Retail infrastructure has an awkward habit: it must be ready for the most exciting day of the year on every unexciting day that comes before it. For AllSaints, the global fashion brand, that meant keeping more than 60 physical and virtual servers prepared for the sudden crush of seasonal traffic—even while much of that machinery sat underused through ordinary shopping weeks.
The arrangement bought resilience at a punishing price. Capacity planned for the rush became capital stranded in the quiet. The website depended on a hybrid estate whose complexity consumed engineering attention, and every layer had to keep working because a slow product page or stalled checkout is not merely a technical problem. In online retail, it is a customer walking out.
AllSaints brought in Pythian to change that equation. The brief was ambitious: consolidate the retailer's infrastructure on Google Cloud, refactor a web of legacy services and improve the economics of peak readiness. The constraint was even sharper. Shoppers should not notice any of it.
01 · The old bargainPeak traffic made idle servers look necessary
The old model was logical in the way many expensive systems are logical. Black Friday, holiday promotions and product drops create steep, temporary demand. Owning enough capacity to meet the highest forecast protects the storefront at its most valuable moment. Yet for the rest of the calendar, AllSaints was paying for a safety margin measured in dormant machines.
Retail makes that mismatch unusually unforgiving. Demand does not rise in a smooth line that an infrastructure team can follow at leisure. It arrives in bursts, sometimes because a campaign worked and sometimes because a product caught fire without warning. A platform provisioned for the average risks failing at precisely the moment the business succeeds. A platform provisioned for the maximum turns preparedness into a standing expense. The technical problem is therefore a financial one: how to keep the upside of spare capacity without paying for all of it in advance.
That fixed footprint also hid a second cost. More than raw compute, the retailer was maintaining a dense collection of dependencies. Pythian's account describes 100 distinct legacy services that needed to be rewritten or decoupled into a distributed microservices model. Moving them unchanged would have relocated the complexity, not removed it.
This was the biggest infrastructure change in the history of the company, and we wanted to do it in such a way that our customers wouldn't notice the change.Andy Dean · Technical Operations Manager, AllSaints
Dean's standard captures the paradox of infrastructure work. The deeper the intervention, the less visible success should be. No grand reopening. No customer tutorial. Just the same storefront responding more quickly while its foundations are replaced beneath it.
02 · The rebuildNot a lift-and-shift, but a change in physics
Pythian designed the new platform around Google Compute Engine and Google Kubernetes Engine. Containers and orchestration gave services a consistent way to run; autoscaling let resources rise and fall with real traffic. The business no longer needed to hold every server at parade rest in case shoppers arrived.
This is the useful distinction in the AllSaints story. A cloud invoice is not automatically smaller than a data-center bill. Savings appear when the operating model changes—when idle capacity can switch off, applications can scale independently and software is shaped for the platform underneath it. The team re-architected rather than simply repackaged.
The architecture shift
The cutover also had to account for geography. AllSaints serves customers across regions, which makes continuous availability less a slogan than a scheduling fact: somewhere, the store is open. The new environment was built to preserve performance while traffic kept moving through it.
03 · The operating systemFaster shipping started behind the storefront
Infrastructure modernization can save money and still leave engineers trapped in yesterday's workflow. Pythian paired the platform change with a continuous integration and delivery pipeline built using Jenkins and HashiCorp Terraform. Service deployments that had taken 20 minutes fell to under five.
The arithmetic matters, but the behavioral change matters more. A shorter deployment is a smaller decision. It reduces the ceremony around releasing an improvement and makes recovery less intimidating when something goes wrong. Engineers can stop treating aging hardware as a patient in permanent care and return attention to features a shopper can actually feel.
Terraform added repeatability to that speed. Instead of relying on a sequence of manual changes remembered by a few specialists, infrastructure could be described in code, reviewed and applied consistently. Jenkins then connected application changes to a delivery path. Together, those tools helped turn deployment from a bespoke event into a routine capability. That distinction is easy to miss in a migration diagram, but it determines whether a modern platform remains modern after the consultants leave.
Visibility had to improve with velocity. Percona Monitoring and Management, working alongside Google Stackdriver, gave AllSaints and Pythian shared, real-time diagnostics across database and application layers. One operational picture meant fewer handoffs between teams and a clearer answer when performance moved in the wrong direction.
04 · The compounding returnSpeed became a commercial result
The migration's most legible number is the 75% reduction in operating costs reported by Pythian. But the other numbers show why this was larger than an infrastructure-efficiency exercise. Page-load speed improved 35%. Conversions increased 20%. The machinery got cheaper at the same time the storefront became more productive.
Those outcomes are connected. Faster pages preserve attention between discovery and checkout. Elastic capacity protects that speed when traffic jumps. Automated releases help teams improve the experience more frequently. Better monitoring shortens the path from a warning to an explanation. None of these elements carries the result alone; together they form a system that learns faster and wastes less.
BigQuery added another layer. By aggregating and processing multi-region e-commerce transactions, the data warehouse gave business intelligence teams a way to turn raw behavior into operational decisions. The platform could now tell the business more about how it was being used, not simply remain upright under the weight.
The loudest outcome came from an invisible change: fewer idle machines, faster pages and more completed purchases.
05 · The takeawayBuild for the rush, pay for the moment
AllSaints' experience offers a practical test for any modernization program. Does it alter the economics of demand, or merely redraw the architecture diagram? Does it make the delivery team faster, or only give them new tools? Does observability become shared and actionable? Most important, does the customer receive the benefit without carrying the disruption?
The project met those tests by treating cloud migration as an operating redesign. Capacity followed traffic. Services became independently manageable. Infrastructure could be expressed and repeated through code. Performance data became visible across team boundaries. The fashion stayed on the page; the heavy machinery receded behind it.
There is something appropriately retail about the final bargain. AllSaints stopped stocking compute for every possible customer before they arrived. It built a platform that could replenish itself in real time. The largest infrastructure change in the company's history succeeded precisely because, for the person adding a jacket to a basket, nothing seemed to change at all.
Reported results and technical details are based on Pythian's AllSaints customer success story.