Background: A Global Media Giant Running Its Streaming Empire on AWS
The company is a major global media organisation operating across three core business pillars: linear broadcasting, digital content production, and direct-to-consumer streaming services. With 27,000 employees spanning content studios, broadcast facilities, regional offices, and technology centres across North America, Europe, and Asia-Pacific, the company's AWS infrastructure is the backbone of its entire digital operation — powering video ingest and transcoding, content delivery to millions of concurrent viewers, real-time audience analytics, advertising technology, and the media asset management systems that support a content library of hundreds of thousands of hours.
The company's AWS footprint had grown dramatically as its streaming platform scaled from launch to tens of millions of subscribers. Amazon CloudFront handled global content delivery. AWS Elemental MediaConvert and MediaLive powered video transcoding and live streaming workflows. Amazon EC2 ran the streaming platform's application tier, recommendation engines, and advertising insertion systems. Amazon S3 stored the content library, production assets, and analytics data lakes. Amazon Redshift and Kinesis supported real-time audience analytics and engagement measurement. The annual AWS bill had grown in parallel with subscriber growth — but cost optimisation had not kept pace with consumption growth, creating significant waste across over-provisioned compute, inefficient data transfer routing, and resources provisioned for peak loads that sat idle during off-peak hours.
With the AWS commitment up for renegotiation and the company's streaming platform entering its scale-up phase, the CTO engaged Redress Compliance to conduct an independent assessment — ensuring the new agreement reflected the company's actual consumption patterns and growth trajectory rather than AWS's revenue maximisation objectives.
"Media and streaming companies are among AWS's highest-value customers — and among the most complex to optimise. The combination of massive egress costs (content delivery to millions of viewers), highly variable compute demand (live events spike 10–50× normal traffic), and the media industry's 'always-on' production culture (where over-provisioning for resilience is the default) creates an AWS bill where 25–40 % of spending typically delivers no incremental business value. The $6.7 million in savings we achieved here came from three sources: eliminating waste in steady-state infrastructure, restructuring data transfer to use the most cost-effective delivery paths, and negotiating commitment structures that accommodate the inherent variability of media consumption patterns."
The Challenges: Streaming Scale, Egress Costs, and Event-Driven Demand
Content Delivery Egress Costs
Data transfer out (egress) represented the single largest component of the company's AWS bill — driven by streaming video delivery to millions of subscribers across multiple regions. The existing architecture routed all content through CloudFront without optimising for regional edge caching efficiency, origin shield configuration, or multi-CDN strategies. High-definition and 4K content multiplied per-viewer egress costs, and the company had not negotiated volume-based data transfer discounts that reflected its scale as one of AWS's largest media customers.
Live Event Demand Spikes
The company's live programming — sports broadcasts, live news, premieres, and tentpole events — created demand spikes of 10–50× normal concurrent viewership. The existing infrastructure handled these spikes through persistent over-provisioning: EC2 instances sized for peak capacity ran 24/7 even though peak events occurred for only a fraction of total operating hours. Live transcoding infrastructure (MediaLive channels) was similarly over-provisioned, with channels running continuously to avoid cold-start latency during live events — even when no live content was scheduled.
Storage and Processing Sprawl
The company's S3 storage had grown to petabytes of content, production assets, and analytics data without systematic lifecycle management. Infrequently accessed content (archive titles, historical production assets, legacy analytics) remained in S3 Standard storage at premium pricing. Video transcoding workflows used MediaConvert at default settings without optimising encoding profiles, instance types, or queue configurations — running high-specification encoding for content that would ultimately be viewed on mobile devices at lower quality tiers. Orphaned processing resources from completed production projects continued running.
Analytics Over-Provisioning
The real-time analytics stack — Redshift clusters, Kinesis streams, and associated compute — had been provisioned for peak audience measurement during the streaming platform's rapid growth phase. As subscriber growth stabilised, the analytics infrastructure remained sized for the growth trajectory rather than actual demand. Redshift clusters ran at 30–40 % average utilisation, Kinesis stream capacity exceeded actual event throughput by 3–4×, and multiple analytics environments (development, staging, pre-production) duplicated the full production configuration unnecessarily.
