Gartner placed AIOps (artificial intelligence for IT Ops) platforms ‘on the rise’ on their Hype Cycle for I&O Automation in 2019. It is only a matter of time that it moves to the ‘at the peak’ stage. As artificial intelligence and machine learning practices evolve, AIOps as a practice is also expected to grow. We anticipate that AI will mature rapidly over the next decade becoming a transformational force for any enterprise. And AIOps will play a significant role in that transformation.
But, AI isn’t there yet. To completely leverage the opportunities and returns that AIOps can deliver — some of which we outline in this earlier blog post — enterprises need to look beyond the hype. AIOps adoption needs to be far more than jumping on the latest bandwagon. It needs to be a strategic initiative, built for the long-term.
In this blog post, we’ll outline key considerations you must keep in mind while adopting AIOps.
True success of AIOps is in its ability to deliver at scale. Therefore, the long-term goal of any AIOps initiative should be to bring the entire enterprise on to a single platform, and automate as many processes as possible. A bottom-up approach of taking one process at a time and automating it will leave you with multiple inefficiencies. AI adoption will remain piecemeal, crippled by the very siloes you wished to eliminate. Moreover, you will be automating processes that were designed for humans, which might not be best suited for machine learning.
Therefore, to make the most of AIOps, you need a top-down approach. Begin by performing a thorough assessment of all assets, applications, processes and structures. Identify which of these can be re-wired for AIOps. Pick use cases that will have maximum impact, and begin your pilot there.
Once you’ve identified the right use case for your pilot, prepare your data. Get an organization-wide process to gather data across application, infrastructure, business and user information. Also get as much historical data as possible for the AI engine to learn from. At this point, the quality of your data might be lower than ideal. Identify areas where you have quality issues and find ways to address them effectively.
In parallel, put together a team that can work will power your AIOps initiative — process engineers, data scientists, application leaders etc. who will come together to set the vision and business-level goals for the success of your endeavour.
This is perhaps the biggest and most important step in your AIOps adoption journey. There are several platforms in the market that allow you to bring AI for specific processes such as helpdesk automation or DevOps automation. There are also general-purpose platforms that can apply to any part of your IT organization.
Considering the maturity of AI tools today, we encourage enterprises to choose a platform that will be adaptable and expansive enough to accommodate the future needs of your organization. An enterprise-grade AIOps platform should be able to do the following:
While shortlisting platforms and comparing them, keep the following in mind.
Identify possible data types and file formats available in your enterprise and ensure your AIOps platform can process that. Depending on the kind of data you generate — code, logs, text etc., to IVR, emails and spreadsheets — how often you generate data and how dynamic it is, choose a tool that can handle that variance.
Ensure that your AIOps platform can integrate seamlessly with your existing tools such as those for trouble ticketing, IT service management (ITSM), DevOps, cloud management, project management etc.
Outline your long-term goals and choose the platform accordingly. If data quality is a matter of concern for you, you might need a platform that has data cleansing and harmonization capabilities. If C-suite decision-making is your primary goal, you will need a platform that has engaging data visualization capabilities.
It also goes without saying that an AIOps platform should also be able to dynamically scale, and adapt to the organization’s needs.
Just because we recommend a top-down approach with a holistic view, you don’t have to adopt AIOps all at once. Take it one step at a time. The first step would be induction — feed historical data to the AI engine, for it to learn about your business, environment and current state. Here, you will define metrics, build dashboards and build reports. Then, get the AIOps engine to build machine learning models for pattern discovery, anomaly detection etc. with the historical data. Post this phase, connect your AI engine to real-time data sources for predictive analytics, root-cause analysis etc.
Remember the AI engine is only as good as the volume, variety and velocity of the data it gets. Ensuring that there is enough accurate and good quality data is an important part of making the most of your AIOps platform. By rushing to the last phase, without ensuing the efficacy of the historical data, you might be setting your AIOps initiative — and yourself — up for failure.