Considering AIOps? First Check What's Under the Covers

While travelling across Europe as a cash-strapped student and frequenting some less than salubrious hotels, I learnt the importance of always checking under the bed and sheets before hitting the hay.

When selecting technology, we also have a tendency to jump in before checking what’s under the covers and getting bitten later. It's not helped when technologies are nascent with vendors attempting to amp up their wares with hype and hyperbole. One example where I see this playing out is Artificial Intelligence for IT Operations - or AIOps for short.

According to Gartner, who can be credited with coming up with the term, AIOps platroms:

... utilize big data, modern machine learning and other advanced analytics technologies to directly and indirectly enhance IT operations (monitoring, automation and service desk) functions with proactive, personal and dynamic insight. AIOps platforms enable the concurrent use of multiple data sources, data collection methods, analytical (real-time and deep) technologies, and presentation technologies.

Now while that's something of a mouthful, the application of artificial intelligence (AI) to ease the burden on a workforce increasingly challenged to maintain customer experience across highly complex digital architecture seems a no brainer. With the advent of cloud and it's associated technology, IT operations are now faced with managing what can only be described as 'complex systems' - systems that act and behave in ways never anticipated. Simply put, we've moved from managing known conditions to a world of "unknown unknowns." So, if algorithms can be trained to spot anomalies in medical images 1000x faster than a medico, or accurately predict mortgage loan defaults, then surely the challenges faced by IT operations can be easily resolved with a quick investment in computational brute force and math smarts.

Of course new and old software vendors will claim to have the answers. Many are refreshing existing monitoring and IT service management products with fresh coats of AIOps paint, while new entrants are releasing innovative, but as yet unproven solutions. And the market is dynamic, with many vendors jostling for a share a share of the AI pie with nifty acquisitions. Like for example, New Relics recent acquisition of SignifAI and Coscale, plus VMware buying Wavefront, Cisco snaffling Appdynamics, ServiceNow with DXcontinuum. Even Atlassian is dipping its toes in the monitoring pond with its Opsgenie buy. Watch this space - more will follow.

So putting exuberance aside, how should organizations go about selecting an AIOps solution?

Well, start by looking under the covers. Here are some pointers:

• Seek out Vision - there's something of a sameness about many vendors AIOps musings - reduce alert fatigue, detect anomalies, pinpoint root-causes etc etc. Don't get me wrong these are important functions, but they hardly convey any sense of vision. It's important therefore to understand how a vendor sees something as powerful as AI materially changing what's traditionally been a cost scrutinized discipline (and the people that support it) for the better. While I'm not predisposed to any one vendor, ServiceNow and their 'future workforce' messaging shows good understanding of AI value.
• Watch for AI Babble!- machine learning, deep neural nets, natural language processing and robotic process automation easily roll off the tongue and look great on PowerPoint. But I'd suggest keeping the 'oohs and ahs' in check by first understanding the applicability of AI methods to your operation, and secondly, determining which vendors are investing actual data science behind all the marketing. There are obvious places to look like patent lists and open source contributions, but also seek out blogs and articles where vendors are prepared detail their AI approaches.
• Beware of Black Boxes - many vendors are augmenting old rule-based alerting with supervised and unsupervised machine learning, yet are often loathe to identify their methods and all the secret math sauce. But to correctly operationalize AIOps it's important to ask vendors tough questions, like - (1) how long will it take to fully train an unsupervised machine learning method across highly complex systems and what data is needed to support it? (2) how does the vendor develop, detail, and expose AI algorithms to its customer base? (3) how is the product moving beyond static dash-boarding to real-time interception and remediation of anomalous conditions?
• Consider the Footprint - everyone talks up the data science, but algorithmic coding snippets are in fact just a small part of an AIOps solution. Behind all the data science, new AIOps platforms may comprise a data lake, ingestion and streaming technology, plus search, log processing and a graph database. While this tech is needed to support the speed, scale and data contextualization requirements of AI, it will introduce new management overheads. Always consider that with AIOps the cost of predictions should be falling, so weigh up all factors when looking to determine a solutions ROI.

With the growing complexity of applications and the high-stakes of running a dynamic digital business, the application of AI to address a growing list of IT operations challenges makes perfect sense.

But don't let shallow data science speak fool you. Any AI business project comes with it's own set of technical and organizational challenges. Seek out AIOps vendors who're prepared to show and explain what's under the covers. Do this and you can start to exploit the collective power of human/machine intelligence - reshaping and redefining IT operations. Fail and prepare to be bitten by added cost and the sting of disappointment.