At a hypergrowth platform, infrastructure and tooling costs were approximately 23% of revenue and the dominant driver of COGS. Targeted fixes that put operations ahead of costly architecture rewrites brought that down to 3% of revenue.
Rapid scale with invisible unit margins.
Technology costs were outpacing top-line growth because leadership could not see what the spend supported, whether it was necessary, or how it affected gross margin before the bills arrived.
Every dollar of unnecessary infrastructure, SaaS, AI or tooling spend that is eliminated goes directly to EBITDA, without requiring additional revenue. Technology spend stayed unmanaged because its costs were spread across disconnected mechanisms, each showing only part of the picture.
Using the Technology Value Creation Method, I evaluated spend across cloud, data platforms, observability, developer tooling and SaaS.
Impact without architecture rewrites.
I took on low-risk, immediate-return opportunities first. That produced fast financial returns without stalling product engineering.
| Priority | Strategic action | Engineering impact |
|---|---|---|
| 1 | Remove idle resources and correct provisioning defaults | None |
| 2 | Correct inefficient tiers, retention, caching and compute configurations | Low |
| 3 | Renegotiate seats, commitments and contract terms | Low |
| 4 | Change architecture only where savings justified the engineering investment | Higher |
Most of the savings did not require an architecture rewrite, a hiring freeze or new product investment. They required connecting spending decisions directly to margin performance.
One measure for executive leadership, tracked monthly and quarterly.
The ratio gave leadership a single benchmark for whether technology costs were growing faster or slower than revenue.
The ALIGN™ framework applied to technology spend. Assess and Lead in Stage 1, Integrate and Grow in Stage 2, Normalize in Stage 3.
Complete portfolio baseline. I calculated baseline spend across cloud infrastructure, AI and LLM model consumption, SaaS licenses, data platforms and developer tooling.
Accountability mapping. Every material spend category was assigned a named owner.
Outputs. An audit-ready cost-to-revenue baseline, an analysis of contract and commitment exposure, and a prioritized remediation plan.
Spend was rebuilt at the product and feature level and mapped against four customer usage indicators. Adoption, depth of use, expansion attachment and perceived customer value. Together they produced a strategic decision framework.
Outcome. A capability that costs more than the revenue it supports does not always need to be cut. It may need to be priced, metered, capped or packaged differently.
Governed savings. Stage 3 stood up a recurring executive reporting package covering cost-to-revenue, run rate against budget, marginal unit cost and upcoming renewals, so the savings held.
Cultural integration. Cost discipline became part of daily technology operations rather than a one time cost cutting exercise.
The pattern in each platform area and the remedy put in place. Figures and names are withheld.
| Platform domain | Common operational finding | Strategic remedy |
|---|---|---|
| Cloud | High-cost provisioning defaults and idle environments | Immediate remediation |
| AI and LLM | Model tier or capacity mismatched to actual workload needs | Immediate reallocation |
| Data platforms | Inefficient queries, storage policies, or interactive compute on scheduled tasks | Immediate configuration and contract fix |
| Developer tooling | Excess CI/CD runner usage, idle environments and poor retention | Immediate cleanup |
| Observability | Unused log ingestion and redundant monitoring agents | Immediate optimization |
| SaaS and contracts | Purchased licenses exceeding active usage, and misaligned minimums | Contract renewal alignment |
Five things that go directly into an operating review or a board meeting.
The free Margin Recovery Scorecard ranks where your cloud, AI and SaaS spend can save the most, with first-year dollar estimates. It takes about five minutes, with no login.