Machine Learning – “A manager’s dilemma; do more with less”

This post is one in a series of how DeployPartners sees the future of Assurance/ITOM with the introduction of Machine Learning capabilities. In this post, we make an argument that every product in the very near future must include Machine Learning capabilities out-of-the-box to be competitive.

How operations will do more with less

Deploypartners has a long history with OSS and BSS tools dating back 25 years in overall experience. Most of the tools that have been around for a longer period use similar technologies of providing scripting or SQL tools to manage fast amounts of data through linear logic. This approach provided successful deployments and strategies where technology was relatively static and conventions were followed strictly. As soon as the environment changed (infrastructure, services, platforms) the linear rules required updating and would often interfere with existing logic.

Current customer environments have ever more dynamic infrastructure (SDN, NFV, SD-WAN, IBN) and dynamic platforms (Serverless, Cloud, Virtualisation, Containerisation). Managing and monitoring these platforms requires more sophisticated tooling and techniques to provide insights into the performance and workings of their services and platforms.

The vendors DeployPartners are working with are all weaponising their products with AI to support more sophisticated use cases. These AI enabled tools support the operators and do not replace them. The most common application of AI currently is for these tools to look at signals in large data volumes and requires operators to provide feedback on the validity of the signal.

DeployPartners’ solution-focus is shifting from providing customised solutions to manage complex infrastructure and services, to helping customers analyse results from Machine Learning processes and identify true-positives from a large data set. The true-positive results are then codified into the management system and the system is able to identify the complex scenarios as they play out.

Implementing these non-linear patterns requires Machine Learning algorithms as well. Techniques like key value pair filtering are backed by pattern identifying algorithms. Where the pattern can be time-based, frequency based, or natural language based.

Machine Learning to reduce incidents

Machine learning is also utilised when reusing knowledge from historical data patterns. Algorithms allow for indexing and tagging of data passing through the system by recording operator actions and providing suggestions when similar data patterns are occurring in the future. The benefit of this combination of big data and Machine Learning is that the period between the occurrences of the patterns can be large and the operators can be different. The knowledge is retrieved and presented to the operator as a suggested workflow as it has benefitted a previous operator. The operator can then determine if that suggested approach makes sense in the context of the current condition.

The end state of this analysis and implementation of non-linear pattern identification is to codify operator interactions into process automation tools (Ansible, Resolve) and robotic process automation (Automation Anywhere, Blue Prism). We have found that customers have significantly reduced their average Mean Time to Repair (MTTR) overall and, in particular, reduced the number of less complex incidents. This frees up operators to have more time troubleshooting and fixing more complex incidents and provide a better customer experience – therefore providing better customer focus and interactions.

Manage larger infrastructures with smaller teams

With AI modern tools are helping Operations and Service managers to keep control of more complex and larger environments with similar team sizes with less tools. Even though the new products do not necessarily reduce the TCO we do see that they provide more business value than the classic tools.

One of the key reasons why AIOps is taking the market so quickly is that, compared to traditional monitoring stacks, they can provide a very short time-to-value. Through the use of open-source products and industry standard protocols (RESTful, JSON, and XML), deployments and integrations are shortened significantly. Delivering a minimal viable product can be achieved in order of weeks to months instead of months to years.

From DeployPartners’ perspective, future vendors will provide more business value through the use of AI technologies like ML, Big Data, and Neuro-linguistic programming (NLP). Customers should acknowledge that system integrators will provide more blended Business and Engineering Analysis services to get the most out of tools with AI capabilities.

Considering Machine Learning, AIOps and virtualisation in a hybrid or mixed vendor environment? Get in touch – we can help.

Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.