Table of Contents
- How Does MLOps Benefit ML?
- Applying MLOps in Practice
- A MLOps Scenario: Customer Churn
- The MLOps Maturity Model
- Delivering on MLOps Maturity
- Conclusion: Proactively Adopt MLOps
- Real World Use Cases
- Annex: An MLOps Maturity Model
- About William McKnight
This report is targeted at Business and IT decision-makers as they look to implement MLOps, which is an approach to deliver Machine Learning- (ML-) based innovation projects. As well as describing how to address the impact of ML across the development cycle, it presents an approach based on maturity levels such that the organization can build on existing progress.
The paper is for practitioners who require practical advice on MLOps, rather than general principles of ML. It references ML-related activities and their importance but does not go into technical detail about the specifics of ML nor associated activities.
While the paper uses Azure Machine Learning and its documentation as a reference point, its guidance will apply to any ML environment.
Drivers to MLOps
Over the decades, data has proven to be a competitive differentiator. Once it was exclusively reports built by IT from overnight-batch-loaded data warehouses, but top performers have moved from passive reporting to predictive and prescriptive analytics, growing their skills in data science, and changing accepted paradigms as they derive insights to drive their businesses forward.
In recent years, rapidly falling costs of processing and increased throughput have unlocked new opportunities for organizations to maximize their data assets. Many companies have spent years or even decades collecting data in their data warehouses, data marts, data lakes, operational hubs, etc., and some now have the infrastructure and tools to act on the data optimally.
Based on established scientific principles, machine learning (ML) can deliver even greater levels of insight from data than traditional approaches, straight to the point of need, and without manual intervention. As shown in Figure 1, ML unlocks a broad range of opportunities across vertical sectors: the rewards will be greatest for those who have the skills, experience, and capabilities they need to deliver on its potential.
Figure 1. Use Cases are Becoming Vast for ML Pioneers Across a Wide Variety of Areas
These are still early days for ML, and success is not a given. Adoption faces several challenges:
- Senior management does not always see ML as strategic, and it can be difficult to measure and manage the value of ML projects.
- ML initiatives can work in isolation from each other, resulting in difficulties aligning workflows between ML and other teams.
- To be effective, ML training requires large quantities of high-quality data, which creates significant overheads across data access, preparation, and ongoing management.
- ML/data science work requires a large amount of trial and error, making it hard to plan the time required to complete a project.
In response, ML adoption requires a cultural shift and a technology environment with people, processes, and platforms operating in the responsive, agile way organizations are looking to operate today: an approach we can call MLOps. Creating such a culture and environment cannot happen overnight: it comes by learning from those at the vanguard of ML how to map the potential of MLOps- driven innovation against an organization’s specific needs and resources.
Based on multiple interviews with those working at the front lines of ML, a pattern emerges that sees the journey to MLOps success in terms of maturity. In this paper we present how to take MLOps strategy to practical reality, using best-practice principles and a maturity model that helps decision-makers assess, define, and enact steps towards MLOps leadership.