MLOps: The New Role in Data Science
Check what skills MLOps needs in Data Science and what this profession of the future is all about.
The demand for consistent, reliable insights in-house has brought about a new role – the machine learning operations (MLOps) analyst. In this Q&A we learn about this role and what it can mean for companies and data science teams.
Machine learning operations (MLOps) analysts have burst onto the scene as demand has grown among businesses for consistent, reliable insights in-house. We speak to Monika Rzepecka, MLOps analyst at the market, consumer and brand intelligence agency, GfK, to hear what the role involves and what it delivers, strategically.
What can the MLOps role involve?
MLOps is a very fresh area. It’s still developing, and companies have differing ideas on what the role is. Many businesses don’t even have it yet but are considering it. In essence, MLOps analysts are a section of the data science team.
Our role is to make machine learning, or AI, projects more systematic, repeatable, and well maintained. Whereas data scientists focus on algorithm design and the interpretation of information, MLOps concentrate on making sure the model is working at its best amid changing demands. We monitor algorithms metrics, improve performance, and set best practice.
Where does MLOps fit into data science projects?
These projects typically have the following stages: scoping, collecting data, training the model, and deploying it in production. After that, MLOps comes into play. We assess and monitor the viability of the model during use.
For example, as different data are added from new points of sale or customer interfaces, or as the business’ environment grows. We also have an important reactive role. This occurs when those using the analytics query the results or notice some misbehavior. At that point, we identify the source of the problem and remedy it as needed.
What advantages do business get from MLOps?
While there is not yet a textbook guide to what MLOps should deliver, the core expectation is that we ensure a systematic approach is taken with machine learning. We ensure the output is consistent in quality and effectiveness, has transparency and trustworthiness.
This applies to both historical and predictive analytics. There are also significant operating efficiencies that we deliver, by allowing data scientists to focus on their core areas, rather than monitoring and interrogating code usage.
How important is MLOps to GfK?
It’s crucial. At GfK we draw in a huge amount of data to fuel our business intelligence insights. We have moved from being a traditional market research company to being a trusted provider of prescriptive data analytics, powered by innovative technology.
As with all companies highly reliant on data science, it is paramount that our data-driven insights are transparent and reliable, and all processes are consistent and efficient. We currently have two full-time MLOps analysts, one position that we’re recruiting for right now, and we expect to recruit more.
What skills does someone in MLOps typically possess?
As a new role, people’s skills and backgrounds currently vary quite widely. My own background is in qualitative methods in economics, working with machine learning algorithms and monitoring tools. What is important is data curiosity. To be interested in what happens.
To be able to drill down into masses of data, spot patterns, and identify what is working or not working. We also need an understanding of machine learning models and algorithms, and the skills and enthusiasm to investigate and drive important change.
How can companies attract and retain talented MLOps?
I would say that many MLOps analysts are looking for the ability to grow in the role and make a tangible difference to how the company operates. In many cases, we are starting with a blank sheet of paper. We’re the kind of people who thrive on being allowed to innovate to find better ways of working.
MLOps analysts like the space to deep dive, and to shape the role as the business advances its use of analytics. Creating this environment of trust and freedom will be important for companies, given that MLOps analysts are in increasing demand.
What excites you about the future of the role?
For me, it’s seeing how MLOps evolves within different organizations as the appetite for the role grows among business leaders. There’s going to be accelerating growth in what MLOps delivers for different businesses, the smart tools available, and what is ultimately possible in collaboration with the data science team. I’m excited to be in at the start of this big movement.