Organizations must show immediate benefits from their AIOps implementation, or else risk losing top-level support for it
Selecting the right AIOps platform is just the beginning. It’s crucial for the technology to be implemented quickly and efficiently, and to demonstrate value quickly. This is true for any major technology investment but it is particularly true of AIOps. Why?
AIOps, and AI in general, has in recent years been the subject of extreme hype. Its promise seems boundless. At the same time, it is poorly understood by those outside — and even inside — of the IT community.
AIOps should not be mysterious but it is to many
This lack of understanding is not so much a function of its novelty; in fact, AI is hardly novel, having been an object of study and even commercial and military implementations almost since the birth of computer science. It is more a function of its complexity, both in terms of what it is meant to accomplish and how it goes about accomplishing that goal.
Is it something like HAL-9000 or Skynet intending to take away our jobs or worse? Is it something mysterious and mystically born inside our hardware and software systems over which we have no control? Does it see the future? I have heard all of these questions asked — and asked seriously. But the problem is not with asking these questions but the levels of expectation they set.
The extraordinary and impossible to comprehend nature of AI ought to deliver massively positive results immediately, and any indicator to the contrary will burst confidence in the whole endeavor, like a pin stuck in a balloon.
What is AIOps?
For us at Moogsoft, of course, AIOps is not so mysterious. It is a category of software that can be explicitly characterized by its enabling of the sequential application of five distinct types of algorithm.
Data selection algorithms clear away data noise and redundancy and surface data items pointing to events of significance taking place in an IT environment. Pattern discovery algorithms find patterns that correlate those significant data items. Causal inference algorithms determine which of those correlational patterns are, in fact, actionable or causal in nature. Collaboration algorithms bring together teams of human and robotic agents that are capable of acting in response to the now correlated and causally analyzed significant events. Finally, automation algorithms execute the plans that emerged from the teams’ collaborative efforts.
That’s it. No more. No less. Still, AIOps is complex and even when explained in a way that does not invoke science fiction and mysticism, it can be somewhat daunting, particularly to the business professional that’s minimally IT literate. So the point we began with remains. Results are needed and fast. Otherwise, executive decision makers will lose confidence in the project and look for ways to wind it down.
Don’t boil the ocean
So what pitfalls should an enterprise avoid? First, don’t try to boil the ocean. AIOps has the five high-level components and, within each, many subcomponents as well — for example, Moogsoft currently implements three distinct types of pattern discovery algorithm.
Furthermore, unlike many technologies associated with IT Operations and Service Management (ITOSM), AIOps is not targeted at a specific technology domain. Its five types of algorithm apply to networks, infrastructures, applications, and end user interfaces in isolation from one another and in every combination. Attempts to do everything at once are bound to end in failure. There is just too much to get it all done in anything less than three years and that is assuming that all steps are taken without mishap.
Instead, enterprises should begin with one or two key business applications. Select two data streams being generated by the application itself — say application logs and events extracted from an APM tool — and combine them with data being generated by the network or infrastructure over which the application is running.
Once these selections have been made, the enterprise should examine where among the ITSOM processes it is feeling the most pain. If issues tend to cluster around availability and performance management, enterprises should consider focusing on pattern discovery. If issues tend to cluster around incident or problem management, enterprises should consider focusing on collaboration.
In all cases, however, enterprises should, at a very early stage, deploy data selection algorithms. These will yield an almost immediate and highly tangible benefit since Moogsoft deployments among our customers show that the typical level of noise and redundancy in the self-describing data sets generated by IT systems is somewhere between 90% and 99%. In summary, start small and focused, putting only those elements of AIOps to work on your most business critical applications.
Prepare for culture change
Another major pitfall for many AIOps deployments is the failure to appreciate and prepare for the culture change they trigger. A common complaint for the last forty years (really ever since the emergence of the minicomputer) is that IT is segregated by technology domains that refuse to communicate with one another. From time to time, solutions emerge that enable a ‘single pane of glass’ that can unite those siloed views.
However, the number of domains grows as IT evolves, and, very soon, the last generation’s single pane of glass becomes a new isolated domain unto itself. AIOps, since it creates unity in an algorithmic way, can provide escape from the constant multiplication of domains. Nothing in any of the five algorithmic types limits any of them to any existing or future technology domain. Hence, AIOps will guarantee a single pane of glass in perpetuity. That alone is a major culture shock. In a sense, AIOps will force all IT professionals to be technology generalists.
But that’s not all. AIOps not only integrates technology domains; it also integrates process domains. When IT departments got interested in ITIL in the early 1980s, there was much discussion about how from the perspective of each process described by ITIL, technology domains would need to be looked at through a common lens. Indeed, the “single pane of glass” technologies described earlier were to some degree inspired by the ITIL vision.
What was not fully appreciated then, and even 20 years later, was that well-defined processes themselves would carve up IT into small enclaves that communicated little if at all with one another. AIOps rips up the process structure and forces IT professionals to collaborate across processes. In fact, if one reviews the sequence of the five algorithmic types, one can see that it is a choreographing of the handoffs and interactions among those elements of ITOSM that receive signals from the IT environment to those elements that respond to those signals.
AIOps will force all IT professionals to regard their rigid process roles (which have been drilled into them for decades) as being fluid and boundaryless. Since this is a question of modifying what they do and not just what they are expected to know, the culture change brought about by this ‘bonfire of the processes’ is even more significant than the one brought about by the ‘bonfire of the domains’ discussed above.
The recognition ITOSM deserves
So the culture change is great. Unless prepared for it, individuals will consciously or unconsciously resist and even work against the success of an AIOps project. Hence, it is critical to recognize the magnitude of transformation that a successful AIOps endeavor will ultimately bring about. This means, of course, thinking through up front not just the business value of AIOps but also how the individual ITOSM staff member stands to benefit.
And what is that benefit? Luckily, that is pretty easy to state. IT has become extremely and necessarily complex as it has evolved to support digital business, and that complexity will only increase in the future. Without AI, that complexity will be impossible to manage. Once that complexity is managed, however, the direct line between IT and business value will be clearer than ever. ITOSM staff will cease to be viewed as workers in the engine room and come to be recognized, at last, as the key enablers of the revenue-generating digital business infrastructure.
About the author
Will studied math and philosophy at university, has been involved in the IT industry for over 30 years, and for most of his professional life has focused on both AI and IT operations management technology and practises. As an analyst at Gartner he is widely credited for having been the first to define the AIOps market before joining Moogsoft as Field CTO. In his spare time, he dabbles in ancient languages.