SaaS Platform Migration: Cutting AWS Compute Costs by 68% with React & Neon PostgreSQL

In early 2026, an Indian B2B SaaS startup reached out to Webeta with a critical bottleneck: their AWS bill had ballooned to $4,200/month despite having under 12,000 active monthly users, and their marketing platform was taking 4.6 seconds to load. Here is the step-by-step engineering teardown of how we rebuilt their architecture, slashed cloud overhead by 68%, and cut page load times to 420 milliseconds.

The Client Problem: Architecture Debt & Cost Explosion

The client operates a SaaS workflow automation platform for freight forwarders. Two years prior, their initial development agency built what they called an "enterprise microservices architecture."

In reality, it was an operational disaster:

• A Dockerized Node.js frontend server doing synchronous SSR on every single page request.
• Separate authentication, billing, tracking, and notification microservices communicating over internal HTTP.
• A provisioned AWS RDS multi-AZ PostgreSQL instance running 24/7 at 12% average CPU utilization.
• A marketing website glued directly to the application container, meaning a DDoS on marketing took down active customer workspaces.

Phase 1: Decoupling Public Marketing from Application State

The first move was isolating public discovery from private application logic. We extracted the marketing site, pricing calculator, and lead intake forms out of the containerized Docker pipeline and migrated them to a pre-rendered React architecture.

With pre-rendering, every marketing page is baked into static HTML files at build time. When a prospective customer lands on the pricing page, Netlify delivers cached HTML directly from the nearest regional point of presence (Edge CDN) in Mumbai, Singapore, or Frankfurt:

// prerender.js - Pre-rendering public routes into static HTML assets
import { renderToString } from 'react-dom/server';
import fs from 'node:fs';

export async function preRenderPublicRoutes(routes, templateHtml) {
  for (const route of routes) {
    const appHtml = renderToString(<StaticRouter location={route}><App /></StaticRouter>);
    const finalHtml = templateHtml.replace('<!--app-root-->', appHtml);
    fs.writeFileSync(`./dist${route}/index.html`, finalHtml);
  }
}

This single change eliminated 3 always-on AWS ECS t3.medium instances that were running solely to serve HTML to web crawlers.

Phase 2: Database Modernization with Neon Serverless Postgres

The largest financial drain on their AWS bill was their provisioned RDS PostgreSQL database. Because the previous team feared connection exhaustion during marketing webinars, they had provisioned an expensive `db.m5.large` instance with 500 IOPS and multi-AZ replication.

We migrated the data to Neon Serverless PostgreSQL. Neon's architecture separates storage from compute:

• During quiet night hours (1 AM to 6 AM IST), compute automatically scales down to 0.25 Compute Units (CU), costing pennies.
• During peak operational hours, compute scales up elastically to 4 CUs without dropping transactions.
• Built-in connection pooling handles over 10,000 concurrent client connections via PgBouncer without requiring a separate AWS ElastiCache layer.

Phase 3: Restructuring the Django API Monolith

Rather than maintaining 4 separate Node.js repositories with inconsistent TypeScript versions, we consolidated business operations into a single modular Django REST Framework (DRF) service:

1. Automated Schema & Migrations: Django's built-in migration engine eliminated manual schema drifting issues that plagued their Node services.
2. Security Hardening: Enforced Turnstile token validation and strict rate-limiting on sensitive endpoints to protect against brute force and API scraping.
3. Webhooks for External Integrations: Background workers asynchronously push customer lead data into HubSpot and Slack in 180ms.

The Results: 60 Days Post-Deployment

Two months after cutover, the metrics spoke for themselves:

• Annual Infrastructure Savings: $34,440 freed up directly into their product growth budget.
• Organic Search Impressions: Rose by 64% within 45 days due to perfect Core Web Vitals (sub-500ms TTFB and 0ms CLS).
• Demo Booking Completion Rate: Rose from 2.1% to 3.8% because the qualification wizard loaded instantaneously on mobile browsers.