Companies are losing money on cloud services. According to Flexera's State of the Cloud report, 84% of organizations are having difficulty controlling their cloud spending, and their budgets are already 17% over budget. On average, companies waste 32% of their cloud technology budget—that's about $232 out of every $1,000 wasted. With global spending on cloud services reaching $723.4 billion this year, that's hundreds of billions of dollars wasted.
But here's what many companies don't realize: some of the biggest losses don't come from obvious sources, such as idle servers or forgotten test environments. They come from the fact that almost all companies ignore logging. Every time an application performs an action, it writes a log. Every click, every transaction, every error is recorded. Companies store mountains of these logs "just in case," paying cloud providers to store data that no one ever looks at.
Oleksandr Shevchenko, a Site Reliability Engineer with published research on engineering cost optimization, has been investigating this problem across different organizations. His background uniquely positions him to tackle this issue—years of data center infrastructure work at IT-Soft, where he reduced downtime by 35% and operating costs by 15% through predictive maintenance systems, combined with expertise in cloud automation that he now shares through technical publications and participation in hackathons like Hackathon Raptors. What he's discovered is shocking.
The Hidden Problem
After many years of working with cloud infrastructure, Shevchenko began to notice a certain pattern. Companies were paying huge sums of money for logging without asking themselves whether they really needed all that data. Organizations have strict requirements for reliability and security, so everyone assumes that these expenses are necessary. But something doesn't add up here.
"I started studying logging pipelines and realized that companies were storing huge amounts of redundant data. The same information was logged multiple times as it passed through different systems, and no one questioned whether we needed all that data," Shevchenko explains.
Across the industry, companies treat logging as insurance—better to have too much than too little. But this approach creates another kind of risk: financial losses that no one notices because the bills come from cloud service providers rather than traditional IT budgets.
Most of the data in logs is noise: repetitive status checks, routine operations that worked perfectly, and debugging information that is only relevant in the event of a failure. Some logs are duplicated three or four times in different systems. Other logs collect information so frequently that even if someone wanted to review them, it would be impossible due to the sheer volume.
Finding the Solution
Instead of simply deleting logs and hoping nothing went wrong, Shevchenko took a systematic approach. He restructured the entire logging architecture, embedding intelligent filtering into the pipelines. Now, the system decides in real time what deserves long-term storage and what can be deleted immediately.
Critical errors? They are flagged and stored with full context. Routine operations completed successfully? They are aggregated or deleted. The result: logging costs have been reduced 20-fold. For a large financial institution that processes millions of transactions daily, this means serious money—potentially millions of dollars a year.
"The key was understanding the business value of different types of data. Not all logs are created equal. Some are absolutely necessary for regulatory compliance and troubleshooting. Others just take up space and cost money," he says.
But cost savings weren't the only benefit. With less noise in the system, it became easier to spot real problems. When everything generates alerts, nothing gets attention. By filtering out irrelevant data, Shevchenko's team was able to focus on the real issues.
The Bigger Picture
Issues related to the cost of cloud technology are usually framed as a question of choosing the right server size or negotiating more favorable rates with suppliers. However, Shevchenko's work points to something else: most companies don't really understand what they are paying for.
At IT-Soft, he helped expand a large data center by 250 racks, optimized cooling systems to reduce energy losses, and built predictive maintenance systems using IoT sensors and machine learning. This project reduced unplanned downtime by 35% and lowered operating costs by 15%.
This experience taught him to look at infrastructure holistically. Everything is interconnected. Logging affects storage costs. Storage affects backup costs. Backup affects network costs. Most companies optimize these elements separately, losing sight of the big picture.
Throughout his career, Shevchenko has implemented a number of key improvements in various organizations. He created automatic "circuit breakers" that smoothly shut down applications when nodes fail, preventing data corruption. He migrated mission-critical systems from outdated, vulnerable software versions to modern, secure ones, eliminating security risks. He created reusable infrastructure templates that reduce deployment time from several days to 30 minutes and reduce configuration errors by 90%.
Making AI Actually Useful
The consequences of unchecked logging extend beyond inflated storage bills; they directly sabotage the adoption of advanced automation. One of the hottest topics in technology right now is AIOps—the use of artificial intelligence to manage IT operations. However, AI models are only as good as the data they are fed.
