AI / FinOps

FinOps-Powered Cost Optimisation and Cloud Cost Visibility for AI Workloads

Client

PlotWeaver AI

Partner

Marlocks Technologies

Focus

FinOps

Overview

PlotWeaver AI is a fast-growing AI startup building generative AI products that depend on heavy inference workloads across Amazon Bedrock, Amazon SageMaker, and GPU-backed EC2 instances. As their user base scaled, cloud spend scaled faster than revenue — a familiar pattern for AI startups operating at the intersection of high-cost foundation models and rapid product iteration.

PlotWeaver engaged Marlocks Technologies to bring discipline, visibility, and structural cost reduction to their AWS environment without compromising the performance their product depends on.

PlotWeaver AI cloud cost overview

Key Challenges

The shape of an AI startup's cloud bill makes it uniquely difficult to control.

Opaque Spend Drivers

Monthly AWS bills grew month-on-month with no clear understanding of which services, environments, or product features were driving the increase.

Untagged Resources

Compute and storage resources had been provisioned organically across multiple accounts with no consistent tagging, making cost allocation to product lines or customer cohorts impossible.

Over-Provisioned Inference

GPU instances ran continuously to serve unpredictable traffic, leaving capacity idle during off-peak hours and inflating cost per inference.

No Commitment Strategy

All consumption was on-demand — no Savings Plans, no Reserved Instances, no Spot adoption — leaving 20–40% of typical savings on the table.

Reactive Cost Management

No budgets, no anomaly detection, no alerts. Cost surprises only surfaced at month-end when the invoice arrived.

The Solution

Marlocks delivered a FinOps-first engagement designed to deliver immediate cost reduction while installing the practices, tooling, and governance for sustained cost discipline.

Cost & Usage AssessmentDeep analysis of AWS Cost and Usage Report (CUR) over a 90-day window using AWS Cost Explorer and a custom Amazon QuickSight dashboard, isolating top cost drivers by service, account, and workload.

Tagging & Cost Allocation StrategyStandardised tagging schema (environment, product, owner, customer-segment) applied across all resources and enforced going forward via AWS Config and tag policies.

Right-Sizing & Compute OptimisationMigration of inference workloads from over-provisioned GPU EC2 instances to Amazon Bedrock on-demand and SageMaker Serverless Inference, with Spot Instances adopted for non-production training and batch jobs.

Commitment-Based DiscountsAnalysis of steady-state baseline consumption and purchase of a tailored Compute Savings Plan covering predictable workloads, with on-demand capacity reserved for elastic spikes.

Real-Time VisibilityAWS Budgets with multi-threshold alerts, AWS Cost Anomaly Detection across all linked accounts, and weekly FinOps dashboards delivered to PlotWeaver's leadership team.

Knowledge TransferEngineering team upskilled through a structured FinOps enablement programme so cost decisions become part of the daily engineering rhythm rather than a quarterly fire-drill.

PlotWeaver FinOps architecture

Key Results

Total AWS Bill Reduction

34% in first 60 days

Sustained through subsequent months as new architecture absorbed user growth

Cost per Inference Reduction

48%

Via right-sizing, Bedrock adoption, and elimination of idle GPU capacity

Cost Attribution Coverage

100% of spend tagged

Attributable to product line, environment, and customer cohort

Anomaly Response Time

Under 48 hours

Reduced from 30+ days — surfacing within 24 hours via automated alerts

Scalability

3x users at 1.4x cost

Breaking the previous linear cost-to-usage relationship

Key Lessons Learned

Tagging discipline is the foundation of FinOps:

Without consistent tagging, no amount of analysis or commitment optimisation can deliver durable savings. Tagging must be designed early and enforced through automation, not policy alone.

Serverless and managed inference shift the cost curve:

Migrating suitable workloads to Bedrock and SageMaker Serverless Inference broke the linear relationship between user growth and infrastructure cost — the single highest-leverage change in the engagement.

FinOps is a cultural shift, not a one-time project:

The most lasting outcome was embedding cost awareness into engineering decision-making. Once developers can see the cost impact of their choices in real time, sustained savings follow without a consultant in the room.