Machine Learning is a much spoken topic nowadays – perhaps more spoken about than available capabilities allow for. There is nothing new about machine learning, neural networks, or artificial intelligence. These concepts have existed for a while with early discoveries in the 1700s. Even recently, there was a huge push in research in AI/ML in 70s and 80s.

The last 5-7 years have witnessed an upsurge in activity in this realm. The reason for this increase in discussion and action is primarily the rapid decrease in compute costs, increase in compute efficiencies and tool that allow for simple, relatively non-programmatic approaches to get these insights.

Insights have driven operations for a long time now. At one point, such action was purely reactive. Then it slowly moved towards being pre-emptive. And now it trending towards being predictive. Not every situation requires or will benefit from predictive insights. While insights must be actionable, not all need predictive traits. And even when predictive, how far predictive is another question all together. While technology groups are being asked to play a role in the move to predictive insights, the percentage of technology organizations that apply such philosophies to their own operations is not all that high. There are few who believe that an organization needs to be of a certain size for it to use prediction capabilities. Others believe that its unaffordable. The reality is that organizations of any size, operations of any scale will benefit for being able to use predictive capabilities.

Technology operations can leverage machine learning techniques generate predictive insights in a variety areas. From predicting where the next alert on their server farm will come to predicting the code from where the next software fault will arise to predicting the set of test cases to use in regression test cycle, technology operations can leverage prediction capabilities. The pre-requisite is that heuristics exist, can be funneled to a data container where insights can be distilled from. Of course, an important element is that the organization carries data science skills in its human resource base. The lack today is more around the skills area than any other. The lack is also around leaders and managers having the vision to stretch themselves and their employees to think differently than in the past.

For those of us to remember, it took years for our software developers (formerly generally called programmers) to move from procedural code to object oriented thinking. For decades, we would find procedural code being written inside of object methods. I wont say that this has gone away completely. I still see literally 100s of lines of code written inside of methods which by the very nature of object oriented thinking were meant to be short and sweet, focused and dedicated. Likewise, our leaders and managers mostly still manage operations like we did 20 years ago, despite the availability of tools that can help us do it differently. We still run 100s of test cases just because they are in the suite, even though we know that the likelihood of many of them in finding a defect is low to none. We still suffer what is known as ‘alert fatigue’ – which then leads to a real alert actually being missed. We still wait for the problem to occur before we can even begin to prepare for it.

Machine learning is a compelling tool. If leveraged in simple and effective ways, it can lead to cost optimization, effort optimization, focus optimization and in the end make for happier technology employees… and of course happier customers.

CP Jois