For years, business leaders have been told that artificial intelligence (AI) can give them a competitive edge. As a result, AI adoption is becoming fairly mainstream. In our recent research, 68% of businesses, globally and across industries, have adopted AI and machine learning (ML).
Yet, many are struggling to scale their AI initiatives and realize business value from these projects. In our study, just 39% of respondents said AI/ML had contributed to significant business value. According to Gartner, just 53% of AI proofs of concept are ever scaled to production, and fewer manage to deliver the intended, measurable business value.
In our work with clients to scale their AI initiatives, we've recognized a make-or-break factor among those who’ve succeeded with elevating their AI maturity: adoption of machine learning operations (MLOps).
MLOps—which combines machine learning, data science, software engineering and cloud infrastructure—requires businesses to master several critical capabilities in the right combinations. These capabilities range from enterprise strategy, culture and talent, to the ability to experiment, develop, deploy, manage and monitor machine learning models at scale, collaboratively, involving various teams from start to finish in a structured and coordinated manner.
Realizing AI business value with MLOps
Think of MLOps as a guiding light on the path to scaled AI. Rather than taking an ad hoc approach to deploying, monitoring and governing AI applications, MLOps structures, standardizes and automates the process, from the exploratory data analysis stage, to the taxonomy and feature buildout, through modeling and production, and, finally, to monitoring and feedback.
With an MLOps-based approach, organizations can plug in their most recent or relevant machine learning models and generate reliable, efficient and consistent access to everyone involved. The business, data, development and production teams can all gain insights at the right time and cadence to make effective decisions.
This democratized approach to data access is consistent with what we are seeing in the industry, where there is a significant shift toward AI initiatives that use low-code/no-code solutions. These solutions allow non-experts to access insights in order to make decisions and use AI applications at scale.
The benefits of MLOps
MLOps delivers many benefits that can boost return on investment (ROI) of AI initiatives:
- Promotes a culture of collaboration and communication to improve the efficiency and effectiveness of ML projects
- Enables organizations to build, deploy and manage machine learning models quickly and efficiently, reducing the time and resources required to get value from AI
- Improves the overall quality and reliability of the AI systems being deployed
- Enables businesses to respond more quickly to changing business needs and take advantage of new opportunities
- Promotes responsible AI, protect sensitive data and models, and ensures compliance with ethics, regulations and standards
Here’s what these benefits look like in action. We recently worked with an asset-intensive organization responsible for nationwide infrastructure maintenance in the UK. The rail operator wanted to use AI and ML to improve service uptime and the overall passenger experience. With 20,000 miles of track, 30,000 bridges, tunnels and viaducts, and thousands of signals, level crossings and stations, maintenance and upkeep is critical to the safety and reliability of the entire system.
At the start of the initiative, we helped establish clear goals and objectives for the rail operator’s ML initiatives, which helped focus its efforts and prioritize the most impactful projects. Next, we implemented a version-control system and process so that the cross-functional teams could track changes to the ML models and ensure they were always working with the most up-to-date versions. In addition, we set up a continuous integration and delivery (CI/CD) pipeline, which automated the process of building, testing and deploying the models, saving time and resources.
We also put in place protocols for monitoring and managing the performance of the ML models, as well as a way to manage “data drift,” a common reason for model performance degradation. Finally, we established a strong focus on security and compliance, and supported the team in building a learning culture that supported the responsible and ethical use of AI.
As a result of this structured, standardized and automated approach, the rail provider was able to deploy a scaled AI initiative that delivered an array of business benefits, including:
- 10% reduction in service-affecting failures
- 15% reduction in passenger delays
- More punctual and predictable passenger journeys through real-time AI-based failure predictions
- Prioritized service interventions, using analytics risk models to determine asset criticality
- Improved worker safety by eliminating the need for manual inspection and trackside visits
Creating the MLOps foundation
This level of success doesn’t just happen—it requires businesses to follow a set of best practices that result in a strong MLOps foundation:
- Clarify the objectives and goals for AI and ML with a roadmap and clearly defined priorities
- Assess the current ML environment to identify pain points or bottlenecks that need to be addressed
- Establish a governance structure for the cross-functional team of data scientists, engineers and other stakeholders to ensure collaboration and foster a culture of learning and continuous improvement
- Create protocols for monitoring the ML model and model performance
- Design security and compliance protocols, as well as guidelines for using responsible and ethical AI
Now that AI is in the business mainstream, AI projects can no longer function like the wild west. With the structured, standardized and automated approach of MLOps, businesses can ensure their AI initiatives deliver on business value and objectives, while continuing to build the confidence of the business users to consume AI effectively.