Cost Attribution, at Three Scales

One discipline — making compute fleets legible and their costs attributable — proven from $10–12M/quarter cloud spend down to invoices I personally pay.

Fleet-Scale Financial Attribution — Apple

Own the CIBO/FinTrack pipeline correcting cost misattribution across roughly 700,000 servers, where quiet errors distort budgets by millions. Plus cloudOS, an internal platform giving Finance, SRE, and execs visibility into livability, deployment state, and financial attribution. Now piloting LLM inference to pre-label project cost hierarchies.

~700K servers
~400 weekly users
Cost attributionPrivate cloudLLMFull-stack platform

Cost-Attribution Pipeline — Roku

Built the pipeline handling $10–12M/quarter of AWS spend and co-led the ECS→Kubernetes migration. The engagement where the theme crystallized: cost data is an engineering artifact, not an accounting afterthought.

$10–12M / qtr
ECS→K8s
AWSKubernetesCost attributionShowback

Trendvesting — Unit Economics in Production

AI signal-intelligence platform live since Feb 2024: multi-mode Go backend, Next.js app, React Native client, Python FastAPI consensus service. A real AI system whose invoice I pay — ~$1.50–2.00 fully loaded per signal, ~$400–500/month steady state. The case study I live in.

$1.50–2.00 / signal
Since Feb 2024
GoNext.jsFastAPIReact Native

TendForm — Infrastructure Discipline, Solo

Solo SaaS form builder with multi-form packets, a markdown DSL, a production MCP integration, and a HIPAA-compliant tier on a CloudNativePG enclave with a full BAA chain. Real customers, real deployments — run like the fleet work, just smaller.

HIPAA tier
Production MCP
React/ViteFastAPIKubernetesRedisR2/KV

Want this outcome?

Book an AI Feasibility Sprint to get your 90-day implementation plan.

By the Numbers

$10–12M
Quarterly AWS spend attributed at Roku
~700K
Servers under fleet-scale attribution at Apple
$1.50–2.00
Fully-loaded cost per AI signal on Trendvesting
~400
Weekly users on the cloudOS internal platform

How I Work

1

Attribute

Map the spend to owners, systems, and units. Nobody owns a cost they can't see.

2

Instrument

Turn cost into an engineering artifact — measured in the system, not reconciled after the fact.

3

Reconcile

Validate the architecture against the invoice. The diagram and the truth have to match.

4

Make it legible

Ship a surface Finance, SRE, and execs can all read — cost data people actually use.

Want a number you can defend to finance?

Get my AI cost-attribution checklist to find where your spend is really going.

No spam. Unsubscribe anytime. Used to send your kit and occasional AI insights.