Sergey Nivens - Fotolia

Get started Bring yourself up to speed with our introductory content.

Decipher real-world AIOps use cases from the hype

There are many AIOps use cases that apply to enterprise IT, but make sure you are familiar with what exactly the technology can -- and can't -- do before adoption.

AIOps -- a term IT pros likely heard a lot in 2019 -- refers to the use of machine learning and AI technologies to streamline IT operations tasks. But while the possibilities might seem endless, AIOps has its limitations.

In this SearchITOperations Q&A, Charley Rich, a research director in the IT Operations Management group at analyst firm Gartner, discusses common AIOps use cases and misconceptions. In doing so, Rich provides a glimpse into some of the topics he'll touch on during his session "Mythbusters: AIOps and What it Really Means for IT Operations" at the Gartner IT Infrastructure, Operations & Cloud Strategies Conference, running December 9-12 in Las Vegas.

Editor's note: The following interview has been edited for clarity and brevity.

What are the biggest AIOps myths or misconceptions that you've heard?

Charley Rich: One misconception is that AI is almost magical, and it can think about everything that's wrong, and it'll automatically fix everything without a lot of human involvement. And that's a bit of a stretch, because that will probably take us years to be able to do that, if ever.

A second one is the conflation or confusion of AI and automation: AI can be used to intelligently decide what automation [does] and how to run it, but it, itself, is not automation.

What are the common challenges enterprises try to solve with AIOps platforms?

Rich: It all depends on what [the organization] is trying to do. The basic problems commercial AI platforms are marketed to solve are generally [to decrease] event noise, improve signal-to-noise [ratios] of events and reduce false alarms, number one. However, we are starting to see it applied to metrics and traces. Number two, doing anomaly detection: Usually, the better solutions for anomaly detection are multivariate, with a notion of seasonality -- automatically detecting seasons and determining that the condition has occurred in a specific time period of the day or day of the week. Alternatively, anomalies can be seen across [device types], such as load balancers. And then the third area is causality, with the capability to use AI [and] machine learning to come up with better root cause analysis.

These platforms have been most successful with event correlation and analysis. An old problem that used to be solved by rules is now solved by math, [with] statistical and probabilistic algorithms. And that's kind of where the situation is today.

Of the AIOps use cases you just described, which are well established and which need further development?

Rich: Well, those cases that I mentioned are what vendors offer today, but there's much more that can be done. The ones that we've talked about are mostly in the area of observation, or monitoring. There are a number of other use cases from the area of IT service management. And in that area, [enterprises] are using machine learning to analyze the effectiveness of service desk personnel [and] the resolution of incidents, and then are using natural language processing with chatbots. These help share the corporate knowledge base both with workers internally and with customers. And then VSAs -- virtual support assistants -- are used to help front-end automation. Think of it as a democratization of automation where, for example, a VSA could greatly simplify the process to invoke the CI/CD process without requiring the user to know all its intricacies. Those are the two main areas.

We'll start to see more of AIOps applied to the entire lifecycle of an application.

And then there is a third area, which is around intelligently driving remediation. It's the intelligence to help companies use automation to resolve a problem. And while there's a lot of talk about that capability in the industry, it's still a very nascent and primitive capability. There are many other potential use cases that could be done. Take the patterns -- the pattern recognition, or clustering -- that machine learning can provide and use it to find [important insights into] customer loyalty, orders, risk and compliance, as well as other business outcomes. But we don't see much of that today in the solutions that are offered in the market. In the near future, we hope to see more AIOps applied to the customer journey, and digital experience.

Your session focuses on myths. What are the potential dangers of AIOps being overhyped?

Rich: Well, some of the danger is that a business purchases a solution based on the myth before they figure out the problem they are trying to solve. It's sort of like the cliché of holding a hammer and thinking that everything is a nail. The other myth is that if we have a whole bunch of data, there must be something useful in it. [Enterprises think,] 'Let's turn AIOps on the data, and something good will come out.' However, we've never seen that work -- ever. The better approach is to understand the problem and then apply the right set of solutions to resolve [it], which may or may not be AIOps.

There's a set of skills or expertise called data science, which can be helpful to understand how to clean and provide proper hygiene for data so that it can be effectively analyzed. Many large companies have that skill, but it's not generally available within infrastructure and operations [I&O]. We believe over time this should become a recommended or required skill that new I&O hires have.

Which industries will benefit the most from AIOps?

Rich: I don't think AIOps is really a vertical solution. I think instead it's most effective for enterprises that are building what Gartner calls Mode 2 applications: [Applications] that are highly agile, related to customer experience and revenue, and change frequently. An example might be to apply AIOps to release management when a DevOps group is deploying multiple versions of an application every day. AIOps could be used to help determine the risk in producing unintended consequences as a side-effect of deploying the build. This can enable the business with the ability to decide if it is worth doing.

How can we expect to see AIOps evolve in 2020?

Rich: I think the trend towards segmenting the AIOps market into domain-centric and domain-agnostic solutions will continue. We'll start to see more of AIOps applied to the entire lifecycle of an application: analyzing code in application development, builds in DevOps, applications in production and engagement in IT service management. We'll see more of it applied to identify possible risks that occur in the lifecycle of an application and especially into analyzing the risk in change management.

Hopefully, by the end of next year, solutions will be available that provide value quickly. And we'll see more enabling in ITSM of natural language processing to help with bots for customer relations, customer improvement, engagement, CRM and improving customer satisfaction. AIOps will become the enabler of observability, engagement and the intelligence behind automation throughout I&O.

Dig Deeper on Managing Cloud-Native Applications

Join the conversation

1 comment

Send me notifications when other members comment.

Please create a username to comment.

How do you think AIOps will evolve?
Cancel

-ADS BY GOOGLE

SearchSoftwareQuality

SearchAppArchitecture

SearchCloudComputing

SearchAWS

TheServerSide.com

SearchDataCenter

SearchServerVirtualization

Close