Phase 1: Comprehensive Cloud Consumption Analysis
Service-by-Service Spend Mapping
Redress conducted a granular analysis of every AWS service the company consumed, mapping spending to business function: streaming delivery (CloudFront, data transfer), video processing (MediaConvert, MediaLive, Elemental), compute (EC2 for platform, recommendations, ad insertion), storage (S3 content library, production assets, analytics data), and analytics (Redshift, Kinesis, EMR). The analysis identified the top 20 cost drivers that represented approximately 85 % of total AWS spend — prioritising optimisation efforts on the highest-impact services rather than attempting to optimise everything simultaneously.
Usage Pattern and Peak/Off-Peak Analysis
Redress mapped the company's consumption patterns across 24-hour, weekly, and seasonal cycles. Streaming viewership peaked in evening hours and weekends, with live events creating irregular demand spikes. Content transcoding was heaviest during post-production cycles. Analytics processing peaked during quarterly business reviews and advertising upfront presentations. This temporal analysis quantified the gap between provisioned capacity and actual utilisation — revealing that average utilisation across the EC2 fleet was approximately 35 % of provisioned capacity, with peaks reaching 70–80 % only during major live events.
Data Transfer and CDN Efficiency Assessment
Redress analysed CloudFront distribution configurations, cache hit ratios, origin fetch patterns, and regional egress costs. The assessment identified three categories of egress waste: sub-optimal edge caching (content fetched from origin more frequently than necessary due to cache configuration), absence of origin shield (multiplying origin requests across edge locations), and premium-tier data transfer routes used for content that could have been delivered via lower-cost paths. The analysis also evaluated multi-CDN opportunities — where supplementing CloudFront with additional CDN providers could reduce per-GB delivery costs for specific regions.
Phase 2: Optimisation — $1.8 M in Waste Eliminated
| Category | Finding | Action Taken | Annual Savings |
|---|---|---|---|
| Compute right-sizing and scheduling | EC2 fleet at ~35 % average utilisation; live event infrastructure running 24/7; orphaned instances from completed productions | Right-sized instances; implemented auto-scaling with event-driven triggers; terminated orphaned resources; converted steady-state to Reserved/Savings Plans | $620 K |
| Data transfer optimisation | Sub-optimal CloudFront caching; no origin shield; premium transfer routes for standard content | Optimised cache policies and TTLs; deployed origin shield; restructured transfer routing; negotiated volume egress discounts | $480 K |
| Storage lifecycle management | Petabytes in S3 Standard including archive content and legacy assets; no lifecycle policies | Implemented tiered storage: S3 Standard for active, Infrequent Access for catalogue, Glacier for archive; automated lifecycle transitions | $390 K |
| Analytics and processing right-sizing | Redshift at 30–40 % utilisation; Kinesis 3–4× over-provisioned; duplicate non-production environments | Right-sized Redshift clusters; reduced Kinesis shard count; consolidated non-production environments; implemented Redshift Serverless for variable workloads | $310 K |
| Total waste eliminated | Annual savings from optimisation alone | $1.8 M | |
Phase 3: Strategic Commitment Restructuring
Volume-Based Egress and Multi-CDN Strategy
Redress restructured the company's content delivery economics through two mechanisms. First, volume-based egress pricing tiers negotiated with AWS that reflected the company's scale — moving from standard published rates to custom pricing that recognised the company as one of AWS's largest media egress customers. Second, a multi-CDN architecture that routes content delivery through CloudFront for regions where AWS offers the best performance and pricing, while supplementing with alternative CDN providers for regions where competing CDNs offer lower per-GB costs — reducing overall content delivery expense while maintaining viewer experience quality.
Event-Driven Scaling
Live event infrastructure was restructured from persistent over-provisioning to event-driven auto-scaling. Pre-configured scaling policies automatically provision additional EC2 capacity, MediaLive channels, and CloudFront distribution capacity 2–4 hours before scheduled live events — and scale down within 1 hour after the event concludes. For unscheduled breaking news events, warm standby pools maintain a minimum responsive capacity that can scale to full live-event levels within minutes. This approach delivers the same live-event capacity at 60–70 % lower cost than persistent provisioning.
