Data observability has been a hot topic for the past few years in the data space. Several recent blogs have explained why this capability is critical to the success of modern data and analytics solutions, which increasingly comprise a complicated collection of disparate tools in a hybrid multi-cloud environment. With the profusion of data sources, various modes of data transformations, and a diverse set of consumers, it’s a miracle that tasks still get accomplished. Until they don’t. Thus, the need for full-stack observability.
Despite a wealth of available information, many organizations are still unclear about appropriate use cases to introduce data observability into their technology stack. That is the goal of this document — to examine key use cases of data observability. This document is not a primer on the topic. If you are keen to learn more, please refer to What is Data Observability?, Data Observability Accelerates Modern Data Stack Adoption, and What Have I Observed about Data Observability?
While the modern data stack abounds with a plethora of specialized technologies, businesses don’t care about the technology soup. They want to derive maximum value from their data assets. For which they expect data to be trustworthy, accessible, and explainable. Hence, one of the most effective ways to convey the importance of data observability is through an examination of business-focused use cases. As an example, a technical use case of data observability may be to enable data governance, but the business user is interested in decreased latency between the creation of data and its consumption.
Data observability extracts key metrics from complex pipelines that help improve business operations as follows:
- Fast: reducing the friction between data producers and consumers, leading to faster time to conduct analytics
- Trustworthy: reducing data’s downtime along with ensuring its high-fidelity quality
- Cheap: exposing and eliminating wastage in data pipelines through cost optimization recommendations
- Efficient: automating tasks in order to optimize resources
- Modern: enabling businesses to intelligently migrate to innovative architectures that better serve customers by analyzing the technical and behavioral state
All the benefits mentioned above are intertwined. For example, automation can help deliver data faster and with higher quality leading to lower costs. In order to achieve these goals, the data observability implementation should reinforce organizations’ overall DataOps and data governance initiatives.
Faster time to analytics
The primary goal of the sales teams in any organization is to generate revenue. To do so, they rely on the marketing team to analyze vast amounts of incoming signals and qualify them as leads and prospects. Faster analysis of this data leads to quicker decision-making, which translates into revenue. To achieve this use case, the IT team needs to deploy an efficient and reliable architecture that can process fast-changing data in the shortest time possible. Amazon CTO Werner Vogels’ statement “everything fails, all the time” is especially true when organizations are running numerous pipelines on high-velocity data on ever-increasing volumes of data.
So, how does data observability play a part?
There is a high opportunity cost in delaying decisions. This may be because of the unavailability of data or the systems that process data. Data observability can deliver reliability in the build and runtime stages of the data lifecycle as follows:
- Build. During the development phase, data observability is used to improve the ability to experiment with data and test various data products before deploying them in production. The goal is to speed up development and testing. Netflix popularized the concept of chaos engineering, i.e., deliberately introducing a fault in an otherwise stable pipeline to proactively determine unknown issues and fix them in the development phase, rather than in production. A data observability product provides visibility into the state of data in testing multiple pipelines.
- Runtime. Once new pipelines are developed and tested, they are pushed into production. At this stage, data observability is used to analyze operations in an automated and continuous manner so that anomalies can be detected as early as possible. The goal of data observability in this phase is to reduce the time to detect failures and deploy fixes with zero or low downtime.
Errors can emanate because of multiple reasons — poor data quality, resource failures, or human errors. The data observability tool’s single-pane-of-glass view provides visibility into quality, performance, and resource utilization aspects of data flows across pipelines so that appropriate actions can be taken in case of an anomaly. In absence of this visibility, errors may go undetected, leading to incorrect analysis. Or when errors take place, they may require time-consuming manual triage and remediation steps.
Your data observability tool should improve the speed of data delivery to consumers. This is especially true when additional data sources are added to the mix. How quickly an organization can provision the newly available data determines its competitive advantage.
Always-available quality data
Businesses are moving faster than ever in history. Oftentimes, they are moving faster than their IT teams. Unfortunately, this sometimes causes the CFO to be the first one to detect a broken dashboard. In such a situation, remediating the fix becomes a reactive fire drill that has a high opportunity cost and reduces the team’s productivity. Broken pipelines compromise trust in the abilities of the IT teams.
Stop using your business users as your QA team. Data observability provides a unique opportunity for business and IT teams to align early in the lifecycle and remediate issues before promoting pipeline into production. Data observability can detect or even predict errors faster than traditional approaches, and business SMEs can provide the needed business domain knowledge to fix them. The goal is to help deliver reliable architecture, where data is always available and trustworthy. Accessible and quality data works on multiple dimensions:
- Timely data. A common scenario is when one of, say, a thousand pipelines did not run. Traditionally, this situation may go undetected. However, the data observability platform, using built-in statistical and AI capabilities, should build a time-series graph of the data’s patterns, including seasonality. This capability helps identify when there is an unexpected skew in the data pipeline. The platform monitors the data flow and notifies if it predicts volume-drift challenges.
- Quality data. Always-available data is one side of the story. The other side is high quality, which is often measured in several dimensions, like uniqueness, null values, distribution of data, etc. A data observability product profiles the data and performs statistical analysis or uses ML algorithms to detect anomalies.
