Contact us
A global white-label auction platform runs live sales for more than 400 branded auction houses, handling thousands of bids a minute. We split the live-auction services apart, folded each house's configuration into one release pipeline, moved partners onto queued events, and gave every tenant one identity system while keeping its data separate, because every release used to mean carrying each house's branding and settings across by hand.
Every release went out four hundred times.
The biggest sale carried the most risk.
As the platform reported.
Each auction house has its own branding, feature settings and partner connections for payments, identity checks, shipping and marketing. Engineers had to carry those differences through every release by hand, without disturbing a live sale, which meant a ten-minute change could take days to reach every instance, and release frequency was capped at a handful a month.
An auction house's biggest events are also the platform's biggest revenue moments, because real money gets bid in real time. Bidding, payments and inventory lived in one application and could only scale together. Partner integrations were wired straight into the core, which meant a misbehaving payment or shipping provider could affect bidding. Because authentication was handled separately on each instance, compliance reviews grew longer with every new auction-house contract.
We started treating the 400 auction houses as one product with 400 configurations, where before there had been 400 separate products. A release gets built once, the pipeline applies each house's settings on top of it, and the live-auction services stay apart from the partner connections so neither can take the other down.
4 decisions, in the order we made them. Each one says what we did and why, with what it did drawn beside it.
The services that handle bidding and payments run on capacity dedicated to them, so other parts of the system can't pull it away.
Configuration-driven pipelines manage branding, feature flags and environment settings across more than 400 deployments. A shared release uses each auction house's configuration, with no separate manual release procedure per house. CodePipeline and CodeBuild automate delivery, and CloudFront serves the branded assets stored in S3. Monitoring tracks every release against the operational and security controls auction houses are audited on.
SNS, SQS and EventBridge carry the message-based integration model. Payment, identity, shipping and marketing connections exchange standardized events instead of adding another direct dependency to every service. Queued communication also lets the platform handle a delayed partner separately from the live bidding path, so a partner outage is contained instead of cascading.
This matters because in a live auction, a slow shipping partner can't be allowed to slow down the bidding.
Cognito centralizes authentication, so credential rotation, MFA and access policy get set once instead of 400 times. IAM and per-tenant data controls keep each white-label customer's data apart, with separation enforced rather than logical. That replaces the work of maintaining a separate identity setup for each branded instance, and gives large auction houses' legal teams the isolation they ask for.
As the platform reported.
Ongoing platform engineering on the same release model, with more auction houses and partners joining through configuration.
This project came before our current offers. Here's how we'd approach the same work now.
Auction rules are exactly the kind nobody wrote down, so the services would be cut out with AK Evolve: AI agents read the old code into specs, characterization tests pin what it does today, and each service is rebuilt against those tests behind a routing layer, going live on its own with a way back. Releases would reach a small share of the 400 auction houses first and roll back on their own if quality, speed or errors cross a line agreed in advance, with the four DORA measures of delivery visible from the first weeks.
AK Run's AI agents would take the first look at every alert and draft the fix, for a named engineer to confirm and ship with a test through the platform's own pipeline, and we'd own each security finding until a tested fix is live.
A free workshop: a few hours with the architects who'd do the work, on your own systems. No cost.
Book a free workshop