Spot and Burst Compute for Transcoding
Video transcoding and post-production processing workloads were restructured to use EC2 Spot Instances for fault-tolerant batch processing (catalogue transcoding, format conversion, quality ladder generation) and Reserved Instances only for time-sensitive live transcoding. MediaConvert jobs were optimised with encoding profiles tailored to output quality tiers — reducing processing costs by avoiding high-specification encoding for content destined for mobile and lower-resolution delivery. Spot pricing delivers 60–90 % discounts compared to on-demand for batch transcoding workloads that can tolerate interruption.
Phase 4: Benchmarking and Negotiation — 32 % Cost Reduction
Redress benchmarked the company's AWS pricing against peer global media companies, streaming platforms, and large-scale content delivery customers — including comparable broadcast-streaming operations and digital-native media companies. The benchmarking revealed pricing premiums across EC2 compute, data transfer egress, and storage — with the largest gap in egress pricing, where the company's scale warranted volume discounts significantly deeper than its current rates.
| Metric | Previous Agreement | New Agreement (Post-Negotiation) |
|---|---|---|
| Annual AWS cost | Baseline (100 %) growing with subscriber base | 68 % of previous — 32 % reduction |
| Commitment structure | Monolithic 3-year commit; no workload differentiation | Tiered: steady-state Reserved + Savings Plans; event-driven auto-scale; Spot for batch processing |
| Data transfer / egress | Standard published rates; single-CDN delivery | Volume-based custom egress pricing; multi-CDN architecture; origin shield deployed |
| Live event capacity | 24/7 persistent over-provisioning for peak events | Event-driven auto-scaling; warm standby for breaking news; 60–70 % cost reduction for live capacity |
| Storage | Petabytes in S3 Standard; no lifecycle management | Tiered lifecycle: Standard → Infrequent Access → Glacier with automated transitions |
| Scaling flexibility | Rigid annual commitment; no provisions for subscriber growth variability | Annual step-up/step-down; subscriber-linked commitment adjustments; quarterly review provisions |
| 3-year total savings | $6.7 million — $1.8 M optimisation + $4.9 M negotiated discounts and structural savings | |
📊 Key Negotiated Concessions
- Custom egress pricing: Volume-based data transfer rates reflecting the company's scale as a top-tier media customer — significantly below standard published rates for CloudFront and inter-region transfer
- Event-driven scaling provisions: Contractual guarantee of EC2 and MediaLive capacity availability during scheduled live events with 4-hour pre-provisioning and 1-hour wind-down — no persistent over-provisioning required
- Subscriber-linked commitment: Annual AWS commitment that adjusts based on actual subscriber count — if subscriber growth exceeds or falls below projections, the commitment adjusts proportionally rather than requiring mid-term renegotiation
- Spot Instance priority: Enhanced Spot capacity availability for the company's batch transcoding workloads — reducing interruption rates that affect content processing timelines
- Savings Plan flexibility: Convertible Savings Plans that allow the company to change instance families and regions as its infrastructure evolves — avoiding lock-in to specific instance types
- Enterprise Discount Programme (EDP): Custom EDP rates with annual step-up provisions that reflect validated growth rather than aspirational projections
- Multi-region data residency: Contractual terms for content and viewer data processing across US, EU, and APAC regions — supporting GDPR, content licensing territorial restrictions, and local data processing requirements
Client Testimonial
"Redress Compliance's expertise in AWS negotiations was instrumental in optimising our cloud strategy. Their insights helped us save substantially while ensuring our infrastructure supports innovation and audience growth. Their support has been transformative."
— Chief Technology Officer, Global Media Company
Phase 5: Governance and FinOps Framework
Real-Time Cloud Cost Monitoring
Redress implemented a FinOps (Cloud Financial Operations) framework providing real-time visibility into AWS spending by business function: streaming delivery, live events, content production, analytics, and corporate. The dashboard tracks cost-per-viewer, cost-per-transcoding-hour, cost-per-GB-delivered, and other media-specific unit economics — enabling the company to understand how AWS costs relate to business metrics rather than just monitoring aggregate cloud spending. Anomaly detection alerts flag unexpected cost spikes within hours, allowing investigation before they compound.
