FinOps in practice: 10 ways to lower cloud costs

The cloud was supposed to be cheaper. In many organizations, however, it has turned out to be a source of monthly stress – the bill grows faster than the business, nobody fully knows what exactly is being paid for, and an attempt at cost reduction ends in a production outage. This phenomenon even has a name: cloud shock – the moment when infrastructure expenses spiral out of control. The answer is not abandoning the cloud, but FinOps – a culture and set of practices that build a bridge between engineering, finance, and business. FinOps is not a one-time “cleanup” project, but a continuous process in which decisions about resources are made consciously, with full visibility of their cost. Well-implemented cloud cost optimization does not mean lowering quality – it means eliminating waste.

Below we present 10 specific ways for cloud cost reduction, ordered from “quick wins” to mature management mechanisms. No generalities – with techniques, thresholds, and metrics you can implement right away.

Quick wins – rightsizing, shutdowns, budgets

Before reaching for advanced mechanisms, gather the “low-hanging fruit.” These are actions with the best ratio of effect to effort.

Method 1: Rightsizing.

The most common source of overpaying is overprovisioning – virtual machines and databases selected “just in case,” utilized at only a dozen or so percent.

• Analyze real CPU, RAM, and I/O usage in a window of at least 2–4 weeks.
• Match the instance size (SKU) to the actual load – rightsizing can achieve a cloud cost reduction for a resource by dozens of percent without affecting performance.
• Treat this as a continuous process, not a one-time action.

Method 2: Automatic shutdown of non-production environments.

Dev/test/staging environments rarely need to run 24/7.

• Set shutdown schedules outside of working hours (e.g., evenings, weekends).
• An environment running 8 hours a day for 5 days accounts for approx. 24% of a full week’s time – the rest is pure savings.
• Automate this (scheduler/tags) instead of relying on manual shutdowns.

Method 3: Budgets and cost alerts.

Do not start optimization with cuts – start with visibility.

• Set budgets at the subscription, project, and team levels.
• Configure threshold alerts (e.g., 50%, 80%, 100% of the budget) before the month ends.
• Budgets without alerts are just numbers in a report – only notifications provide real control.

Reservations and Savings Plans – break-even thresholds

The on-demand model (pay-per-hour usage) is flexible, but the most expensive. For stable and predictable workloads, committing to longer usage yields significant discounts.

Method 4: Reservations (Reserved Instances).

A commitment for 1 or 3 years in exchange for a significant discount compared to on-demand – in practice, even up to dozens of percent. They work best for a stable, known baseline workload.

Method 5: Savings Plans.

A more flexible form of commitment – you declare a fixed hourly spending amount (e.g., in USD/h), and the discount applies to eligible compute services, offering greater flexibility than classic reservations. Savings Plans are a good choice when the workload is stable in terms of volume, but variable regarding specific machines.

How to determine the break-even threshold? The key is the resource utilization rate on a monthly scale:

Practical rule: if a resource runs stably for a significant part of the month, e.g., above 60–70% of the time, it is worth calculating the cost-effectiveness of reservations or Savings Plans based on historical data.

Below this threshold, on-demand flexibility is usually worth its price. Do not make reservations “blindly”- base your reservation decisions on hard usage data.

Storage and transfer – tiering, archive, limiting egress

Data storage and transfer costs are an area that grows quietly and is notoriously overlooked.

Method 6: Storage tiering.

Not all data requires the fastest (and most expensive) tier.

• Move data between tiers based on access frequency: hot (frequent), cool (infrequent), archive (rare, long-term).
• Automate storage tiering using lifecycle policies that automatically move older objects to cheaper tiers.
• The archive tier can be many times cheaper than hot – at the cost of access time, which is acceptable for backups and historical data.

Method 7: Limiting egress

Data transfer into the cloud is usually free, but egress – outgoing data traffic – can be an expensive surprise.

• Keep processing close to the data – avoid sending large volumes between regions or clouds.
• Use caching and CDNs for externally served data to limit repetitive egress.
• Design architecture consciously – transfer cost is often a hidden bill component that only reveals itself at scale.

