Way back in 1978 there was a Clint Eastwood movie called “Every Which Way But Loose”. That was a lifetime ago technologically, but the title is a great way to sum up the frequent and scattered use of the term AIOPS in the ITOM marketplace today. It seems we are in the upward trajectory in the hype cycle in relation to vendor’s touts and mentions of their AIOPS capability.
I have listened to multiple vendors mention the use of artificial intelligence, machine-learning, and AIOPS in presentations, marketing briefs, and industry events, and clearly the marketing gurus are playing a leading role in pushing the term. Even the term itself has seen shifts in its meaning. AIOPS was originally coined by Gartner in 2016 to be short for Algorithmic IT Operations, but just a year later was updated to denote Artificial Intelligence for IT Operations. Make no mistake — I believe that AIOPS is an important step in the evolution of enterprise management systems and capabilities. The explosion in monitoring data sources, velocity, and the need to ingest at rapid rates, combined with the virtualized, ephemeral nature of compute resources today have created a major problem in problem identification, correlation, and root-cause analysis. Solving these problems transcends past approaches based on rules, static topological relationships, or inference engines. It is largely a big data problem that must be solved by applying algorithms, analytics, and machine-learning concepts at the underlying telemetry. That is the promise of AIOPS, but we are at the outset of a shaking out period as the platform-based solutions are competing for our attention in showcasing their abilities to solve that problem. With that, it is important that we scrutinize and understand what these next generation platforms are offering, how they do it, and ultimately how that fits and addresses the problems we are trying to solve.
Given the push to capture attention through use of the term, customers must look to try to separate fact from fiction. What is called AIOPS can have a broad interpretation, but generally should include machine learning, anomaly detection, root-cause identification, predictive analysis and performance baselining. Many so-called AIOPS vendors offer no more than the existing rules and behavior model-based approaches from their legacy products. Other vendors have emerged with new platforms built for purpose around an algorithmic approach. The vendors arrive at their solutions in different ways, and may tilt their messaging toward what they have “under the hood”. Therefore, customers should examine how various AIOPS solution derive their outcomes — showing by what steps the data transforms into a result. The mix of algorithms that are used (or purpose-built and created in some cases) will ultimately play a significant role in the outcomes. Algorithms can be patented, so the ones utilized from vendor to vendor will necessarily differ, and therefore the outcomes. And then, some algorithms require a training period in which to identify patterns and “tune” the system. Other approaches purport to engage in immediate pattern identification. These factors will ultimately impact the results and the platform’s ability to realize the problem(s) it was built to solve.
AIOPS is not magic but it does hold promise for solving some long-standing and vexing issues in service management. It is being pushed as a solution for a number of issues valued in IT Operations Management, such as event correlation, automated problem response, and incident management among others. The first step a customer should undertake is to determine the problem(s) they are trying to solve. Not every customer will be in the same place with the issues before them, the management infrastructure they may already have in place, and their priorities for service management and IT Operations. For some enterprises, it may benefit them to migrate to a brand new platform that directly addresses their mix of data sources, infrastructure, and applications. For others, an upgrade to their existing management platform that leverages algorithmic operations to improve their service management objective(s) may be the right approach at the time.
To be sure, there will be continued momentum toward an AIOPS approach. The market for AIOPS is expected to grow from $4 billion today to $11 billion by 2024, with enterprise adoption surpassing 50% by just 2020 alone, according to Gartner. But given the “noise” surrounding all the vendors rushing in with their AIOPS claims, customers should exercise caution, evaluate the claims, and compare against their own real needs. In due course, the products that truly deliver on their promises, such as increasing event management accuracy, automated problem resolution, and better root-cause identification will rise in prominence, and market share. In the meantime, do your homework on vendor’s positioning of their tools and capabilities. As I recall, there was a sequel to that Clint Eastwood movie…it was called “Any Which Way You Can”.