of AI and ML models in enterprises at large scale.
His thesis, Infrastructure for model management and model diagnosticsoffers ModelDB, a system for tracking the provenance and performance of ML-based workflows.
“While the tools for developing production-ready code are well-developed, scalable, and robust, the tools and processes for developing ML models are nascent and fragile,” she said. “Between the difficulty of managing model versions, rewriting research models for production, and streamlining data ingestion, developing and deploying production-ready models is a massive battle for small and large companies.”
Model management systems are essential for MLOps to be operational at scale in enterprises, she explained, increasing the likelihood of modeling efforts to succeed. Model iterations can easily get lost, and it’s surprising how many companies don’t do model versioning despite having large teams of AI and ML specialists and data scientists on staff.
Having a scalable model management system in place is key to scaling AI in a business. AI and ML model developers and data scientists tell VentureBeat the potential to achieve DevOps-level returns from MLOps is there; the challenge is to iterate the models and manage them more efficiently, capitalizing on the lessons learned from each iteration.
VentureBeat is seeing strong demand from companies experimenting with MLOps. This observation is supported by IDC prediction that 60% of companies will have operationalized their ML workflows using MLOps by 2024. And, Deloitte predicts that the market for MLOps solutions will grow from $350 million in 2019 to $4 billion by 2025.
Increase the power of MLOps
Supporting MLOps development with new tools and workflows is key to scaling models in a business and extracting business value from them.
On the one hand, improved version control of model management is crucial for business growth. MLOps teams need model management systems to integrate or scale and cover staging, packaging, deployment, and models running in production. What is needed are platforms that can provide extensibility across large-scale ML model lifecycles.
Additionally, organizations need a more consistent operationalization process for models. How an MLOps team and business unit work together to operationalize a model varies by use case and team, which reduces the number of models an organization can promote to production. The lack of consistency is pushing MLOps teams to adopt a more standardized approach to MLOps that capitalizes on continuous integration and delivery (CI/CD). The objective is to obtain a greater visibility on the life cycle of each ML model by having a more complete and coherent operationalization process.
Finally, companies need to automate model maintenance to increase throughput rates. The more automated model maintenance can become, the more efficient the entire MLOps process will be and there will be a greater likelihood that a model will go into production. MLOps and data management platform vendors need to accelerate their personality-based support for a wider variety of roles to provide customers with a more effective management and governance framework.
MLOps vendors include public cloud platform vendors, ML platforms, and data management vendors. Public cloud providers AWS, Google Cloud, and Microsoft Azure all provide support for the MLOps platform.
DataRobot, Dataiku, Iguazio, Cloudera and DataBricks are the major competing vendors in the data management market.
How LeadCrunch uses ML modeling to generate more customer leads
Cloud-Based Lead Generation Company LeadCrunch uses AI and a patented ML methodology to analyze B2B data to identify prospects most likely to become high-value customers.
However, updates and revisions to ML models were slow, and the company needed a more efficient approach to regularly updating models to provide customers with better lead recommendations. LeadCrunch’s data science team regularly updates and refines ML models, but with over 10 sub-models and an ever-changing stack, implementation has been slow. New models were rolled out only a few times a year.
It was also difficult to get an overview of the experiences. Each model was managed differently, which was inefficient. Data scientists have struggled to get a holistic view of all the experiments going on. This lack of insight further slowed the development of new models.
Deploying and maintaining models often required a lot of time and effort from LeadCrunch’s engineering team. But as a small business, those hours were often not available. LeadCrunch evaluated a series of MLOps platforms while seeing how they could streamline model management. After extensive research, they chose Verta AI to streamline every phase of ML model development, release management, production, and ongoing maintenance.
Verta AI freed LeadCrunch data scientists from version tracking and organizing so many models. This allowed data scientists to do more exploratory modeling. During the initial rollout, LeadCrunch also had 21 issues to address, with Verta AI addressing 20 immediately after implementation. Most importantly, Verta AI increased model production speed 5x and helped LeadCrunch achieve one deployment per month, up from two per year.
The powerful potential of MLOps
The potential of MLOps to deliver models at the scale and speed of DevOps is the main motivation for companies that continue to invest in this process. Improving model yield rates starts with an improved model management system that can “learn” from each model recycling.
There needs to be greater standardization of the operationalization process, and the CI/CD model needs to be applied not as a constraint, but as a supporting framework for MLOps to realize their potential.
VentureBeat’s mission is to be a digital public square for technical decision makers to learn about transformative enterprise technology and conduct transactions. Discover our Briefings.