She Said Yes

The situation

The client had spent real money on a launch campaign before they came to us, and the storefront underneath it could not take the weight. Product pages took several seconds to become interactive on a mid-range phone, which is what most of their audience was using.

The immediate risk was not a slow site. It was a slow site on the one day of the year when every visitor arrives at once and none of them come back if it fails.

What we found

Product discovery was the first problem. The catalogue was flat, so a shopper looking for one occasion had to scroll past everything else. The filters that existed filtered on fields the merchandising team did not actually use.

The second was the checkout. Shipping was calculated on a separate step that reloaded the page, and that step was where the analytics showed people leaving.

What we did

We rebuilt the storefront around how the brand actually merchandises: collections became landing pages with their own copy and imagery rather than automatic category listings, so a campaign could point at something purpose-built.

Image delivery was moved to responsive sizes served from a CDN, which is where most of the page-weight problem lived. Filters were rebuilt on the attributes the team tags products with in practice.

The checkout lost a step. Shipping now calculates in place, and the form asks only for what the warehouse needs to pick and dispatch an order.

How it went

The grand-opening campaign ran without a performance incident. Product pages render in under a second on the connection profile their customers actually have, and the checkout holds people through the shipping step that used to lose them.

What we would do differently

We load-tested the storefront but not the payment gateway, on the assumption that a third party would handle its own scale. It did, and we would still not make that assumption again — the one part of the flow we could not fix on the day was the part we had not tested.