Artificial intelligence for IT operations, more commonly known as AIOps, is a hot topic these days but few seem to talk about the path to AIOps. In order to implement AIOps successfully, it’s important that businesses understand the process and the tools required at each stage. With the start of the new year, now is a great time for businesses to consider the integration of AIOps into their IT programs.
What is AIOps?
AIOps is the application of machine learning (ML) methods to sift through incoming data and identify the critical information allowing for more effective management of IT Operations. In many cases, AIOps can identify a problem before it impacts the end users of the application.
This is especially important for revenue-generating applications. A recent study of 1100 organizations found that the risk of not evolving to AIOps is $1.2M on incident escalations that could often be avoided.
With its advanced capabilities, AIOps can quickly identify and resolve potentially damaging problems, often before it impacts users, by identifying other conditions or situations that precede the problem. Enterprises put AIOps in place to analyze IT operational data to augment and automate all major IT operations.
The Path to AIOps in 2021
The road to AIOps begins in 2005 with the introduction to the concept of big data. Then, in 2014 analysts created the term IT operations analytics (ITOA)—the application of big data analytics to IT operations data. ITOA provided the analysis and reporting of huge volumes of IT operations data to provide a big picture of a business’s IT estate and help enterprises make more informed decisions.
It wasn’t until 2017 the term AIOps came into the lexicon as the need for organizations to assimilate all their data in real time grew. The term continued to evolve in 2019 with the addition of “multi-domain AIOps” to account for the differentiation of domain agnostic vs domain specific solutions and the concept of an AIOps overlay of existing tools. Since the onset of COVID-19 has accelerated the move to the cloud, the need for AIOps has only increased.
Business leaders need to recognize the benefits AIOps provides and its success in better managing company IT operations and infrastructure. Below are five steps businesses can take to start integrating AIOps into their IT programs and start 2021 with enterprise automation.
Ensure AIOps aligns to business goals. AIOps is a multi-domain technology. Recent research found it supports, on average, eight different domain-specific roles and 11 cross-domain roles. For a project with a large scope it is essential to get executive buy in and support from C-suite leaders. A good way to gather this approval is to set up a small test project to solve a single pain point. From there, IT professionals can leverage this success to recruit the executive teams support and alignment with further integrating AIOps into the enterprise.
Observe and incorporate everything. A core concept in AIOps is to pull data from everywhere—networks, servers, applications, etc. The reason for this is twofold—AI loves data (the more the better), and the more domains covered the more likely it is to find the source of the problem faster. For example, user experience monitoring might signal a performance problem before it shows up elsewhere in a noticeable way. When AI has that data, it can look at the rest of the information to find what infrastructure item might be the source of the problem. Incorporating all enterprise assets into an AIOps program is therefore essential when starting out.
Integrate domain-agnostic AI. After businesses have conducted a holistic integration of IT operations data – IT professionals should consider domain agnostic AI as an appropriate next step. Many domain tools these days provide AI capabilities but in their own silo. While domain tools provide some benefits, they often fall short of a desired solution. There is a benefit when a server monitor tool’s AI flags something amiss, but with most problems there is a root-cause event with multiple symptomatic events. AI in one domain won’t be able to correlate events outside its domain, so operators are forced to investigate the symptom alerts until they are fine the correct one, which can be time-consuming and inefficient.
Instead, enterprises should seek a first-time fix by employing a system that points to the correct problem with the first alert it shares. This is a huge time saver on many fronts, as the problem is identified faster requiring less employee intervention, and fewer tickets raised.
Centralize the data lake. When businesses have siloed AI they are unable to have a comprehensive view because each dataset is in its own tool. Typically, the next step is creating a centralized data lake to better integrate AIOps. Using a central data lake where all data is kept provides multiple advantages—for example, ML can process data from multiple domains and the ability to baseline data from a domain tool that doesn’t baseline. It is much easier to construct displays that show cross domain data. Additionally, when AI has all the data in one place, analysts are able to examine it to identify patterns and issues that can lead to more informed decisions that will benefit the business.
Move toward connected AI automation. The goal of implementing an AIOps solution is to find and fix problems faster. One user started by interviewing teams to find their common repeatedly occurring problems. Once they had the list, they used the root cause from ML and connected it to a workflow they found or built that would fix the problem. In the first year they saved $1 million, in the third year they saw $4 million in savings. Connecting ML problem analysis to an automated workflow that can fix the problem should be a significant element of any AIOps program.
2021 is the Year
Undoubtedly, 2021 is the year enterprises should focus their attention on progressing AIOps as a key component of better IT application and infrastructure management. Where to start on the AIOps journey is different for each business, and the order of the steps may change, but most importantly, the aforementioned process is necessary to ultimately see success.