Monte Carlo puts AI agents to work on data reliability

Big-data observability startup Monte Carlo Data Inc. is the latest technology company to jump on the agentic artificial intelligence craze, launching a suite of Observability Agents designed to automate monitoring and speed up incident response times.

The suite consists of two AI agents at this time, including a monitoring agent that helps out by recommending and implementing rules regarding data quality thresholds. The other is a troubleshooting agent that can investigate, verify, explain the root cause and suggest fixes for any data quality issues that arise.

Monte Carlo is the creator of a popular observability platform that helps companies to keep tabs on the “quality” of their data assets. It’s based on the same principles that guide application observability tools such as Datadog and AppDynamics, only it’s applied to data pipelines instead of app metrics. It works by using machine learning algorithms to understand the normal behavior of a customer’s data streams, so it can warn them if anything abnormal occurs.

Data quality is key to ensure that companies can trust their analytics-based insights, which are critical to business decision-making in today’s data-driven economy. It’s said that data is the new oil, after all, fueling enterprise operations. But it needs to be the right stuff, and that’s why Monte Carlo wants to reassure its customers with its new AI agents.

AI agents are a new kind of AI system that go beyond the capabilities of generative AI chatbots, as they’re programmed to take actions on user’s behalf with minimum supervision. By automating mundane manual tasks, they can free up human workers to focus on higher value work.

Monte Carlo said its AI agents are different from others because they don’t just make recommendations based on data profiles. Instead, they leverage a “sophisticated network” of large language models, native integrations and sub-agents to obtain full visibility into customer’s data estates.

According to Monte Carlo co-founder and Chief Technology Officer Lior Gavish, this extensive network is necessary because the capabilities of AI agents are defined by how well informed they are.

“Our AI agents can execute more sophisticated analyses that are truly useful because they are reviewing data samples to determine what the data looks like, metadata to understand the larger contextual meaning, and query logs to understand how the data is used,” he said.

Data quality rules

Generally available starting today, Monte Carlo’s monitoring agent promises to save data teams a lot of time that has historically been spent on defining, writing and deploying rules that are used as a baseline to monitor data quality. For instance, these rules include things such as a currency ratio field must never contain a negative value. If it does have a negative, that would indicate a problem with the data.

The challenge is that, in order to create these data quality rules, teams have to consider all of the ways that the data might “break.” In addition, teams need to get the threshold alert balance right, so it’s neither too noisy nor too permissive.

The company says its monitoring agent can automate all of this work, identifying sophisticated patterns and relationships across datasets that would likely be missed by humans.

“For example, the monitoring agent may identify that a product SKU ID field always starts with “950” followed by 4 unique digits for certain product categories and not others, or that a certain product SKU always has a higher order amount than another,” the company explained. “It uses context on how fields are used to prioritize and rank the most critical alerts – providing the most coverage with the least amount of noise.”

Should it ever spot deviations from these trends, it can file an alert to warn data teams of a potential quality issue. According to Gavish, early tests show its alerts and recommendations have an impressive 60% acceptance rate, helping improve monitoring deployment efficiency by 30% or more for the average customer.

Root cause analysis

As for the troubleshooting agent, this is meant to help human data engineers by dramatically reducing the time it takes to investigate data quality issues.

Monte Carlo says that’s necessary because data and AI systems are becoming more complex, with numerous interdependent moving parts. With so many components, it can take hours to investigate and identify what’s causing a problem when one occurs, and time very often means money when it comes to the enterprise applications that rely on this data.

When it launches later in the second quarter, the troubleshooting agent will be able to automate these investigations, exploring hundreds of possible causes across all of the relevant tables in the affected data to try and understand what the problem is. It will be able to identify if the problem is the result of bad data from the source, an extract, transform and load system failure, or an incorrect model output, for example.

Monte Carlo said this investigative process involves dozens of sub-agents working in parallel, enabling it to check every possible cause in a matter of minutes. The result is that teams can reduce the average time to resolution by 80%, it promised.

Constellation Research Inc. analyst Michael Ni told SiliconANGLE that Monte Carlo’s agent-based shift is rooted in its machine learning and observability expertise, mirroring the pattern of AI agents that have already transformed data analytics definitions and business performance monitoring.

“Monte Carlo’s new agents smartly define observation rules for increasingly complex datasets and accelerate data teams through interactive root cause analysis,” he said. “For enterprises, this means faster data fixes and greater reliability, critical for scaling analytics and AI.”

The company said this is just the start of its agentic AI push, with plans to enhance the capabilities of its AI agents in the coming months.

Image: SiliconANGLE/Meta AI