DevOps and Site Reliability Engineering (SRE) are known to be fast-paced, high-stress jobs. It's no wonder, given these professionals are responsible for preventing and remediating unplanned service interruptions — and each second of downtime can cost an organization thousands of dollars in revenue. According to one previous industry survey, a large majority of SREs reported significant post-incident stress, including changes in mood, concentration and ability to sleep. The same survey also found that having a "supportive team" can reduce a lot of the stress that DevOps and SRE professionals regularly deal with.

That's why we were concerned by the prevalence of another trend revealed in our recent survey: internal disputes over what data to keep and what to discard for observability purposes. DevOps and SRE teams need access to their log data to resolve incidents in a timely manner. However, our survey reveals that a whopping 83% of DevOps and SRE professionals report internal company disputes over these matters.

This unfortunate dilemma is due to a growing avalanche of data that risks rendering some observability initiatives cost-prohibitive. Unfortunately, observability costs scale linearly with data volumes, which have increased an average of five-fold over the past three years. 93% of respondents in our survey noted they experience overages or unexpected spikes in observability costs at least a few times per quarter, if not more. Perhaps most noteworthy, only one percent of respondents said their observability costs are not rising.

How are organizations dealing with this conundrum? Hint: they're not increasing their budgets.

As observability and monitoring costs come under increasing scrutiny from company leadership, the vast majority of businesses (98%) attempt to remedy this issue by limiting the data ingested by the observability platform. In one-third of all cases, the decision of what data to keep and what data to discard is completely random. Unfortunately, the consequences of this "data down the drain" approach can be severe, including increased risk or compliance challenges; losing out on valuable insights and analytics, and failure to detect a production issue or outage. It's no wonder such decisions often lead to anxiety, discontent, and bad blood.

Organizations should no longer be forced to make the unacceptable compromise between ingesting and paying for data that ultimately goes ignored, and discarding data sets, leading to disputes and running the risk of unanticipated blind spots. Given that data growth is not going to slow any time soon, a fundamental paradigm shift is badly needed, one that reduces both the cost and noise of observability monitoring.

The key lies in leveraging AI and machine learning to analyze data at its source, as it's being generated, and identifying and ingesting only the most useful data sets. By distilling only those data sets that organizations access most frequently or might want an alert on, organizations can drastically reduce the number of metrics ingested. This can be the key to helping teams realize more value and efficiency from observability, without creating unnecessary stress and arguments.

For DevOps and SRE professionals, dealing with incidents is stressful enough. We don't need to make it worse by introducing avoidable discord. We also don't need to deprive our colleagues of the data they need to do their jobs, nor do we need to hoard all data needlessly and pay cloud service providers excessively for a lot of data that is ultimately never used. Leveraging advances in AI and machine learning can be the key to realizing significant ROI from observability initiatives and keeping costs in control, while also maintaining team harmony and peace of mind for DevOps and SRE professionals.