How Accrete AI Went from Flying Blind on Cloud Costs to Product Margin Visibility and Significant Savings

Accrete had strong visibility into overall AWS spend and cost allocation at higher levels. However, like many organizations operating shared Kubernetes environments, they encountered a common limitation: cost attribution broke down inside shared compute.

Specifically:

• Costs could not be reliably attributed to individual workloads or services
• Customer-level cost and margin were difficult to model
• Investigating cost anomalies required manual correlation across multiple systems
• AI usage could not be easily mapped to specific services without added operational overhead

While tagging and billing tools provided useful top-level insights, they did not reflect how resources were actually consumed at runtime within shared clusters.

Accrete deployed Attribute to observe real runtime behavior using eBPF, enabling cost attribution based on actual system activity rather than tags or estimates.

As Jason Moore, Principal DevOps Engineer at Accrete, described it: “It’s like a network monitor for your budget.”

Deployment required no changes to existing workloads, tagging strategies, or pipelines. Within days, the platform team gained:

• Cost visibility at the workload and service level
• Insight into traffic patterns driving infrastructure spend
• The ability to correlate AI usage with specific services
• A unified view of infrastructure and AI-related costs

This allowed cost data to align directly with how the platform actually operated.

$46K per Month in Inefficiencies – Eliminated

Attribute surfaced a previously invisible network traffic pattern driving significant infrastructure cost.

A high-throughput service was routing large volumes of data through an inefficient network path. Once identified, the team implemented a targeted fix to optimize routing.

Result:

• Monthly cost reduced from $63K to $17K
• ~$46K/month in unnecessary spend eliminated

Without runtime-level visibility, identifying this issue would have required significant manual investigation and may have persisted unnoticed.

Improved AI Governance Visibility

Runtime attribution also revealed inconsistencies in API key usage across environments.

This enabled the team to:

• identify policy deviations
• correct usage patterns
• reduce operational and compliance risk

Shared Source of Truth for Cost of Goods Sold

For the first time, Accrete established a consistent view of cost across engineering and finance.

They can now:

• understand cost per workload and service
• align engineering decisions with financial impact
• move beyond estimated allocation toward actual consumption

Foundation for Per-Customer Unit Economics

With runtime attribution in place, Accrete is extending its model to incorporate customer-level identifiers.

This will enable:

• Accurate cost per customer
• Margin analysis at the account level
• Data-driven packaging and pricing decisions grounded in actual consumption, not estimates

Accrete continues to expand this use case, moving from infrastructure cost visibility into product margin analysis and pricing strategy. With true cost-to-serve visible at the account level, their pricing and product teams now have the data to set rates that reflect how the platform is actually used.

By shifting from tag-based estimation to runtime-based attribution, Accrete gained visibility into how resources are truly consumed within shared infrastructure.

This enabled immediate cost optimization and established a foundation for aligning platform operations with business economics at scale. As Accrete expands into product margin analysis and pricing, runtime attribution gives their teams a single source of truth: not just what the platform costs, but what it costs to serve each customer, and what that should mean for how they price.