By using website you agree to our use of cookies as described in our cookie policy. Learn More

TDWI Upside - Where Data Means Business

How MLOps Delivers Business Value

Machine learning projects too often get stuck in the lab. Bringing the discipline and efficiency of DevOps into AI/ML practices helps move your models to production with ease, even at scale and in real time, and manage and monitor them over time for optimal results.

Data science has been driving a change in the way businesses solve complex problems. With the availability of very large data sets, significant advances in machine learning (ML) research and affordable computing power, an AI-fueled transformation is happening across industries. In a highly competitive market where consumer expectations are high and rising, ML/AI applications that detect fraud, decrease churn, deliver suggestions in real time, and predictively manage maintenance on infrastructure can be the critical differentiator.

For Further Reading:

The Machine Learning Data Dilemma

Data Requirements for Machine Learning

The Rise of Automated Machine Learning

Although AI/ML projects are emerging as a mainstream business need, companies are discovering just how hard it is to go from data science to business value.

No AI-driven project is an easy win. ML projects can rapidly explode technical debt absent a devoted operations practice. In the worst-case scenarios, ML projects don’t even make it out of the lab. Indeed, most ML projects don’t make it to production according to various estimates. It’s not that the science is hard or that talent is lacking. Instead, it’s everything surrounding the development of the model that slows the ML process down and prevents it from seeing daylight.

MLOps bridges this gap. As the name indicates, MLOps brings the discipline and efficiency of DevOps into AI/ML practices. Its goal is to create continuous integration and delivery (CI/CD) of ML intensive applications.

ML systems in particular can accumulate technical debt because they have all the maintenance problems of traditional code amplified by an additional set of data and ML-specific issues. To tackle this, MLOps streamlines the entire life cycle of ML, from data collection and preparation, model development, testing, and deployment to monitoring, governance, and business metrics. MLOps creates an environment where ML technologies can generate business value by swiftly and reliably building, testing, scaling, and deploying ML technology into production.


Companies including Netflix and Uber achieved market dominance in part by leveraging AI/ML as a core competency. In companies where AI/ML is not part of the DNA, building an AI culture is certainly a journey, but pragmatic steps can make it feasible. MLOps ensures that ML projects can be deployed and quickly generate value. Here are some ways MLOps enables efficiency and shortens time to market:

Developing ML Models with MLOps

Automation: Generally, ML teams spend the great majority of their time on everything besides the science. MLOps enables these scientists to get back to the science by automating the labor-intensive tasks of data collection, preparation, training, testing, and deployment.

Scaling up: ML teams need enormous amounts of computational power and storage to produce accurate machine learning models. Data scientists shouldn’t need to invest time on scaling up infrastructure as the project moves along. MLOps improves your team’s workflow with methods such as serverless technology, which enables your team to use computing power on demand as the need arises and manages resources efficiently. Your data scientists and your bottom line will thank you.

Deploying models with MLOps: Deploying a model to production is just the beginning of your work. ML teams need to plan for tuning or retraining models. Many times, real-world data can change unexpectedly in contrast to the sterile environment of the lab. Streaming data requires preparation and scaled-up computing power. Much like DevOps, MLOps accelerates time to value by deploying models seamlessly and continuously by building in a way that is production-ready from the beginning and by automating much of the packaging and production tasks along the way.

Managing Models with MLOps

Governance: Clearly, control and compliance is a potential roadblock for an industry such as fintech. Recent scandals vividly demonstrate just how critical governance is for any model. MLOps delivers access control, traceability, and audit trails to mitigate risk and ensure regulatory compliance.

Monitoring: The life cycle of a model has only just begun once it is deployed to production. A key role of MLOps is monitoring a model for concept drift. The abrupt and extreme changes in global patterns resulting from COVID-19 are a drastic example of how important it is to be able to promptly react to rapid change with fresh, real-time data. By implementing MLOps practices, teams can collaborate, iterate, and reuse models without duplicating effort while auto-scaling resources.

MLOps: The Critical Enabler

In a crowded market, a competitor that delivers business value faster inevitably surpasses the competition. The practices that enable teams to predictably, reliably, and quickly deliver business value are well established in software development. In the product development world, it is expressed as a method of working from hypothesis to validation and iteration as quickly as possible. The same principles can be applied to getting ROI from ML applications, though the practice is much more complex.

Even in a bull market, companies look for ways to rapidly produce value and decrease costs. Many companies have suddenly found themselves in an existential fight for survival due to the COVID-19 crisis, and getting AI applications to market can be a critical differentiator when the dust clears on this difficult phase. Inaccurate predictions, application downtime, or delays in bringing models to production are avoidable costs that are particularly perilous now. Against the backdrop of uncertain times, decreasing time to market with MLOps is mission-critical.

The tumultuousness of 2020 should also cause companies to carefully consider the operational challenge of monitoring and maintaining a single ML model, let alone several. The COVID-19 pandemic caused ML models across many industries to go haywire in response to rapidly changing conditions combined with a lack of relevant data sets to inform a correction. Quick interventions enabled by a proper established MLOps practice are critical.

MLOps Affects the Entire Team

Preparing data, building models, and getting them out of the lab and into the real world is a complicated hurdle that requires coordination across many roles in a company. As companies across industries embed AI/ML applications into their processes, IT leaders must invest in MLOps to drive impact.

About the Author

Yaron Haviv is a serial entrepreneur who has been applying his deep technological experience in data, cloud, AI, and networking to Israel’s leading enterprise companies since the late 1990s. As the co-founder and CTO of Iguazio, Yaron drives the strategy for the company’s data science platform and leads the shift towards real-time AI. He also initiated and built Nuclio, an open source serverless platform with over 3,400 Github stars and MLRun, Iguazio’s open source pipeline orchestration framework. You can contact the author via LinkedIn or Twitter.

TDWI Membership

Accelerate Your Projects,
and Your Career

TDWI Members have access to exclusive research reports, publications, communities and training.

Individual, Student, and Team memberships available.