Cost governance – tags, chargeback/showback, alerts

Savings techniques without organizational order quickly drift apart. Governance is the layer that sustains results and prevents waste from returning.

Method 8: Consistent tagging strategy.

Without consistent tags, you do not know who generates the cost.

• Define a mandatory set of tags: owner, team, project, environment, cost center.
• Enforce tagging via policies (e.g., blocking the creation of untagged resources).
• Consistent tags are the foundation of every cost report – without them, chargeback and showback are impossible.

Method 9: Chargeback vs showback.

Two models for allocating costs to departments:

• Showback – you show owners how much their resources cost, but without formally charging their budget. Builds awareness and shared responsibility.
• Chargeback – you actually charge departmental budgets with their cloud costs. A stronger disciplining mechanism that requires organizational maturity.

It is usually worth starting with showback, and implementing chargeback once tagging and reporting are reliable.

Method 10: Proactive anomaly alerts.

Beyond budgets, it is worth detecting unusual spikes.

• Configure cost anomaly detection (sudden spend increase vs historical trend).
• Route alerts to resource owners, not just the central team.
• An early anomaly signal is the difference between a correction of a few hundred vs tens of thousands of dollars.

FinOps KPIs – unit cost, coverage, forecast accuracy

Mature FinOps is measured by metrics, not impressions. Three of them are key.

• Unit cost – cloud cost per unit of business value (e.g., per customer, transaction, order). This is the most important indicator because it shows efficiency regardless of scale. Growing spend with a falling unit cost means healthy growth, not a problem.
• Coverage (reservation coverage) – what percentage of eligible workload is covered by reservations or Savings Plans. Too low means overpaying in on-demand; too high – the risk of paying for unused commitments. The goal is a conscious balance.
• Forecast accuracy – how accurately you predict spending. High accuracy is the foundation of trust between IT and finance and a prerequisite for budget planning.

These KPIs turn cloud cost optimization from a one-off action into a measurable, repeatable process.

Most common cloud cost optimization mistakes

1. Silos between IT, finance, and business. When engineers do not see costs and finance does not understand architecture, optimization stalls. FinOps is primarily a cultural shift.
2. “Just in case” resources. Purchasing excess capacity without analyzing workloads is the simplest path to overpaying – the exact opposite of rightsizing.
3. Lack of automation. Manual cost management is time-consuming, error-prone, and unscalable. Without automation, optimization results quickly regress.
4. Lack of visibility. Without consistent tags and reporting, it is impossible to identify sources of inefficiency or make accurate decisions.
5. One-off approach. “Once a year” optimization does not work – the cloud changes daily, so cost control must be continuous.

FinOps is a process, not a one-off optimization

The cloud gives organizations immense flexibility, but without proper processes, it is easy to lose control over costs. That is why effective FinOps does not boil down to individual optimization actions, but relies on continuous monitoring of resource utilization, spending analysis, and collaboration between technical, business, and financial teams.

Organizations that implement the described practices are able not only to limit costs, but also to better forecast expenses and utilize cloud potential more effectively.

Don’t know where to start? Check your cloud’s savings potential

Many organizations know they incur unnecessary costs related to cloud usage, but lack full knowledge of which services account for the largest expenses and where the greatest optimization potential lies. Therefore, the first step should be a cloud cost audit.

During an audit, we analyze real resource utilization, identify unused or overprovisioned services, evaluate opportunities to use Azure reservations and Savings Plans, and point out specific actions that can lead to cloud cost reduction.

This approach brings tangible results. For a large financial institution, we conducted an analysis of their Microsoft Azure environment covering virtual machines, disks, and database services. By implementing the recommended changes, the client reduced costs by approximately EUR 8,000 per month, translating to nearly EUR 96,000 in annual savings, without impacting performance or service availability.

If you want to check “how to reduce Azure bill” or see what optimization potential is hidden in your environment, it is worth starting with a cloud cost audit.