FleaMail

The build

We built the replacement as a single Next.js 16 application on TypeScript, Prisma and MySQL: 55 data models, the public site, a customer account area, checkout, a 50-screen admin and a versioned API for a new Flutter mobile app. It was built from FleaMail's own design system — warm cream, coral for action only, Australian spelling throughout — rather than a template.

The data came across from a read-only mirror of the production WordPress database, through idempotent migration scripts that can be re-run at any time without creating duplicates. A reconciliation gate compares the two systems and exits with an error on any mismatch in counts or monthly recurring revenue; it is the thing that decides whether cutover can happen, not a person's confidence. It earned its place early: the first pass imported order headers without their line items, addresses and refunds, and the original gate only checked header counts, so it passed. We tightened the gate to check every relationship, backfilled the missing records, and it now reconciles green to the cent.

Billing continuity is structural rather than hopeful. The new platform uses the same Stripe and PayPal merchant accounts, so every stored customer and subscription reference carries over and nobody re-enters a card. Customers log in with their existing WordPress password, which is verified against the old hash and upgraded to bcrypt on first login. Webhook handling is idempotent, so a payment event delivered twice is processed once. And the whole platform runs in a shadow mode first: it receives the same payment events as the live site and applies them, but sends no email and charges nobody, while a nightly drift report compares it to WordPress subscription by subscription.

The treatment logic was carried across verbatim. The feline, canine and worming lookup tables are seeded from the business's own data and editable in the admin; the code is a pure lookup, and where a record has no code the platform shows none rather than generating one. Postcode decides the parasite tier, weight and species decide the product.

For search, we audited every URL that had earned traffic in the previous twelve months — 227 of them — and found that 118 would have returned a 404 on the new platform, almost all old WooCommerce product pages and WordPress archives. Each now has a 301 to its closest live equivalent, managed from a redirect table in the admin rather than hard-coded, alongside the migrated Yoast titles and descriptions, the 92 blog posts, structured data on every article and plan, and a 404 monitor for the weeks after launch.

The new platform is built and reconciling. Cutover follows the written runbook: a parallel run of at least four weeks with a clean drift report, a freeze on the old checkout, a final sync, the reconciliation gate, and only then the payment webhooks and DNS — with WordPress kept read-only for ninety days as a fallback.