Most implementations fail because companies feed these models raw, uncurated logs. When an AI attempts to learn from a system generating millions of useless lines of data, it cannot distinguish a critical signal from background static. Instead of detecting true incidents, the model trains on "noise patterns," resulting in a system that generates more alerts than it eliminates.
Many organizations attempt to bypass this complexity by relying entirely on managed services like AWS DevOps Guru, Azure Monitor's smart detection, or Google Cloud's Log Analytics anomaly detection. While these platforms offer powerful, pre-trained models capable of identifying standard patterns—such as memory leaks or latency spikes—they are not immune to the data quality problem. When fed unoptimized, noisy application logs, even these premium services struggle to distinguish between a genuine outage and a harmless warning, often leading teams to mute the very tools meant to save them.
He's led AIOps implementations with specific, measurable targets: reduce alert noise by 95%, cut the time to detect problems by 60%, and improve the time to fix problems by 40%. These projects consistently hit their targets.
"AI is not magic. It is a tool that requires careful tuning. It needs to be trained on real incident models, not theoretical thresholds. The difference between a useful AI system and an annoying one is whether the operations team trusts it," Shevchenko notes.
Now the system filters out false alarms and highlights real problems. Engineers stopped ignoring alerts because they actually meant something. This is the difference between artificial intelligence as a buzzword and artificial intelligence as a practical tool.
Beyond Simple Filtering
But filtering alerts is just the beginning. Shevchenko's more advanced implementations take AI integration further. When the system detects a problem, it doesn't just notify engineers—it checks existing documentation to verify that recommended solutions are still current. Infrastructure changes over time. Solutions that worked six months ago might not work today. AI can catch these discrepancies before an engineer follows outdated instructions at 3 AM.
The alerts engineers receive now include potential solutions from verified documentation. Instead of "Database connection pool exhausted," they get "Database connection pool exhausted—verified solution: increase pool size from 50 to 100 in config.yaml, last successful resolution 2 days ago." This cuts troubleshooting time dramatically.
Some organizations push this concept even further with automated remediation, having AI not just suggest fixes but implement them. Shevchenko remains: "Automated fixes can work for well-defined, low-risk scenarios like restarting a crashed service or clearing a full disk. But you need strict guardrails. I've seen systems where automated remediation created cascading failures because the AI didn't understand the broader context."
Balancing speed with safety is key. Automating simple, repetitive fixes frees engineers to focus on complex problems. But every automated action needs monitoring, rollback capabilities, and clear boundaries.
This layered approach connects back to the original logging problem. Smarter alerts reduce wasted time, which reduces operational costs. Verified documentation prevents expensive mistakes. Careful automation handles routine work that burns out good engineers. Each piece reinforces the others, creating systems that are both more reliable and more cost-effective.
What This Means for Other Companies
Shevchenko's work offers a blueprint for organizations drowning in cloud costs. The solution isn't just about buying less or negotiating harder. It's about questioning assumptions. Why are we storing this data? Who will use it? What happens if we don't have it? These simple questions can reveal enormous inefficiencies.
The tech industry faces particular pressure right now. Regulations are tightening, cyber threats are increasing, and there's constant pressure to cut costs without breaking things. Engineers who can deliver both reliability and efficiency, not one at the expense of the other, are becoming essential.
Shevchenko also contributes to the broader technology community by publishing his research. His work appears in InfoTechLead, The American Journal of Engineering and Technology, TechMediaToday, and AllTechMagazine, where he discusses infrastructure automation and cost optimization strategies.
But the real lesson to be learned from his work is not about any particular tool or method. It's about changing our attitude toward cloud infrastructure. The pay-as-you-go model promised to make IT costs more predictable and controllable. Instead, it has created new opportunities for waste that many organizations still don't recognize.
84% of companies struggle to manage their cloud technology costs, so the question is not whether optimization is needed, but why more organizations are not approaching it systematically. Shevchenko's 20-fold reduction in logging costs was the result of his asking basic questions that most people never ask. How many more opportunities like this are hidden in plain sight?
