I heard about the term AIops term last week at The AI Conference, presented by O’Reilly and Intel AI. Is this a real practice? My answer is yes and here’s why.

Consider that DevOps is the practice to align developers and operations on the agility, speed, and stability of making software releases. DevOps aims to align a multidisciplinary team on conflicting missions and priorities by standarding CI/CD pipelines, increasing monitoring, and other DevOps best practices.

Then consider DataOps, the emerging discipline of align data professionals including data-driven business managers, data scientists, citizen data scientists, data engineers, data stewards, database architects, ETL developers, and DBAs to align on strategies and practices to ingest, cleanse, store, govern, manage, and deliver data and analytics to the organization and its customers.

Both DataOps and DevOps align multi-skilled professionals on mission, values, technologies and practices to achieve short term goals and deliver on longer term competitive value.

So here’s my definition of AIops:

AIops defines the mission, principles and practices that drive collaborative AI experimentation that delivers business results

AIops aligns key practices in the AI journey

AI also requires aligning multi-skilled professionals. In addition to AI specialists, it requires support from business managers, subject matter experts, data engineers, and technologists to align on mission, data sources, platforms and desired outcomes.

The team needs direction on experiments that drive business value.  This requires collaboration as defining overly bold experiments may be unachievable while missions that target marginal business value may not justify the investment.

To be successful in AI, the team needs access to a large volume of relatively clean data. The AIOps team then must take steps to tag data for supervised learning AIs or define reward functions for unsupervised problems. The practice of organizing and standardizing data for AI experiments is an AIops practice.

AI requires selecting platforms, tools, and infrastructure that needs to be ramped up and down as experiments are conducted. Teams can consider a multitude of platforms (TensorFlow, Keras, PyTorch, Caffe), cloud provider (AWS, Azure, Bluemix, Google), and a growing number of collaboration platforms (Dataiku, H20.ai, Databricks, Anodot, Clusterone and others) as part of their AI and machine learning environment.

With data ready, the team needs a working process for running experiments. I’ve suggested that agile experimentation is required for AI and the team needs to establish trackers to capture metadata and results of the trials conducted.

Once experiments are conducted, the results need to be analysed. The team needs to determine the success of the overall experiment and what follow-on experiments to prioritize.

When results yield satisfactory results, the team then needs to determine how to establish a production process to run new data through the AI models.

Why AIOps should be formalized

Organizations dedicating resources to AI experimentation recognize the journey that needs to be led by a collaborative, aligned team:

• From early stages where people, partners, and platforms are established
• Middle stages where the team develops its practices, grows data sets, and automates processing steps
• Later stages where agile AI experimentation begins to show results and production practices are established