Operations Managers: The Unsung Heroes of MLOps

By on
Read more about author Audrey Reznik.

Operations managers play a critical role in the MLOps life cycle. Without the proper infrastructure, the number of CPU nodes, or security checks, models, and applications built by data science and development teams are more likely to fail – if they even reach the deployment phase.

That’s why operations managers should be brought into the MLOps process from the beginning. Doing so can yield many benefits. For example, operations managers may be able to point out infrastructure compatibility issues as applications are being developed. This can save developers and scientists from having to go back and revise their work when development is almost complete. When an operations manager is involved, projects can run more smoothly, efficiently, and at speed – right from the start.

Let’s look more closely at how organizations can adjust their MLOps processes to involve operations managers throughout the entire development cycle.

Why It’s Important to Involve Operations Early

The value that operations managers bring to the development cycle cannot be overstated. They  support the entire MLOps process, from conception to deployment. Consider that:

  • Containerized applications cannot run smoothly without the proper number of clusters.
  • Programs cannot launch without the correct amount of compute power. 
  • Development and data science teams cannot collaborate successfully without the right underlying infrastructure. 

All of these fall under the purview of operations managers. But these are only some of the reasons why operations managers should be involved in the MLOps process from the beginning. Others include:

  • Ensuring an application runs well is easier to do when operations managers understand the purpose behind the application and how it was created, from the modeling phase on up.
  • Operations managers can point out flaws in the process before applications reach the deployment stage, helping to mitigate the need for significant rework at the end. Examples of flaws could include vulnerabilities, compatibility issues, and other challenges that could prevent applications from running optimally.
  • Operations managers can offer unique perspectives on how to make an application run better. For instance, they might recommend using a different or more efficient programming language for a particular application.

When an operations manager is invited to participate in the application development process, they become a true part of the MLOps team. The invitation to participate shows them that their work and perspectives are highly valued and are core to the success of the project.

Those who are invited to participate and collaborate with data science and development teams are more likely to feel like they’re co-owners of the project. In my experience, they will get excited about the project and feel compelled to participate in its development proactively. Their active participation will ultimately help the project run more smoothly and successfully, leading to more agile application development and ensuring that models and applications have the right infrastructure support.

How to Involve Operations Managers Early 

Getting operations managers involved in the MLOps cycle from the beginning does not have to be difficult. It simply requires involving the operations team in project discussions and planning from the outset. A simple email or chat message will often do the trick: “Hey, I wanted to reach out about a new data science project we have coming up with which we could use your support.”

Perhaps the most challenging part of this initial outreach will be managing the reactions of different team members. I’ve found that, while operations managers, data scientists, and developers are often surprised by the invitation for operations to join initial planning discussions, that’s only because it’s so unusual – a breaking of the IT cultural norm. Once that initial surprise wears off, however, everyone generally settles into their roles and coalesces as a team.

The others in the process – data scientists and developers – more often than not quickly begin to really appreciate the value that operations managers bring to the project. For example, it’s not uncommon to hear a data scientist express appreciation for the work that operations managers do because it helps them better understand how their models will ultimately be deployed and put to use.

Likewise, operations managers will often state how appreciative they are to have access to data science teams. It helps them understand the type of infrastructure those teams will need to make their models run. A side benefit is that everyone ends up learning from each other and picking up different skills.

How to Keep Operations Managers Engaged

From that point forward, it’s important to keep operations managers engaged throughout the cycle. This means giving them access to all of the communication resources data scientists and developers use to collaborate with one another. Tactics and tools that will help with continuous collaboration include:

  • Assigning a point person to facilitate interactions between data science, development, and operations teams. While the idea is to get different team members to freely interact with one another, it’s still a good policy to have a single person making introductions and managing the process – at least initially.
  • Setting up group chats and channels that team members can use to collaborate and bounce ideas off each other. Team members may end up having side conversations in these channels. That’s to be expected and encouraged. It means that everyone, operations included, is engaged.
  • Holding weekly or bi-weekly meetings with all team members. These will give everyone a chance to provide status updates and exchange ideas on how to improve processes.
  • Having an “open door policy” that invites operations managers to check in multiple times a day or week with suggestions and questions. This policy gives operations managers the opportunity to approach the point person or other team members with questions, suggestions, or new ideas outside of the normal meeting cadence.

How Operations Managers Make Projects More Efficient and Cost-Effective 

The unique ideas and experience that operations managers bring to the MLOps process can lead to measurable process improvements, including greater efficiency and cost savings. These improvements can benefit both the team and the organization they work for.

For example, I recently developed a workshop. At the beginning of the project, I asked an operations manager to help set up the appropriate infrastructure to support the project. I told them what I had planned for this one-time event, and the manager pointed something out that I had not considered:

“You’re probably going to run this workshop more than once,” they said. “Let’s set this up so that it’s failproof. I’ll write up some scripts to bring the cluster down when the initial workshop is completed, and then back up again when you need it. That way, you don’t have to reinvent the wheel every time, and instead of it taking days to spin back up, it’ll take about 30 minutes.”

That simple, unprompted, outside-the-box suggestion will save me and my organization enormous time and money later on. There are many other examples of the different perspectives operations managers bring – ideas about how to better secure the production process, for example, to better set up a Git repository, or simply leveraging past experiences and learnings to optimize application development.

Indeed, given everything operations managers bring to the application development table, it makes sense to bring them to that table before it’s set up. They’ll appreciate the gesture and their involvement will make the production process run smoother and more efficiently, resulting in better overall products.

We use technologies such as cookies to understand how you use our site and to provide a better user experience. This includes personalizing content, using analytics and improving site operations. We may share your information about your use of our site with third parties in accordance with our Privacy Policy. You can change your cookie settings as described here at any time, but parts of our site may not function correctly without them. By continuing to use our site, you agree that we can save cookies on your device, unless you have disabled cookies.
I Accept