WEB DEVELOPMENT

SaaS Platform Rebuild: From 6s to 0.4s LCP

Complete platform rebuild using Astro + Preact Islands, cutting load times by 87% and doubling conversion rates.

Client: CloudMetrics (illustrative)

Illustrative example. Client name and metrics are an illustrative example of a representative engagement, not a specific real customer.
98
Lighthouse
+142%
Traffic
0.4s
LCP
+96%
Conversions

"RexBunny didn't just rebuild our site — they rebuilt our business. The performance gains translated directly to revenue."

— Illustrative client
Sample quote shown for illustration; not attributed to a real named customer.

The Challenge

A representative SaaS client arrived with a monolithic React app that took 6.2 seconds to load on mobile. Their Lighthouse score was 34. Bounce rates were climbing. They needed a complete rethink, not a patch.

Our Approach

Architecture

We migrated from a single-page React application to Astro with Preact Islands. Content pages became static HTML (zero JavaScript). Interactive dashboards became isolated Preact components that load only when needed.

Performance Engineering

  • Font optimization: Subset Inter to Latin characters only, preload critical fonts
  • Image pipeline: Automatic WebP conversion, responsive srcsets, lazy loading
  • Code splitting: Every page ships only the JavaScript it needs
  • Edge deployment: Content served from 200+ edge locations globally

AI Search Optimization

  • Added llms.txt with structured business information
  • Implemented LocalBusiness and SoftwareApplication schema
  • Configured robots.txt to allow AI crawlers
  • Created citation-friendly content structure

Results

MetricBeforeAfterChange
Lighthouse3498+188%
LCP6.2s0.4s-93%
CLS0.220.01-95%
TBT1,200ms42ms-97%
Bounce Rate68%31%-54%
Conversion Rate2.1%4.1%+96%

The performance improvements also boosted AI visibility. In a representative case, the site moved from zero AI citations to being surfaced in ChatGPT responses for “cloud analytics platform” queries within 60 days — a new channel the client hadn’t reached before.


Related reading: