Shipped software, with real proof.

The hard part

The hard part was the count, not any one piece. Five surfaces share one domain model, so a change to how an order works is a change in five places, five release cycles and five sets of edge cases — and the whole thing only functions when all five agree. We scoped this as an app with some panels attached, and it is not: it is five products that happen to ship together. Anything similar we now scope as five.

The hard part

Shipping 300+ tools created a discovery problem we underestimated — the index page became the product, and the first version of it was unusable. Search and categorisation had to be rebuilt after launch.

The hard part

Rate parity was the hard part, not the code. Undercutting the OTAs directly would have risked the resort's listings, so the direct channel had to win on inclusions and guarantees rather than on headline price. That was a commercial decision we had to work through with the owner before a line of the booking flow made sense.

The hard part

Report generation was slower than promised in the first build because the templates were doing layout work at request time. Moving generation to a queued job fixed it, but it meant reworking how counsellors were notified — a change we should have anticipated during scoping and did not.

The hard part

The site renders its content on the client, which search engines handle unevenly. It works for patients but it holds back the ranking this rebuild was meant to win, and moving it to server-rendered pages is the honest next step — we would scope that differently today.