BizOps, MarketingOps, DevOps, AIOps, MLOps, DataOps. As the names suggest, they're all cross-functional, but do companies need all of them, some of them, or none of them? That depends on one's point of view. Clearly, organizations are at different stages of maturity based on their industry, size, age, culture, tech adoption and budgets. However, increasingly, organizations need the benefits of what the different kinds of Ops provide.
Like DevOps, the various types of Ops aim to accelerate processes and improve the quality of what they're delivering: software (DevOps); data (DataOps); AI models (MLOps); and analytics insights (AIOps). Some consider the different Ops types important since the expertise required for each type differs.
Others believe it's just hype, specifically relabeling what already exists and/or there's a risk that the fragmentation created by all the different groups may create extra bureaucracy that frustrates faster value delivery.
XOps began with DevOps
Agile software development practices have been bubbling up to the business for some time. Since the dawn of the millennium, business leaders have been told their companies need to be more agile just to stay competitive.
Meanwhile, many agile software development teams have adopted DevOps and increasingly they've gone a step further by embracing continuous integration/continuous delivery (CI/CD) which automates additional tasks to enable an end-to-end pipeline which provides visibility throughout and smoother process flows than the traditional waterfall handoffs. Like DevOps, DataOps, MLOps, and AIOps are cross-functional endeavors focused on continuous improvement, efficiency and process improvement.
The XOps landscape
The Ops landscape continues to expand but this article focuses specifically on DataOps, MLOps, and AIOps.
DataOps is a process-oriented methodology that uses automation to improve the speed of data-related tasks and ultimately the quality of insights. According to Arvind Prabhakar, CTO of of DataOps platform StreamSets, DataOps enables change management agility for companies with complex data infrastructures.
"The number of people in the data supply chain has gone through the roof so you're now looking at the modern enterprise trying to keep up. Just at the data infrastructure level, it's an order of magnitude more complex than what it used to be 10 years ago," said Prabhakar.
Like DevOps, DataOps involves rapid iteration, measuring and monitoring to facilitate end-to-end understanding. The key role here is the data engineer. And, in companies that have them, the chief data officer (CDO) wants to drive process optimization that ensures data reliability and governance.
MLOps bridges the creation of machine learning models, their deployment, and operation in production. DataOps and MLOps are considered more closely related than AIOps because AIOps is a higher (application) level process than the other two.
Management and technology consulting firm Booz Allen Hamilton started Its MLOps journey by looking at DevOps concepts and asking how they could be applied to MLOps.
"You have to bring together technologists, data architects, modelers [and] security experts," said John Larson, senior vice president at Booz Allen and leader in the firm’s analytics business. "It's not intuitive in that modeling framework to think about building something that needs to be containerized so it can be scaled and deployed. The first principle of modeling tends to be what is my algorithm?"
Booz Allen uses MLOps to understand model performance and drift in real time as well as whether the model was deployed as intended.
"What we're emphasizing is the importance of data version, model version and deployment," said Larson. "This integration of the MLOps with DataOps and DevSecOps, in a DevOps framework enables you to have those types of insights. I think it's important to scaling what we're doing and the adoption of what we're doing because it's going to give us the mechanisms, the tools, the traceability to understand what's happening in the model when it's deployed, how it's drifting, how it's refreshed."
Gartner's definition of AIOps is that it "combines big data and machine learning to automate IT operations processes, including event correlation, anomaly detection and causality determination."
The key benefit of AIOps is actionable insights. In fact, AIOps specifically benefits analytics.
On the other hand, is XOps just hype?
Forrester Principal Analyst Charles Betz advises against getting caught up in the XOps hype.
"To some extent, there's the usual kind of buzz and people throwing ideas against the wall to see what'll stick, so I don't really worry about this BizOps, DevOps, DevSecOps, BizDevSecOps, Marketing Ops. It all basically does the same thing and that is a movement toward a more cross-functional way of working," said Betz.
One reason not to chase the various forms of Ops is because it contradicts IT staffing trends. Traditionally, specialization was considered a good thing but more modernly, IT roles are becoming broader (e.g., full-stack developer or site reliability engineer). In fact, modern HR professionals will say it's more important for a department or team to have the right mix of skills as opposed to the right mix of titles or roles.
"The large companies are telling me they have a coordination problem. I've had two senior vice presidents say to me in the same week, 'I don't have enough DBAs for every product team to get [one] so we wind up with a shared services problem." said Betz. "There's a need for expertise and so how do we solve this problem without going back to the bad old days of it takes six months to get anybody to get back to me when I need something. That's why people want the cross-functional world."
Betz also thinks that Scrum and Agile methods should have the right orientation and that the default should be the product team.
"The ideal is that the product team has the resources and approvals it needs to get the job done. You do not have an operating model that relies on interchange and transaction, you have an operating model that relies on collaboration," said Betz. "The fundamental point to all this is that the operating model pivots from high-transaction process overhead to pure collaboration between these focused teams with process and transactional friction being the exception, not the rule."
While there's merit in the argument that different types of Ops require different types of expertise, beware of unnecessary complexity and bureaucracy.
"The evidence is overwhelming that process gridlock results when you overspecialize," said Betz. "You get a bunch of middle managers and they're all trying to increase their kingdoms because your pay is tied to the number of people who report to you."
Ultimately, Betz considers the elements of XOps no more than a set of terms used for internal marketing purposes.
"I don't object to it, but I don't care about it and I don't have a lot of patience for it because the deeper current is this trend towards more collaborative product team models [in which] process friction is the exception," said Betz. "And in many cases, the biggest transactional friction is between those who create versus those who operate and react, so that's why all of this stuff is XOps."
More types of DevOps-like organizations are forming in today's enterprises and vendors are responding with solutions such as for DataOps. However, from a competitive standpoint, having the right capabilities is place is more important than what those capabilities are called. Given the organizational uniqueness from one company to another, whatever businesses consider "XOps" will differ from one company to another, like DevOps.
Follow up with these articles on DataOps, MLOps, and AIOps: