How to Spot AIOps Snake Oil

Step right up folks! Don’t be shy! Be prepared to witness the marvel of Artificial Intelligence for IT Operations – or AIOps for short. See how powerful machine learning algorithms can resolve issues faster, and stand in awe as you watch deep neural networks predicting future problems in real-time.

Hurry, hurry, hurry!

I jest, but it’s easy to get carried away with the hype and hyperbole surrounding every new technology – especially those where the funky and often esoteric nature of the subject matter, coupled with the very real challenges it purports to address, allow vendors to pimp up some magic cure-all product – snake oil.

AIOps, with the focus on utilizing big data, machine learning and other analytical goodness to enhance processes under the purview of IT operations is now gaining close attention. With the promise of freeing up already overburdened teams and shifting cognitive capacity from the cost-centric mundane towards value-generation it all makes good sense. Not least because we’re well past the point where human brute force and heroic endeavors alone are enough to wrangle the massive complexity of today’s modern digital systems.

But against this backdrop its easy to get carried away and succumb to exaggerated claims about new product capabilities. After all, analytics is all dark science kind of guff, so it’s natural to believe what’s written on the label, take a swig from the bottle, and wait for the magic to happen, right?

Well maybe not, but before trawling through the marketing fodder from a growing list AIOps vendors (23 and counting according to Gartner’s, 2018 Market Guide for AIOps Platforms) consider asking these four key questions before you sign-up and drink the cool-aid:

• What’s actually in the bottle? – be on the lookout for classic claims about analytical black boxes that don’t require data science expertise. Sure, we shouldn’t need to become math whizzes to reap AIOps benefits, but it’s still important to understand what algorithmic recipes your vendors are proposing and their applicability to your operation. Consider too that many machine learning methods aren’t new at all, so give marks to vendors who’ve applied proven field-tested methods (not necessarily in IT) and have the chops to explain when, why and how they work.
• How many quality ingredients? – by nature, AIOps machine learning – where the software learns and improves – is dependent upon crunching vast amounts of data for any task it’s learning to perform. Take for example, AIOps software that’s being trained to classify optimum cloud instance configurations based on demand patterns, or learning which coding practices correlate to the best sales outcomes. These can only get better when the math is given more diverse and contextualized data to process. It’s important then to consider collection inhibitors that prevent richer insight together with costs associated with data capture and storage.
• Can it help cure invisible illnesses? – a goodly proportion of AIOps work has been focused on developing statistical methods to find signals in the noise - the false positives. That’s great, because old static rule-based thresholds using single metrics no longer work for complex and systems. However, to be more valuable, AIOps solutions must advance beyond basic event correlation, and towards methods that yield deeper insights within today’s complex systems that can and will perform in unpredictable ways. To this end, I’d seriously consider basing any AIOps product evaluation not only on its ability to tell you what sort of know already, but also how it can identify unexpected and unforeseeable conditions – the unknown unknowns – that’s where an organisation will almost always get the greatest payback.
• What are the side-effects? – while nascent, AIOps has the potential to simplify, accelerate, and improve many aspects of IT Operations. That said, perhaps the greatest benefit is how the technology will cause IT leaders to rethink and re-calibrate organizational structures, re-evaluate existing IT operations practices, and develop new processes. One perfect example is help desk support, where many organizations are replacing inefficient and costly three-tiered models with dynamic mechanisms in which teams collaborate (or swarm) around complex problems and their resolutions. It’s within this type of modern process that AIOps can really shine – especially since an optimum swarming model is conditional upon precise root cause determination and contextualized intelligence.

It’s easy to get warm and fuzzy about modern tech, especially when it’s tagged with an AI moniker. But don’t be fooled by grandiose claims about a products AIOps capability until a vendor can clearly explain and demonstrate its value within your own operational context. Use the four pointers above to help determine what’s real and beneficial in all the marketing make believe and always be ready for snake oil.