Egress reduction: $2.1M per year recovered.

The customer's AWS bill was running roughly $9M annually, of which $3.8M — over 40 percent — was data transfer. Outbound transfer to internet, CloudFront origin fetches, and NAT Gateway data processing were the three largest contributors. The customer had largely treated data transfer as a fixed cost of doing business and had never opened the topic with AWS.

A new VP of Engineering had run the numbers against industry benchmarks and concluded the data transfer share of bill was at least 2x what comparable retail e-commerce operations were paying. The diagnostic was correct; the customer was significantly overpaying on egress, both contractually and architecturally.

What the customer needed

• A clear breakdown of what was actually driving the egress spend
• An architectural read on what was negotiable contractually vs. what required workload changes
• A CloudFront private pricing structure that matched the actual delivery volume
• A negotiated PPA overlay on the residual direct-to-internet egress

Egress is the most negotiable category in the AWS bill at any spend level above roughly $250K monthly, and the most under-negotiated. AWS's account teams rarely volunteer the available overlays; the customer has to ask, and the customer has to have the benchmarks to know what to ask for.

Phase 1 — Decompose the spend (weeks 1-3)

We pulled 90 days of detailed billing data and tagged every transfer line by source service, destination, and traffic profile. The decomposition produced a clear breakdown: 48 percent CloudFront origin fetches, 27 percent direct outbound from EC2/S3, 18 percent NAT Gateway data processing, and 7 percent miscellaneous (PrivateLink, cross-region, Transit Gateway).

The NAT Gateway line was a surprise. The customer had not realized that several million dollars of annual spend was running through NAT for service-to-service calls that should have been using VPC endpoints. That diagnosis is purely architectural; the negotiation lever does not help.

Phase 2 — Architectural quick wins (weeks 4-5)

We replaced direct NAT Gateway data paths with VPC Gateway and Interface endpoints for the three largest services (S3, DynamoDB, and SQS). The change took the engineering team five working days to implement and removed roughly $340K of recurring annual NAT data processing charges with no negotiation involved.

Phase 3 — CloudFront and egress PPA (weeks 6-9)

In parallel, we opened the contract conversation with AWS. We benchmarked the CloudFront line against the customer's actual delivered volume and pulled the corresponding private pricing tier that AWS had available but had not offered. We added a PPA overlay on the residual direct outbound to internet, sized against the customer's three-year forecast.

The final contract structure included a CloudFront private pricing tier producing 38 percent reduction on origin fetches, a PPA overlay on direct outbound producing 44 percent reduction, and a re-baselining clause that protects the customer's discount if delivered volume scales above forecast.

The combined architectural and contractual changes produced $2.1M in annual recurring savings against the prior 12-month run rate. The savings break across three buckets.

Where the savings came from

• CloudFront private pricing — $1.1M annual savings from the new private pricing tier on origin-fetch volume
• Direct egress PPA overlay — $660K annual savings from the new overlay on direct-to-internet outbound transfer
• NAT Gateway architectural change — $340K annual savings from moving service-to-service calls onto VPC endpoints

What the customer did with the savings

The freed budget funded the customer's first material investment in personalization infrastructure, which had been deferred for two quarters due to budget pressure. The remaining savings dropped to gross margin.

The re-baselining clause is the long-lived change. The customer's delivered video and image volume has grown 40 percent since the engagement closed; the private pricing tier has held, and the negotiated discount has scaled with the volume instead of degrading.