Automated Resource Lifecycle Management
The governance framework includes automated lifecycle policies for every resource category: S3 objects transition through storage tiers based on access patterns, EC2 instances are tagged with business function and production status (allowing automated identification of orphaned resources), MediaConvert jobs use pre-configured encoding templates optimised for cost-quality balance, and non-production environments automatically scale down outside business hours. These automation policies prevent the resource sprawl that characterised the previous environment.
Quarterly Optimisation Reviews and Training
Quarterly AWS reviews compare actual consumption against commitment levels, assess upcoming content launches and live event schedules (to pre-plan scaling), evaluate new AWS service offerings for potential cost or capability improvements, and prepare adjustment recommendations for the annual commitment review. Redress delivered targeted training for the company's infrastructure, streaming, and production engineering teams covering AWS pricing models, reservation strategies, egress optimisation, and the governance procedures for provisioning new resources — embedding cost-awareness into the engineering culture rather than treating cloud cost management as a finance function.
Outcome: Before and After
| Metric | Before Redress Engagement | After Redress Engagement |
|---|---|---|
| Annual AWS cost | Growing with subscriber base; no cost-per-unit optimisation | 32 % reduction; $6.7 M saved over 3 years; FinOps governance established |
| Content delivery | Standard egress rates; single-CDN; sub-optimal caching | Volume-based custom pricing; multi-CDN; origin shield; optimised cache policies |
| Live event infrastructure | Persistent 24/7 over-provisioning; peak-sized continuously | Event-driven auto-scaling; warm standby; 60–70 % live-event cost reduction |
| Video processing | Default encoding settings; on-demand pricing for batch work | Optimised encoding profiles; Spot for batch; Reserved for live only |
| Storage | Petabytes in S3 Standard; no lifecycle management | Automated tiered lifecycle; archive content in Glacier; 40–60 % storage cost reduction |
| Total financial impact | $6.7 million saved over 3 years — 32 % annual cost reduction | |
Lessons for Media and Streaming Companies
Negotiate Egress Pricing Aggressively — It's Your Largest Cost Lever
For media and streaming companies, data transfer out (egress) typically represents 30–50 % of the total AWS bill — driven by video delivery to viewers. AWS's published egress rates are the starting point for negotiation, not the final price. Companies delivering petabytes of content monthly are among AWS's most valuable customers and should negotiate custom volume-based egress pricing that reflects their scale. Combined with CDN optimisation (origin shield, cache policy tuning, multi-CDN for regional cost arbitrage), egress cost reductions of 25–40 % are achievable without any impact on viewer experience.
Replace Persistent Over-Provisioning with Event-Driven Scaling
The media industry's default approach to live events is over-provisioning — keeping peak-sized infrastructure running 24/7 to ensure availability when live events occur. This approach is enormously wasteful because peak events represent a small fraction of total operating hours. Modern auto-scaling with pre-configured event policies can deliver the same peak capacity at 60–70 % lower cost by provisioning infrastructure hours before events and scaling down hours after. For companies with predictable event schedules (sports, premieres, upfront presentations), the savings are especially significant.
Implement Storage Lifecycle Management Immediately
Media companies generate and store massive volumes of content that follow predictable access patterns: new releases are accessed frequently, catalogue titles are accessed occasionally, and archive content is accessed rarely. Yet most media companies store everything in S3 Standard — the most expensive tier — because lifecycle policies were never implemented. Automated transitions (Standard → Infrequent Access after 90 days → Glacier after 1 year) typically reduce storage costs by 40–60 % with zero impact on content availability for the vast majority of the library. This is the easiest, lowest-risk optimisation available.
Build FinOps Around Media Unit Economics
Cloud cost management in media companies should be measured in business-relevant unit economics — cost-per-viewer, cost-per-stream-hour, cost-per-transcoding-hour — not just aggregate AWS spend. When engineering teams see that a specific architecture decision costs $0.003 per viewer-hour versus $0.001 per viewer-hour, they make different design choices than when they see only the aggregate monthly bill. Embedding unit-economics visibility into the engineering culture transforms cloud cost management from a finance exercise into an engineering discipline — which is where the largest sustained savings come from.