Platform Economics & Cloud Cost

A financial operations practice that brings together finance, engineering, and product teams to optimise cloud spending — treating cost as a metric to be understood and managed continuously, not just a finance concern. The FinOps Foundation defines a crawl/walk/run maturity model for adoption.

"We introduced FinOps after our cloud bill grew 3× in 18 months with no corresponding revenue increase. The FinOps practitioner embedded in the engineering team identified that 35% of spend was idle or over-provisioned resources — recoverable savings without any performance impact."

A cloud pricing model (AWS, Azure, GCP) where you commit to using a specific resource type for 1 or 3 years in exchange for a discount of 30–72% compared to on-demand pricing. Best suited for stable, predictable workloads. Unused reserved capacity is still charged.

"We converted our database tier to 1-year Reserved Instances after 6 months of stable utilisation data — the m5.2xlarge instances run 24/7 with 85% CPU utilisation. The RI commitment saves 42% compared to on-demand, paying back in 11 months. We avoided 3-year commitments given the pace of infrastructure change."

Cloud compute capacity available at up to 90% discount compared to on-demand prices, in exchange for accepting that the instance can be reclaimed (interrupted) by the cloud provider with 2 minutes notice when capacity is needed elsewhere. Suitable for fault-tolerant, stateless, or checkpoint-aware workloads.

"Our ML training pipeline runs entirely on Spot Instances — we save 75% on compute costs. Jobs checkpoint every 10 minutes so a Spot interruption means we lose at most 10 minutes of training progress. We use a Spot interruption handler to drain work gracefully and requeue on new capacity."

The process of matching cloud resource type and size to actual workload requirements — changing over-provisioned large instances to smaller ones that still meet performance targets. One of the most impactful cloud cost optimisation actions; typical savings are 20–40% of compute spend.

"Rightsizing analysis showed 60% of our EC2 instances were below 20% average CPU utilisation — they were originally provisioned for peak load that never materialised. Moving m5.4xlarges to m5.xlarge reduced compute costs by 38% with no performance degradation, verified by 2 weeks of load testing."

A metadata label applied to cloud resources to attribute costs to specific teams, products, environments, or business units. The foundation of cloud cost visibility — without consistent tagging, all spend appears as undifferentiated infrastructure cost with no ownership.

"We enforce cost allocation tags at provisioning time via Terraform policy: every resource must have team, product, environment, and cost-centre tags. Resources without mandatory tags are flagged daily and auto-decommissioned after 14 days if not remediated. Tag coverage is now at 98.4% — up from 61% a year ago."

A FinOps practice where teams are shown their cloud cost allocations for visibility and accountability, without those costs actually affecting their budget or P&L. A precursor to chargeback — used to build cost awareness before holding teams financially accountable.

"We ran showback for two quarters before moving to chargeback — teams could see their allocated cloud costs in real time without it affecting their budget. The visibility alone drove a 20% cost reduction as engineers self-optimised once they could see the numbers. That cultural shift made the chargeback transition smoother."

A FinOps model where actual cloud costs are charged back to the consuming business unit's P&L or budget, creating direct financial accountability. More powerful than showback for driving cost-conscious engineering, but requires accurate cost allocation and a culturally ready organisation.

"After 6 months of showback, we moved to chargeback: cloud costs now appear on each product team's monthly P&L. In the first quarter, three teams reduced their footprint significantly — not because they were asked to, but because the costs were now visible against their budget targets."

The cost and revenue associated with a single unit of product or service — cost per customer, cost per API request, cost per GB processed, cost per transaction. Unit economics connect infrastructure decisions to business performance and are the most powerful signal of whether costs are scaling efficiently.

"Our unit economics dashboard shows cost per active customer: it was £0.84 in Q1 and is now £0.71 after the database query optimisation work. Despite 40% more customers, our total infrastructure cost only grew 18% — the unit economics prove the architecture is scaling efficiently."

Cloud resources that are running but not delivering value — idle instances, over-provisioned services, orphaned snapshots, unused load balancers, forgotten test environments. Industry estimates put cloud waste at 32–35% of total cloud spend for most organisations.

"Our cloud waste audit identified four categories: idle EC2 instances with <5% CPU (12% of spend), unused EBS volumes from terminated instances (8%), over-provisioned RDS databases (11%), and forgotten dev environments never decommissioned (9%). Total recoverable waste was 40% of the monthly bill."

A flexible AWS pricing model that offers discounts (similar to Reserved Instances) in exchange for committing to a minimum spend per hour over 1 or 3 years — applicable across instance types, sizes, and families. More flexible than Standard RIs because the commitment is monetary, not tied to a specific instance type.

"We chose Compute Savings Plans over EC2 Reserved Instances because our instance mix changes as we rightsize — the Savings Plan commitment applies regardless of which instance families we use. We committed to $4.20/hour which covers 72% of our on-demand eligible compute, at 56% discount to on-demand rates."

A comprehensive assessment of all costs associated with a technology decision over its full lifetime — including acquisition, operation, maintenance, training, migration, and decommissioning costs. Used to compare cloud vs. on-premises, build vs. buy, or competing architecture choices on a level basis.

"The TCO analysis for cloud vs. on-premises surprised leadership: on-premises hardware was 40% cheaper on compute cost alone, but when we included staffing, data centre rent, power, cooling, network, and the 3-year hardware refresh cycle, cloud was 18% cheaper on true TCO and offered elasticity on-premises couldn't match."

An unexpected spike or unusual pattern in cloud spending detected by automated monitoring — indicating a misconfiguration, runaway process, data transfer incident, or accidental resource proliferation. Cost anomaly detection is a FinOps best practice for preventing bill shock.

"AWS Cost Anomaly Detection alerted us to a $12,000 unexpected charge on a Tuesday afternoon — an engineer had accidentally deployed a data pipeline to the wrong environment and was writing 4 TB/hour of logs to S3. The alert caught it 40 minutes after it started, limiting the impact to $340 instead of a potential $50,000 monthly bill."