8 weeks post-launch monitoring complete · CloudWatch dashboards & PagerDuty alerting live · Client engineering team trained on new architecture
Who we worked with
An Australian e-commerce retailer generating $14M in annual online revenue, running their entire platform on a 9-year-old PHP 5.6 monolith hosted on two aging on-premise Dell servers in a Sydney co-location facility.
Every Black Friday and mid-year sale, the site buckled under load — averaging 47 minutes of downtime per peak event and roughly $38K in lost orders per outage. The platform couldn't handle more than 1,200 concurrent users before response times crossed 8 seconds and the checkout flow timed out.
| Metric | Before | After | Change |
|---|---|---|---|
| Peak traffic capacity | 1,200 concurrent users | 15,000+ concurrent users | 12× increase |
| Downtime per peak event | ~47 minutes avg | < 2 minutes (auto-recovery) | 96% reduction |
| Average page load time | 6.4 seconds | 1.1 seconds | 83% faster |
| Order processing time | 14 seconds per transaction | 2.8 seconds per transaction | 80% reduction |
| Deployment frequency | Every 2–3 weeks (manual) | 3–4× per week (automated) | 8× more frequent |
| Quarterly revenue | $3.5M baseline | $4.7M (first quarter post-launch) | +34% increase |
Every sale was a business risk
A 9-year-old PHP monolith, two physical servers, and no auto-scaling meant every promotional event carried a real chance of crashing the site — and the business had accepted it as inevitable.
Repeated Peak-Season Crashes
The PHP monolith running on two Dell PowerEdge servers couldn't handle more than 1,200 concurrent sessions before MySQL connection pools exhausted and checkout began timing out — averaging 47 minutes of downtime per peak event at ~$38K per outage. A single pricing bug in March 2023 took 11 hours to roll back with no automated rollback mechanism in place.
47 min avg downtime · $38K per eventDeployments as a Full-Team Emergency
Every change — including a minor promotional banner update — required a full application deployment: SSH into production, manual migration scripts, and hoping nothing broke. Engineers spent roughly 30% of their time on deployment firefighting rather than building features.
2-day manual deploysNightly Inventory Sync Causing Overselling
Inventory synced from the warehouse system via a nightly batch job. During peak traffic, customers purchased items already out of stock — 180 oversold orders per month at $22/incident in labor and goodwill credits, totaling ~$47K/year in avoidable costs.
180 oversold orders/month · $47K/yearZero Personalization, Flat Average Order Value
Every customer saw the same homepage, same product order, and same promotions regardless of browsing or purchase history. Average order value had been flat at $67 for 3 consecutive years — the business had no data infrastructure to run anything beyond basic email blasts.
$67 AOV flat for 3 yearsFive phases, zero customer-facing downtime
A structured 24-week delivery — assessment through cutover — with a phased traffic migration (10% → 50% → 100% over 5 days) so customers never experienced a disruption.