Focus on data reliability improves trust and fosters a strong data culture. As your data observability matures, it leads to a more sophisticated design of a fault-tolerant data and analytics environment.
IT costs have increased precipitously as complexity has increased. They have also increased as more users take advantage of the ease of cloud computing, only to see consumption-based operational expenses rise unpredictably. As a result, many organizations’ focus has shifted from performance optimization to cost optimization.
Cloud computing has taken the effort away from building the data analytics infrastructure. This is good news because this effort adds little strategic value to the bottom line. Cloud service providers (CSPs) provide excellent building blocks but lack adequate guardrails to protect against cost overruns. This has led to the rise of FinOps, the financial management discipline around getting maximum business value from modern data stacks.
Data observability products play a key role as they are the map of how resources are being consumed and the cost they incur in the process. A well-charted data and analytics journey saves build costs in the short term and runtime costs in the long term. Data observability helps derive cost savings by exposing necessary metrics to help across:
- Error handling. Data observability’s monitoring and root cause analysis abilities help pinpoint errors quickly in case of unexpected outcomes. The faster an organization can detect errors, the lower the cost to fix and the resulting downtime. This is the idea behind “shift left.”
- Good pipes. Your data may meet quality gates, but it’s critical that the underlying infrastructure hums. To avoid congestion, often organizations over-provision their resources. This allows IT teams to meet service level agreements (SLA), but it also leads to unwanted and wasted costs.
Various data observability deployments have proven that the cost-savings aspects themselves are reason enough to pay for the tool.
Automation and efficiency
Digital transformation is accelerating. And pipelines are becoming ever more complex. In recent years, organizations have witnessed significant increases in the number of data sources and data consumers as they deploy more applications and use cases. Recent studies have shown that the average small business uses 102 different SaaS apps, and the average enterprise uses 288 different apps across the organization.
One IDC study shows that the number of applications that will be built in the next two years will be more than all the applications built in the last forty years. They estimate the number of new logical applications to be 750 million.
DataOps has risen as a set of best practices and technologies to streamline and efficiently handle this onslaught of new upcoming workloads. Its key pillars include testing, orchestration, and automation. Data observability is closely linked to successful DataOps. For example, the data observability tool’s lineage of pipeline tasks not only helps visualize various data flows, it should incorporate automated monitoring and altering to proactively detect anomalies and reduce the mean time to repair (MTTR).
One of the biggest problems of any monitoring system is “alert fatigue,”’ when the operational staff is inundated with so many alerts that they stop paying attention. The data observability tool should have the ability to separate noise from signals.
In summary, if data observability is done right, error rates in production should tend towards zero.
Sometimes an organization may want to go slower to go faster. Just focussing on speeding up the current state may not be the right solution for the long-run. However, the business goals and expectations do not change. The business has always craved for their data to be available faster, cheaper, and better irrespective of the underlying technical infrastructure.
Recent technology innovations have made some of these desires possible. However, they may not always involve optimizing existing platforms to extract ever more efficiencies, but may require a complete replatforming. Some of these approaches vary from microservices architecture, cloud migrations, and hardware advancements, to the ease of leveraging machine learning algorithms. Data observability can inject a degree of efficiency into these modernization approaches by providing the essential visibility of the current state and future state. This intelligence will improve the return on investment (ROI) of these modernization initiatives.
Take cloud migration as a case point. Traditionally, early migration attempts were knee-jerk attempts where decades of on-premises data assets were migrated into the cloud without intelligence about their usage and quality. This approach leads to higher cost to both migrate and maintain the target platform. Often, it required yet another migration project to optimize the cloud deployment.
Data observability provides an innovative way to determine the state of critical data elements and migrate the data in an optimized manner. Some ways by which data observability helps include:
- Identify deprecated data. Duplicates any temporary tables that should not be migrated. Use the tool to build a lineage between data elements and consumers.
- Resource consumption. Data observability can determine the right CPU, memory, network, and IO resource usage on the source and recommend a right-sized architecture in the cloud. This benefits not only in optimizing cost but also in ensuring that the target architecture will handle normal and peak workloads without downtime because of congestion, outages, and other anomalies.
- Establish SLAs and SLOs. Service level agreements and objectives are common in the DevOps infrastructure space, but time has arrived to apply the same rigor to data pipelines.
Data observability is the catalyst that accelerates the cloud migration journey and makes it cost effective.
Data observability has risen as a key component of the modern data stack as it promises to fulfill the promise of building and operating data products. Consistent access to high-quality, trusted data is the cornerstone for any successful analytics program. It is also what differentiates a successful business.
Data observability is proving to be an indispensable component of this journey, especially as pipeline complexity is constantly increasing in heterogeneous hybrid multi-cloud environments. This research has looked at some of the key business use cases whose major benefits are in improving reliability and in reducing costs.
There are more use cases that are handled by observability tools in adjacent areas, such as machine learning, security, and privacy. All these adjoining areas work on the same metadata substrate and in the future may consolidate into a single product.