INSIGHTS 24 March 2021 | 3 min read

Dave Lewis An accomplished Professional Services and technology leader with experience across a diverse range of enterprise software. Passionately customer focused and results driven with an entrepreneurial character, having built and managed businesses from startup. Significant experience working with international blue-chip organisations in a variety of markets including banking & financial services, telecommunications, public sector, AEC & FM, staffing, retail and transport.

MLOps is what MLOps does

A DevOps lady, an IT operator and a data scientist walk into a bar…

Stop me if you’ve heard this one.  

“The data to pinpoint the cause of an issue and the knowledge of how to resolve it automatically, perhaps even proactively, exists within the business. The data science, DevOps and ITOps teams have known for years that this data could be surfaced and used to automate and speed up issue identification and resolution. However, trust in technology in each other is historically lacking for collaboration to implement a solution that will add exponential value to the business.”

The notion of cross-departmental mistrust and instinctive wariness of technological automation has become so profoundly generic in businesses and our working lives, that it could be the foundation of at least five seasons of an alright (ish) US sitcom.

Why wouldn’t it be? The ongoing drama created by the human relationships and grudging dependence on each other in terms of data science and ITOps or DevOps is the stuff of comedy gold.

The trouble is that, for corporations everywhere, it’s really expensive and really not that funny.

Corporates often find themselves with the keys to a space rocket that either remains parked in their garage or is being driven like a lawn mower.

DevOps, ITOps and data science needs ML

For the avoidance of doubt, mostly everybody agrees that ML can be a game changer for businesses. A vision to combine a highly functioning, operational IT department or production team, with the very best insights from automation and machine learning isn’t news.

The drive for more insights, more often, more consistently and with real time speed has been the stated goal of progressive organisations everywhere, for a long time.

So why is it that so often the entire process is reduced to a series of poorly conceived experiments (seemingly) for their own sake?

Teams from either side bemoan that the project’s actual purpose is to placate egos, spend R&D budgets, say some buzzwords and then get back to the relentless daily grind of putting out fires in operations.

The eventual upshot being, “not for us thank you, we’ll soldier on doing what we always did. Even if it means getting what we always got.”

What MLOps is

MLOps is undoubtedly the answer, for two reasons. The first being what MLOps is, and the second being what MLOps does.

If you consider all the iterative stages of a process, combining data engineering, data science and deployment, all of these processes are highly specific and complex, all of them normally have specific tools or systems and, simply put, it doesn’t integrate that well, even before any human involvement. There’s often human touchpoints and points of handover that add further complexity (particularly in the politically charged environments of big corporations). High level visibility of the problem is often hard to establish when transparency between departments is hard to achieve.

An MLOps platform is the holistic automation of the processes involved in solving a business-impacting technology problem. It’s that simple.

 What MLOps does

An MLOps platform draws every process together. MLOps applies the gathered intelligence that can be trusted to solve an issue at speed.

  • More insights - MLOps drives the collaboration of both skill sets (data science and operations) and as such it drives more efficient ML.
  • More often - MLOps ensures best practice and adherence to regulation, allowing your operations team to operate unburdened.
  • More consistently – MLOps reduces distance and friction between the data and operations, unblocking bottlenecks created by lower functioning / non intuitive algorithms.

 Most importantly, the goals of the business are at the core of an MLOps strategy. Your MLOps specialists will ensure,

  • A strategy for a systematic, finite project (especially in the early use cases)
  • The data requirements of all parties for the project are met
  • That data compliance is at the top of the agenda
  • That success is based on collaboration and that every skill set is leveraged to deliver the best outcomes and the best products.
  • The delivered solution will have been developed to be maintainable to keep pace with the evolving business