Assessment
Wks 1–3Infrastructure & Data
Wks 3–8Microservices & Storefront
Wks 6–18AI Recommendations
Wks 14–20CI/CD & Cutover
Wks 18–24- Legacy Health Score™ assessment — scored 41/50 (Critical) across 5 dimensions; mapped all 23 modules in the PHP monolith and designed target-state 12-microservice architecture
- Stood up AWS environment: VPC, ECS Fargate clusters, RDS PostgreSQL (Multi-AZ), ElastiCache Redis, S3 + CloudFront CDN
- Migrated 1.8M product SKUs, 2.3M customer records, and 4.1M orders (6 years of history) from MySQL to PostgreSQL with zero data loss
- Ran legacy site and new infrastructure in parallel for 2 weeks to validate complete data integrity before proceeding
- Built 12 microservices in Node.js (TypeScript) — each with its own database schema, API contracts, and independent deployment pipeline
- Rebuilt storefront in React + Next.js with server-side rendering — page load from 6.4s to 1.1s; Mobile Lighthouse score 31 → 87
- Replaced legacy PayPal redirect with Stripe — eliminated 4.2% checkout abandonment rate caused by redirect-based flow
- Replaced nightly inventory batch with real-time event-driven sync via Amazon SNS/SQS — overselling eliminated entirely
- Deployed Amazon Personalize trained on 18 months of purchase and browsing data from 340K+ customer profiles — A/B tested for 4 weeks
- Built full CI/CD: GitHub Actions → Docker → ECR → ECS blue-green deployment with automatic health checks and rollback on failure
- Ran 3 weeks of load testing simulating up to 20,000 concurrent users before cutover
- Phased traffic migration — 10% → 50% → 100% over 5 days with zero customer-facing downtime
- Deployments went from 2-day manual process to 22-minute automated push with automatic rollback on failure
- Set up CloudWatch dashboards for real-time performance visibility across all 12 microservices
- PagerDuty alerting with automated escalation and cost monitoring dashboards
- Trained the client's 3-person engineering team on the new architecture, deployment process, and how to extend each microservice independently
- Provided architectural documentation and runbooks for every service and operational procedure
Managed — no ML team required client-side
340K+ customer profiles
AOV: $67 → $75 (+12%)
Every choice made for e-commerce at scale
| Technology | Role | Why This Choice |
|---|---|---|
| AWS (ECS Fargate, RDS, S3, CloudFront) | Cloud infrastructure | Auto-scaling without managing servers; Fargate eliminates container host management |
| Node.js (TypeScript) | Microservices backend | Fast async I/O for e-commerce workloads; TypeScript for type safety across 12 services |
| React + Next.js | Storefront frontend | Server-side rendering for SEO and speed; Lighthouse 31 → 87 |
| PostgreSQL (RDS Multi-AZ) | Primary database | ACID compliance for transactions; Multi-AZ for high availability |
| Redis (ElastiCache) | Caching layer | Sub-millisecond reads for sessions, catalog cache, and cart state |
| Amazon Personalize | AI recommendation engine | Managed ML; no in-house ML team required on the client side |
| Stripe | Payment processing | Lower abandonment vs. redirect-based PayPal (eliminated 4.2% drop-off) |
| Amazon SNS / SQS | Event-driven messaging | Real-time inventory sync; decoupled inter-service communication |
| GitHub Actions + Docker + ECR | CI/CD pipeline | Blue-green deployments with automatic rollback in 22 minutes |
| CloudWatch + PagerDuty | Monitoring & alerting | Real-time dashboards and automated incident escalation |
Results that transformed the business
Across performance, revenue, reliability, and engineering velocity — all delivered in 24 weeks with zero customer-facing downtime during migration.
Capacity
First Quarter
Prevented (2 Seasons)
on Overselling
6 engineers, one cohesive delivery
A compact, full-stack team covering cloud architecture, backend services, modern frontend, AI/ML, DevOps, and quality — with post-launch client training built into the engagement.
Solution Architect
Legacy Health Score™ assessment, 12-service architecture design, AWS infrastructure planning
Backend Developer × 2
Node.js (TypeScript) microservices, Stripe integration, SNS/SQS event-driven inventory sync
Frontend Developer
React + Next.js storefront, SSR implementation, mobile performance optimisation
DevOps Engineer
ECS Fargate, Terraform, GitHub Actions CI/CD, blue-green deployments, CloudWatch + PagerDuty
QA Engineer
Load testing (20K concurrent users), end-to-end checkout validation, phased cutover monitoring
24-Week Full Delivery
Zero customer-facing downtime · Phased 5-day traffic migration · Client engineering team fully trained
What the client said
"Our old platform was costing us real money every time we ran a sale — and we just accepted it as normal. I-Verve's team showed us exactly what was broken, rebuilt it in a way our small engineering team can actually maintain, and the AI recommendations are generating revenue we never had before. The first Black Friday on the new platform was the first one where I didn't get a 2 AM call about the site being down."
Let’s bring your idea to life
Your innovative idea deserves a team that can bring it to life. Reach out to us today to discuss your project, and we’ll work with you every step of the way.