Found where the cloud money went, then stood up FinOps to keep it.

Situation

S1 was negotiating a large multi-year AWS contract and had no FinOps function, no cost org and no data lake. I saw large room for savings in cloud spend, especially in Presto. S1 was a Google shop, so any analysis had to live in Google tools.

Objective

Model cost and spend well enough to guide the AWS negotiation, find and deliver the savings, and give the company a FinOps function that would outlast my own analysis.

Approach

• Cost intelligence in Sheets. I built the whole stack in Google Sheets and Apps Script, fed from raw AWS billing plus exports from DataSet/Scalyr, BigQuery and Pileus. I modeled cost against feature and capability, mapped regional cost normalized by agents deployed over time, and modeled expansion into Canada, India and Australia: minimal footprint by feature and break-even math.
• Three cost curves. For the AWS negotiation I put forward likely growth on current spend, finance’s linear-growth model, and my own model for an optimized environment. Asked about cost-cutting, I named the top lever: accelerate the Presto-to-Scalyr migration, a 9 to 12-month effort.
• What the data showed.
  1. Event volume drives cost, not agent count.
  2. Onboarding defaulted every event category on. A new customer’s data, and cost, stayed high unless someone tuned it.
  3. Kubernetes Requests, Limits, HPA and VPA settings were not tuned.
  4. Each region used one default node size and count, with the same blunt sizing for Postgres and other resources.
  5. A new region needed a large deployed-agent base to break even.
• A major healthcare customer had concerns about query performance and data export ahead of a QBR. In their Presto query activity and agent event volume I found a recurring export pattern driving high query cost against modest ingest. I presented it remotely to the customer’s security team, about 30 people including their CISO. They adjusted their queries, their query costs dropped sharply, and the relationship improved. Internally, the same analysis showed how fast Presto query cost grew on AWS.
• FinOps, stood up and handed off. I interviewed Finance’s 2 to 3-person analytics group and, with my sign-off, they became S1’s official FinOps function. I trained them and handed off reservation right-sizing, spend-trend reporting and the rest, then fed them a prioritized list of cuts that got delivered: idle S3 and EBS cleanup, CloudWatch and CloudTrail pruning, RDS migration to IOPS-optimized instances, Graviton adoption and more.
• Pricing on the real driver. A task force was spun up to optimize event volume and filter excess flows. Over about three months I presented the evidence repeatedly, pushing to make data ingest and storage its own SKU, with customers shown how to tune event flows to stay within their data budget.
• Kubernetes cost. I drove the rollout of KubeCost, then brought in and rolled out ScaleOps, plus forced resource tagging, integrated usage monitoring, right-sizing for Postgres and narrower node scope.
• Allocation and credits. I built a namespace-by-cluster cost-allocation model mapping each Kubernetes namespace to the correct OpEx bucket, SaaS cost of goods or DevEx/R&D, and drove its adoption by Accounting and Finance. I set up a repeatable practice to redeem AWS POC credits on time.
• The room. The CTO and my manager brought me into VP-and-above meetings to present the charts showing where engineering spend could be cut. My job in those rooms was the contrarian case, backed by evidence; leadership then carried the change through.
• Purchasing. I set up procurement and security leadership to sign off on purchases through the AWS Marketplace.

Net results

• S1 posted its first quarterly non-GAAP operating profit in Q2 FY26, the quarter ended July 2025. The cost discipline I built and handed to FinOps was part of what got it there.
• The cost-allocation model materially improved the accuracy of cloud-cost allocation and stood up to external audit.
• AWS POC-credit redemption recovered a substantial amount over the period.
• FinOps ran on its own, funded for better tooling, which freed me for the SLA, SRE and QA work below.

Takeaway: price on the driver of cost. Billing on agents while event volume drove cost misaligned price, profitability and customer incentives at once.

SRE: incident response and uptime intelligence, 2021 to 2023

S1 needed a structured incident-response process and a way to answer SLA-breach claims with data. My team and I built the process. I was repeatedly pulled in to support incident response during outages, and I built real-time dashboards and reports on Scalyr, the logging platform:

• Service health dashboards showed average response time, error rates and more, grouped by feature and the services under it, surfacing the slow and error-prone services.
• Customer and console health tracked error rates by console, correlated to accounts via Salesforce and to uptime via Pingdom. A single console might host dozens to hundreds of smaller clients.
• I mapped which team owned which API-endpoint group in code, by inspecting URL paths against the source repos and Kubernetes manifests. That turned “something’s slow” into “this team’s service is hurting these customers.”

Net result: S1 could tie uptime and outages to impacted customers and accept or reject SLA claims on evidence, with internal Scalyr dashboards and an external third-party status page, including dedicated reporting for specific customers. The work formalized into a dedicated SRE function for tactical remediation, root-cause analysis and corrective-action tracking, which cut the impact, duration and recurrence of customer-impacting incidents.

QA: severity and SLA tracking, then a new QA org, 2022 to 2023

I was asked to dig into bug and vulnerability severity rating and tracking against remediation SLAs. I built Jira dashboards giving VP-downward reporting of security issues, vulnerabilities and bugs, including SLA violations, which drove better prioritization and tighter issue tracking across engineering. I then wrote a proposal, which S1 implemented: QA needed dedicated Sr. Director leadership and resources to set testing standards across the whole development cycle. The org was staffed with automation engineers for compliance, UI/UX testing and more, and partnered with SRE, engineering and security to accelerate remediations and improve pre- and post-production testing.

For both SRE and QA I helped hire, built the value proposition, and handed off the reporting as each team came online. Throughout, I partnered with the Enterprise Data & Analytics org and was senior leadership’s go-to for ad-hoc data analysis.

Scope and technologies

Google Sheets and Apps Script, the entire cost-analytics stack; AWS cost and usage data; Presto query-cost analysis; Scalyr as the migration target; Kubernetes Requests, Limits, HPA, VPA and node right-sizing; KubeCost; ScaleOps; resource tagging; Postgres right-sizing; reservation optimization; AWS Marketplace; Scalyr/DataSet dashboards for service and console health; BigQuery and Pileus for cost data; Salesforce and Pingdom for account and uptime correlation; Jira vulnerability and bug SLA dashboards; third-party status-page tooling; incident response with RCA and CAT; endpoint-to-team mapping via path, repo and Kubernetes-manifest inspection; a namespace-to-OpEx cost-allocation model; AWS POC-credit redemption.