Stabilising the largest US floral marketplace

Challenge

The marketplace brings together more than 4,000 independent florists across the US: a customer orders a bouquet, and the nearest shop arranges and delivers it. Almost all revenue comes in a few days a year: Valentine’s Day, Mother’s Day and the holiday season.

Those were exactly the days the platform struggled. Checkout depended on synchronous calls to external systems, and under load shoppers got timeouts. Florists, meanwhile, could not update their range or prices without help from support.

Solution

• Asynchronous order queueWe moved order processing into a RabbitMQ queue. Shoppers get confirmation immediately, while handing the order to the florist and back-office systems runs in the background and survives temporary failures.
• Caching and searchWe moved caching to Redis and search to Elasticsearch, so the catalogue and florist lookup by address no longer load the database on peak days.
• Florist dashboardWe built a simple dashboard where each shop manages its own range, prices, delivery area and hours.
• Salesforce integrationWe synced orders and florist data with Salesforce so support and sales see the full picture.

How we worked

1. Analysing peaks

   We studied logs and metrics from past holidays to find exactly where checkout hit its limits.

2. Changes before the season

   Major changes were scheduled between holidays, with a release freeze ahead of every peak.

3. Load testing

   Before each season we ran load tests on scenarios that replay real holiday traffic.

4. On call during peaks

   On key days the team was on call, watching queues and response times in real time.

Outcome

Checkout timeouts went away, and holiday peaks became business as usual rather than a fire drill. Florists run their own storefronts, and support has less to handle.