INDEPENDENT AWS CLOUD ECONOMICS

AWS cost engineering, not cost reporting.

Mission FinOps traces AWS cost to the architecture and the contract that produced it: per-tenant unit economics, Bedrock and AI attribution, and commitment structure.

Nearly seven years inside AWS. Enterprise estates from $1M to over $10M a month. Independent, with no reseller or partner quota.

Book a cost-question review See a sample investigation

30 minutes, free. Bring the cost question or the contract.

Where this comes from.

Nearly seven years inside AWS as a Senior Solutions Architect and Senior Technical Account Manager, working with large, complex enterprise environments including national banks and telecommunications organizations. Before that, infrastructure and network engineering.

I am Yuvdeep Singh, and I do not work for AWS anymore. There is no reseller or partner quota behind a recommendation. More on my background →

Five ways in.

Bedrock Cost Attribution

Bedrock inference cost can be attributed to the calling IAM principal from April 2026 onward.

Bedrock cost attribution →

EDP & Private Pricing Advisory

An Enterprise Discount Program or Private Pricing Agreement commits spend over a multi-year term in exchange for a discount.

EDP and private pricing →

Per-Tenant Cost Engineering

Multi-tenant SaaS architectures share infrastructure across customers, so cost allocation tags describe the resource rather than how its capacity was consumed.

Per-tenant cost engineering →

AI Cost Management Training

FinOps and platform teams are being handed responsibility for AI spend faster than the practices exist.

AI cost management training →

Migration Cost Modeling

A migration business case is built on assumptions about volume, architecture, rate, and commitment structure.

Migration cost modeling →

How the work is evidenced.

Every engagement produces the same structure: what the evidence supports, what it contradicts, what is missing, who owns the cost, and where confidence stops. The synthetic example below works a data transfer question rather than one of the engagements above, because the structure is the point, not the subject.

SYNTHETIC WORKED EXAMPLE

Executive conclusion & confidence

NAT Gateway data processing is the primary contributor to the spend movement. Confidence: low until the source workload is confirmed.

Supporting

Cost Explorer service breakdown and usage-type anomaly detection (z-score 2.7) both isolate NAT Gateway processing.

Contradicting / incomplete

No VPC Flow Log evidence. No tag-based ownership resolved.

Ownership, missing evidence & next decision

Account identified; workload owner not yet resolved. Next: enable VPC Flow Logs, reconcile the NAT Gateway to a subnet and workload, then re-run.

See the full worked example →

kulshan report · v0.6.2 · Account: 012345678901 (synthetic) · Period: 2026-06-01 to 2026-06-30

Findings: 7

Also available.

Kulshan is an open-source Python CLI for local AWS cost analysis, published to PyPI under Apache 2.0. The Mission FinOps Agent is in Early Access with selected design partners. Both are engineering tools that support the work above.

Start with the question.

An AI cost forecast that has already moved, an EDP or renewal decision that depends on a number nobody has modeled, or a gross margin that cannot be traced to the customers producing it are all valid starting points.

calendar → book