Sometimes the best defense is a good offense. That’s never truer than when it comes to data quality and reliability.
The classic approach to data quality monitoring arose in the data warehouse and business intelligence (BI) era. It went something like this. An alert pops up or, worse yet, an end user calls with a problem. The data engineer springs into action and starts to investigate to see whether the issue can be traced back to poor data quality or some other cause.
This reactive, defensive approach worked fine when end users “didn’t mind”, there was time to re-run the process, and you were sure all issues would be caught. Unfortunately, that’s rarely the case today, and with each incident end-user trust in data and the data team is eroded. Companies need tools that help them go on the offense and take a proactive approach to ensure data reliability by predicting and preventing issues before they occur.
Alerts are Too Late and “Slow is the New Down”
Today, data-driven organizations are leveraging data and analytics throughout operations in near real-time, with automated processes and embedded into applications.
You simply can’t interrupt the business while you troubleshoot issues. Moreover, slow data can be almost as bad as failed processes, since it forces employees, customers, partners, suppliers and others to either wait for information or give up and rely on guesswork.
Every organization typically has some basic checks that they implement to catch obvious and/or critical failures in data quality. These “usual suspects” often include missing data or wrong format. However, today's data can often be more unpredictable. Examples include:
- Distributed data. Data resides in many locations and technology environments, often beyond your control. You can’t always assume that it will arrive as expected.
- Change management gaps. The structure of data can change unexpectedly--schema drift--breaking downstream processes. This is increasingly a challenge as organizations leverage data that’s external, unstructured, or dynamic in nature.
- Good data is not enough. Data can be of high quality and yet still yield poor results. Many analytic models are built on historical data, and as the world changes, they become less accurate. Monitoring for trends and anomalies--data drift--ensures that models get retrained as needed, ensuring consistent accuracy.
These are just a few examples of why only monitoring for the usual suspects doesn’t cut it and why traditional data quality tools and approaches don't ensure data reliability.
Death by a Thousand Cuts
Traditional approaches to data quality often require technical expertise, domain knowledge, manual work, and a lot of time. With so many things that can go wrong and so many data assets, it’s no wonder organizations have huge gaps in data quality coverage, particularly in large, complex data environments.
Organizations often resort to:
- Hiring an army of data engineers to manually build low-level checks on every data element and process. Few organizations can afford to do this.
- Hoping and praying that the data is good enough, data users will fix things, and they don’t get woken up in the middle of the night. That rarely leads to career success and in some cases leads to legal action.
- Backlog innovation by keeping data inaccessible until engineers get around to validating it. Meanwhile, the competition eats your lunch.
As more data and analytics use cases are implemented, data management gets more complicated and a lot more expensive. The only viable option is for organizations to leverage automation to improve productivity and scale data governance.
Observe Everything, Engineer as Needed
Acceldata’s Torch data reliability and governance solution brings a Data Observability approach that improves the reliability, productivity and cost of data management.
Acceldata Torch is:
- Comprehensive: Torch monitors the usual data quality concerns as well as many other risks, including: reconciling data in motion, schema drift, and data trends and anomalies to provide comprehensive data reliability
- Automated: Torch leverages machine learning and an easy-to-master, user-friendly UI to make quick work of managing data reliability across large and diverse data environments
- Scalable: Torch leverages a distributed processing framework that runs both on-premise and on all major cloud providers to ensure data can move at the speed of modern business