AWS cost and reliability, for funded startups
Your cost dashboard doesn't merge PRs.
I help funded startups turn AWS cost and reliability risk into approved engineering changes: priced backlogs, Terraform PRs, billing-verified results.
AWS already tells you what to fix. It still doesn't get fixed.
Industry surveys put self-reported cloud waste at 27 to 29 percent of spend, and the number went up in 2026, driven by AI workloads. Not because teams lack recommendations. Compute Optimizer, Cost Optimization Hub, and every FinOps tool on the market produce recommendations all day.
Recommendations without ownership, risk scoring, rollback plans, and verification do not get merged. Your engineers look at "downsize this instance," think about the incident it might cause during their on-call week, and close the tab. The waste survives because nobody priced the finding, scored the risk, and named how to prove the bill actually moved.
That gap between the dashboard and an approved PR is the entire job. It is what I sell, and it is the only thing I sell.
Cloud + AI Cost and Reliability Assessment
$5k to $7.5k fixed. Read-only access only. Delivered as a written report within ten business days of access grant. Every dollar figure computed from your own Cost Explorer and CUR data, never estimated.
Spend baseline
13-month trend and current-quarter decomposition by service, account, region, and team, from your own billing data.
Ranked savings backlog
15 to 25 items, each priced in first-year dollars, scored for effort and production risk, with rollback note and named verification method.
Commitment posture
Savings Plans and RI coverage, utilization, expiring commitments, and renewal risk, before they become bill shock.
AI workload attribution
Bedrock, SageMaker, and GPU spend by workload and team, unit-cost trends, and the common overspend patterns checked against your actuals.
Failure-surface map
Where the infrastructure breaks under growth or partial failure, each exposure tied to a business consequence, not a best-practice lecture.
Implementation menu
Fixed-price sprint and retainer options mapped to your backlog, plus a 45-minute walkthrough. Your team ships it, or I do.
What a backlog line actually looks like.
Three rows from the sample report. Every item survives being forwarded to your sharpest engineer: numbers trace to billing data, claims survive the follow-up question.
| # | Finding | First-year | Effort | Risk |
|---|---|---|---|---|
| 1 | EKS right-sizing: workloads request 3.2x their P95 CPU usage; node utilization P95 34%. Fix requests, enable consolidation, conservative 25% node reduction. verify: EKS node line in CUR, staged by namespace, non-prod first | $46,000 | L | Medium |
| 2 | Non-prod scheduling: staging and dev instance-hours run 24/7 dead flat (weekend ratio 1.00). Working-hours schedule, ~62% reduction. verify: daily-granularity spend on non-prod accounts; disable to revert | $24,000 | M | Low |
| 7 | Bedrock prompt caching on a 2.4k-token shared prefix. One API parameter; revert by removing it. verify: InputTokenCount line drops 70 to 75%, cache-read line appears | $20,000 | S | None |
From the downloadable sample report: a synthetic organization at $80k/month, built to be representative. $175k identified across 12 items. In a real engagement every figure is computed from your CUR, rounded down.
The documents do the selling. Calls are for closing.
One email
You get the sample report and a one-page scope doc. If it fits, we close in writing. No discovery-call gauntlet.
Read-only access
IAM role template provided, data-handling checklist countersigned before anything is granted. Nothing I run can change your infrastructure.
Ten business days
Deterministic analysis over your Cost Explorer and CUR exports. Models help interpret and prioritize; they never produce a number.
Report and walkthrough
Ranked backlog, failure-surface map, implementation menu, 45-minute walkthrough. Access revoked, confirmed in writing.
If you want the fixes shipped, that's fixed-price too.
Every engagement is scoped from your own backlog, and your own numbers do the selling. If your team ships the backlog themselves: genuinely, good. That was the point.
-
$15k to $35k
Cost Reduction Sprint
Terraform PRs for approved backlog items, written approval before anything touches production, rollback plan on every change. Results verified against CUR after the billing lag, in writing.
-
$10k to $25k
Reliability Hardening Sprint
The failure-surface map's top items: failover remediation, alarm coverage on revenue paths, backup verification, quota headroom. Runbooks and monitoring included, so your on-call inherits something operable.
-
$2k to $5k/mo
Cost-Watch Retainer
Continuous billing monitoring with anomaly alerts, a monthly written spend note with deltas explained, and quarterly savings PRs from newly surfaced items. The backlog never goes stale.
-
$3k to $6k/mo
Reliability Retainer
For systems hardened in a sprint: continuous monitoring with escalation handling. Sprint-first clients only.
The credit rule
The assessment fee is credited in full against any engagement over $5k within 90 days of report delivery. If you go on to fix things with me, the assessment cost you nothing.
The kind of finding this produces.
identified compute savings, ~$40k/yr
I priced the service fleet both ways. The EKS platform I had already built came out up to 80% cheaper, quantified against live production workloads.
regulated digital-asset trading platform, names withheld
- EKS platforms delivered twice: one carrying production trading traffic, one greenfield at a second company. The platform build is a repeatable capability, not a one-off.
- Observability spend cut at the same platform, primarily Datadog log indexing: one of the most common silent line-item bloats.
- Ten years in regulated Canadian fintech and trading, where deployments run against live order flow and downtime is measured in money.
- Infrastructure as code throughout: Terraform and Pulumi delivery, OpenTelemetry and Grafana observability builds.
Built for a specific kind of company.
This works when
- +You're funded or revenue-generating, roughly 30 to 300 people, with a real engineering team.
- +AWS spend is $15k/month or more. The sweet spot is $50k to $150k.
- +A technical sponsor owns the problem: CTO, VP Engineering, platform lead, or technical founder.
- +You'll grant read-only billing and resource access. That's all the access there is.
Not a fit
- xUnder ~$10k/month AWS spend. Run the free self-audit instead; it will find you real money.
- xShopping for credits, resale discounts, or a managed billing layer.
- xWanting guaranteed savings. Be suspicious of anyone who will guarantee them.
- xA fully staffed FinOps function already shipping fixes weekly. You don't need me.
One senior engineer. No handoffs.
I'm Justin Henderiks, a staff-level platform engineer. The majority of my ten years has been in the regulated Canadian fintech and trading space: compliant order systems under best-execution requirements, platform architecture on AWS and Kubernetes, infrastructure defined in Pulumi and Terraform. The person who prices your backlog is the person who writes the PRs.
I run an AI-first delivery stack, which is the margin that makes fixed pricing work at these numbers. The analysis itself stays deterministic: models interpret and prioritize, and every dollar figure in a report traces to computation over your billing data.
Toronto-based (EST), async-first by preference. The documents do the selling; calls are rare, short, and for closing.
