Responders are better at resolving production concerns.

Responders are better at resolving production concerns.

Consider the following scenario: Your team is notified when a new incident occurs in production. A rising number of individuals join the conference bridge or the Slack channel to troubleshoot. That appears to be a lot of discussions as well. Everyone is striving to pinpoint the source of the problem.

However, it rapidly becomes apparent that considerably more people are working on this than is required. And I’d have to guess. I’d say there are 50% more people than are needed. There is too much noise and insufficient knowledge regarding the incident’s core cause. Having so many people engaged is inefficient and increases the time it takes to resolve this situation.

As an incident first responder, my primary responsibility is to figure out what went wrong and where it went wrong, so I try to enlist the help of the appropriate people. It’s not always possible to respond to these queries right away, so we bring in more people in the hopes of quickly collaborating and mitigating the situation. That, however, is never the case. It reduces the team’s production and efficacy in the long run.

In this blog, I’ll explain how to use a feature in our Applied Intelligence product to address incidents and fix production issues faster and some technical specifics about how it works. A feature called suggested responders is part of the Issue Intelligence capabilities. It augments every new incident by selecting the most relevant team members to help resolve it in real-time.

If you’re a customer of Incident Intelligence, suggested responders are available without any configuration or setup.

Machine learning powered

Will augment future events in real-time with those best equipped to help resolve them as soon as the model training is completed—which will happen when more users interact with the system. The list of responders is included in the issue notification payload, so you can see it right where you usually reply to issues.

You can provide input in a thumbs up/down vote for each recommendation, allowing the model to grow even more accurately.

​

How the machine learning model works

The proposed responder’s algorithm combines historical violations with user analytics data to predict the most likely responders for new infractions. Supervised pattern recognition, label spreading, and a recommender engine are the three phases of the model. Responders are better at resolving production concerns.

Creates a tagged dataset with only those infractions for which the user who closed them is known. After that, we train a binary classifier that considers the users’ actions as features and determines whether they completed the violations as the target. In the next stage, suggested responders will use this trained model.

Label spreading: This stage aims to expand the coverage of our labeled dataset by adding infractions for which the user who closed it is unknown. We create an unlabeled dataset that connects each violation to the user behaviors that took place while the violations were open. The unlabeled dataset is subsequently fed into the trained model. This stage produces a responders table that assigns the likelihood that the user closed the violation for each violation-user pair. In the next phase, we will use this table to determine which users are most likely to fix a specific infringement. The table is updated regularly to reflect the most current infractions.

Recommender engine: The model’s third and final step leverages the responder’s table developed in the previous stage to propose new infractions of responders. When a new violation occurs, the model generates a similarity score between it and the breaches in the responder’s table. The similarity score is calculated using a variety of factors, including the type of product, target id, policy id, condition id, golden signal, and violation components. The similarity score is then used as a weight to construct a weighted score for each user across all of the table’s violations. The (predicted) level of participation a user had with similar breaches in the past can be understood as the weighted score. The model returns the users with the highest weighted score. The suggested responders are listed below.

How to use suggested responders to route incident notifications

Now that you understand how suggested responders function, you can apply it in new ways to improve your response efficiency and handle production issues more quickly. One method to accomplish this is to configure a pathway logic that routes alert notifications to a specific channel whenever it anticipates certain users as suggested responders for a given situation. It lets you ensure that the appropriate persons are notified of the relevant issues.

Get started now and resolve production issues faster

To get started with suggested responders, all you have to do is sign up for Incident Intelligence and start ingesting your violations. About 24 hours after opting in, the model trains itself and makes recommendations. The more you participate, the more suggestions you’ll get.

Enteros

About Enteros

Enteros offers a patented database performance management SaaS platform. It proactively identifies root causes of complex business-impacting database scalability and performance issues across a growing number of RDBMS, NoSQL, and machine learning database platforms.