- 6country sites live
- 12×peak sale capacity
- -70%manual order handling
- 180msavg. overseas first paint
Background
The client had been running on a third-party SaaS storefront. As SKUs and markets multiplied, three problems became unavoidable: promotions couldn’t be customized, orders reached the ERP by manual spreadsheet export, and platform fees and traffic rules were outside their control. They decided to build their own.
System shape
Storefront Catalogue · cart · checkout · member centre (multi-language / multi-currency)
│
Middle office Product · order · inventory · promotion · member · settlement
│
Integration Payment gateways · carrier APIs · overseas warehouse WMS · ERP · tax · support desk
Key design decisions
Multi-site, multi-language
A “single codebase plus site configuration” model: the site determines language, currency, tax rules, payment methods, sellable SKUs and logistics options. Launching a new country is configuration, not code or a separate deployment.
- Copy and product content are separated: translations live in their own i18n content tables and can be overridden per site
- Pricing: base price plus per-site FX policy plus tax (with inclusive/exclusive display switching) plus promotion stacking rules
- Per-site SEO: domain or subdirectory strategy, hreflang, structured data, per-site sitemaps
Payments and risk
Stripe, PayPal and local payment methods sit behind a unified payment router that selects a channel by site, amount and card scheme, with fallback on failure. Risk controls cover device fingerprinting, order rate limiting, blocklists and 3DS trigger rules, keeping chargebacks below the industry average.
Automated fulfilment
After an order is created the system automatically reserves inventory, runs risk review, splits shipments across warehouses and parcels, assigns overseas warehouse or direct mail, generates labels, ingests tracking updates and syncs to the ERP. Exceptions land in a human workbench; manual handling fell about 70%.
Peak-sale stability
- Product and listing pages static and CDN-served, with hot data cached at several levels
- Inventory decremented in Redis first and persisted asynchronously via queue, guaranteeing no oversell
- Full-path load testing with capacity planned at 1.5× the forecast peak, plus a throttle-and-degrade rehearsal a week before
- Result: peak traffic reached 12× normal with zero incidents
Outcomes
- Six country sites live, with operations able to launch new ones unaided
- Platform commissions eliminated, materially improving the long-run cost structure
- Fulfilment automated, reducing support and operations headcount pressure
- 180ms average first paint overseas (CDN, edge caching, adaptive images)
Related
Before committing, we ran a round of technology consulting comparing three-year total cost of “SaaS plus plugins” against building in-house. Delivery followed our software development model.
Once the platform had been running for a while the bottleneck moved from technology to operations, and we layered three AI scenarios on top — see AI Operations Copilot for Cross-Border Commerce.