Cloud cost visibility needs resource attribution so teams can trace increases to a specific project or environment.
2
A short weekly cost review can replace noisy alerts and give teams a way to investigate changes before they grow.
3
Storage policies, unused-volume cleanup, pricing-model choices, utilization checks, and budget controls can reduce infrastructure spending.
Summary
Jose Navaro explains how Cleo tracks and reduces data infrastructure costs. He starts with visibility, since cloud-provider dashboards often show which service became more expensive without showing which team or project caused the change. Resource tags such as environment and project provide that attribution. Cleo reviews cost changes before sprint planning, then investigates substantial changes and creates follow-up work when needed. Jose also describes ways to reduce spending: move infrequently accessed data to cheaper storage tiers, delete or relocate unused volumes, switch stable workloads from on-demand to provisioned capacity, use spot instances where interruption is acceptable, reserve predictable compute and database capacity, and move underused resources to cheaper instance types. For experiments, a budget with an automated deny policy can stop access when spending reaches an agreed limit. His approach gives product teams responsibility for their own costs while keeping the process small enough to maintain.
Jose says provider dashboards can show that total cost rose by 30 percent and identify the service responsible, but that service may be shared across the organisation. Tags add a path back to the source of the spend. Cleo uses environment and project tags, then drills into the increased service cost. In his example, environment-level costs rose uniformly, while project-level data showed that Project C caused the increase. The platform team could then speak with that project team and investigate the change quickly.
A small weekly review can work better than constant alerts
Jose finds cloud-cost alerts and anomaly detection difficult to tune. Too many notifications create alert fatigue, so people stop following them. Cleo checks cost before sprint planning and compares it with the previous week. If spending is stable, the team moves on. If there is a substantial change, it uses attribution tags to locate the affected infrastructure and may create a sprint ticket for investigation or remediation. Jose says this review takes about two or three minutes each week.
Product teams should share responsibility for infrastructure cost
Attribution lets product teams see the costs connected to their own work instead of leaving cost ownership with the platform team. Jose describes this as a shared culture of keeping an eye on spend. The process does not require a large finance program. Teams can review their assigned costs during the weekly check and investigate changes while the context is still fresh.
Storage lifecycle policies reduce the cost of retained data
AI companies often ingest large amounts of data and keep it indefinitely. Jose asks teams to check whether all of that data needs to remain in the default storage tier. Lifecycle policies can move data that is accessed infrequently into cheaper tiers. He also recommends checking for volumes that are still attached or retained despite having no recent use. Teams can snapshot and delete them, or move their contents to a cheaper location after confirming that the data is still needed.
Stable workloads can move away from on-demand pricing
On-demand capacity is useful when a project is new and its usage is unknown. As usage becomes predictable, leaving everything on demand can cost more than necessary. Jose has reduced spending by examining service usage, moving stable workloads to provisioned capacity, and setting a default capacity. Scheduled scaling can handle periods when the workload becomes larger during a day or week.
Spot and reserved capacity fit different workload patterns
Spot instances use spare cloud capacity and can cost as much as 90 percent less than regular on-demand instances, according to Jose. The tradeoff is that the provider can reclaim them with only a short warning, so workloads must tolerate losing an instance. Reserved instances suit compute or database workloads with predictable annual usage. Jose says teams can pay 40 percent less when they understand their steady requirements and reserve capacity for a year.
Budgets can stop experimental services from running away
Jose describes a budget with an automated deny-access policy for an expensive service used by an experimentation team. The team agreed on a spending limit, and the policy would stop access when the budget reached 100 percent. This approach does not fit production workloads, where an automatic denial could cause an outage. It can protect experiments from a forgotten service creating a very large bill.
"Attribution is just leaving some breadcrumbs along the way so that you can always follow it back and understand where the cost is coming from."Jose Navaro04:03
Who should watch
Platform engineers who need to explain a cloud-cost increase to a product or finance team.
Teams whose cost alerts have become noisy and are no longer being reviewed.
Engineers managing large data stores, predictable workloads, or experimental services with high spending risk.
