5 Common Challenges to MLOps Capabilities & Development

Enterprise interest in artificial intelligence and machine learning (AI / ML) is exploding with experts forecasting a 10x increase in spending over the next five years. However, the complex nature and exacting demands of effective AI / ML are above and beyond those experienced with traditional, legacy systems. As a result, a new discipline, MLOps, is emerging to manage these capabilities and tackle challenges.

Searching for more MLOps insights? Watch the replay of our latest webinar “MLOps in Action: Real-World Examples for Establishing Best Practices”.

Naturally, new capabilities are always met with unique challenges, and those around AI / ML are especially complex. With that in mind, here are the five most common challenges a business enterprise might face when establishing and strengthening MLOps capabilities: 

1. Struggling to Gain Executive Buy-In and Support

Getting buy-in and support from the top is always important, but this is especially true for MLOps projects. Processes are more complex, and results that generate ROI may be slow to develop. Not only does this call for consistent executive support, but it also requires the alignment of incentives so that the enterprise gets on board as well. AI / ML is a departure from the ordinary and requires specific feeding and handling that keeps an eye on the big picture. 

2. Maintaining Data Quality and Availability

Data is often referred to as the oil of the modern enterprise, and AI / ML is no exception. New sources of data are often required, and data quality is doubly important. Both data availability and quality are essential for MLOps to develop and flourish. Data management is a discipline in and of itself, and it should be closely aligned with MLOps efforts. Poor data will lead to slower development times and diminished insights. 

3. Changing One Thing Without Changing Everything

Given the complex interplay involved in AI / ML models, any change made to one part of the delivery chain will likely substantially impact the overall program. Because this level of sensitivity comes with the territory, it is incumbent that an MLOps regime accounts for both the upstream and downstream impacts that a change to any input or part of the models will have.

4. Preparing for Eventual Model Decay

In addition to factoring in the “change one thing, change everything” paradigm, an MLOps development regime needs to account for AI /ML models that will be called on to evolve as new insights emerge and business conditions evolve. MLOps must also account for the fact that models invariably decay over time. Such decay should be monitored, and allowances for maintenance and new development should be built in and accounted for.

5. Contending With Model Locality

It can be tempting to point an AI / ML model that worked in one area at a new problem, but this is not likely to produce useful results. Unlike commodity dashboards and code, ML models are less likely to deliver utility if an attempt is made to apply them to a new area. For that reason, MLOps will require more resources and processes, and procedures should be implemented to monitor and measure models as they are utilized. 

Getting the Most Out of MLOps

No matter where your business is, be prepared to devote resources to develop MLOps capabilities as these revolutionary new technologies and techniques are incorporated. That said, for a number of reasons, success with AI / ML can be difficult to achieve on your own — but that doesn’t mean it can’t be done. With the right partner, you can approach your next AI / ML project with eyes wide open and maximize help from outside of the enterprise, thus allowing your team to focus on other areas of the business.

To learn more about Maven Wave / Atos’ approach to MLOps, download our recent white paper “Delivering on the Promise of AI / ML: The Emergence of MLOps”.

Maven Wave / Atos drives the future of industry with innovative business outcomes, fueled by cloud, with risk top of mind. To help organizations maximize economic outcomes and advancements, we bring a rich blend of industry-specific technological expertise, agile-integrated design, and best practices for transformation. Contact us to learn more.

About the Author

Brian Ray
Brian Ray is a Managing Director and Data Science ML/AI Horizontal Practice Lead for Maven Wave Partners. Mr. Ray is heading the group’s mission of solving complex analytical problems for major businesses worldwide through the power of Data Science with enablement in the cloud. Prior to Maven Wave, Brian has 20 years of hands-on experience in Engineering around the Sciences. A big picture strategist, team builder, and influential top technologist, he has extensive expertise in agile delivery of Data Science — from ideation/discovery, feasibility, exploration, modeling, to engineering and architecture, to hands-on integration and deployment of best-in-class solutions.
February 27th, 2023

Get the latest industry news and insights delivered straight to your inbox.

Sign up for our Newsletter