Cut page load from 12s to under 400ms for three luxury brands.

The problem

MyTCS is the post-booking portal for TCS World Travel, Smithsonian Journeys and Four Seasons. Salesforce is the source of truth, and every page waited on it.

What we did

• Put a Redis getOrSet cache in front of Salesforce queries.
• Moved slow operations onto SQS FIFO queues so pages stopped waiting on them.
• Lazy-loaded components, prefetched images and split CSS per route.

Key decisions

1. 01

   Redis getOrSet cache in front of Salesforce

   Trade-off: Accepted brief staleness on cached reads in exchange for sub-second pages.

2. 02

   SQS FIFO queues for async operations

   Trade-off: Writes became eventually consistent; ordering was kept over raw throughput.

3. 03

   Semantic token theming for three brands

   Trade-off: More upfront token design instead of maintaining per-brand forks.

4. 04

   Lazy-loading and code-splitting

   Trade-off: Small extra requests on navigation for a much faster first load.

Architecture

Screenshots

• 

  MyTCS dashboard, Four Seasons brand

• 

  MyTCS destination page, Smithsonian Journeys brand

Result

Page load dropped from 12s to under 400ms, on one codebase serving three brands.

What I learned

1. 01

   Measure where time goes before optimising: the wait was on Salesforce, not React.

2. 02

   A cache in front of a system of record needs an agreed staleness budget with the business.

Stack

• React 18
• TypeScript
• tRPC
• NestJS
• Redis
• AWS SQS
• Prisma
• PostgreSQL
• Auth0
• Datadog