Web & Conversion Optimization Testing — PacSun

01

From tests to a practice

Going into 2020, the problems were measurable and known: weak movement from the homepage toward checkout, high mobile homepage bounce, and experience work increasingly organized around conversion. Testing was becoming part of roadmaps, working schedules, and a dedicated Fall 2020 strategy.

I did not think of that work as a pile of isolated website tweaks. As Senior Manager, eCommerce, I created and owned the testing and optimization strategy end-to-end. We were trying to increase conversion, reduce bounce, and make each next change answerable to what the previous one had shown.

I made the practice more deliberate by forming hypotheses, placing them in a sequenced backlog, configuring the experience or getting it built, launching when it was ready, measuring what happened, and carrying a recommendation into the next decision. I prioritized the backlog by audience size, potential KPI lift, and operational effort. The important shift was that another worthwhile test was always expected to enter the cadence.

02

Recommendations as the starting point

Product-page recommendations were effectively the starting point for the more aggressive practice. They gave me a contained place to ask merchandising questions that could be expressed as competing experiences: cross-sell against up-sell, one category or brand strategy against another, and different product mixes against each other.

The 2020 planning material names cross-sell versus up-sell denim, branded-affinity recommendations, product-page algorithm and widget variations, header testing, and reporting by gender, strategy, brand, and category. Some items were marked launching or ongoing, and one widget test was described as live. Those labels tell me what was in flight, not which version ultimately won.

Dynamic Yield made much of this directly executable without a full development cycle. I shaped the hypothesis, configured many of the experiences, decided when they were ready to launch, read the results, and made the recommendation that followed.

03

Onto the homepage

Once recommendation testing showed enough value to keep pushing, I expanded the practice into more visible homepage real estate. That made the questions more exposed: the homepage carried more traffic, more stakeholders, and a larger share of the customer's first impression.

The program expanded to a hero A/B test, a shortened-homepage test, a test that hid homepage recommendations, recommendation-audience updates and targeting, and homepage auto-slider speed. Some were plans or launching work, so I do not treat their status as a completed result.

The change in visibility also raised the implementation bar. I shipped much of the work directly through Dynamic Yield. When an experience went beyond what I could configure there, especially larger or site-wide changes, PacSun development and other partners handled the implementation. The optimization strategy and the decision loop stayed with me.

04

“King of the Hill Tests”

Sort order on category listing pages gave the practice its clearest repeatable form. The roadmaps contemplated strategies such as Most Popular, Most Liked, and Personalized, and a working schedule lists PLP Sort A/B Tests as launching. That status says the tests were starting, not which sort won.

I called the method “King of the Hill Tests.” The current sort remained the control while a challenger ran against it. If the challenger lost, the control stayed. If a better challenger won, it became the new benchmark and the next useful hypothesis had to beat it.

We could keep going until the ideas worth testing were exhausted. That mattered more than finding one permanent answer. The control was not sacred, and a winner was not final; each one was only the strongest supported choice we had at that point.

05

Losing tests belonged

Plenty of challengers lost, and no single loss was the story.

A negative result was ordinary evidence. It removed one answer, exposed a weak assumption, or made the next hypothesis more precise. A losing challenger could still improve the next test even when nothing from that version deserved to replace the control.

The mandate was to keep testing. A loss did not break the program or leave the cadence empty; it became part of what we knew when we decided what to put into the next week or month.

06

What the practice became

As I remember it, Dynamic Yield was already in use at PacSun by about 2017–2018. Experimentation did not begin with the approximately 2020–2021 replatform, and the replatform did not introduce the platform. Afterward, the modernized storefront gave the more aggressive testing practice a stronger foundation and helped it accelerate.

Over time, the same habits pointed toward deeper personalization. Recommendation strategies and audiences had made relevance testable; the broader practice made it normal to target an experience, measure the response, and revise what came next.

Testing also changed how I handled some stakeholder disagreements. When competing opinions could be turned into testable propositions, customer behavior could add evidence and give the conversation a constructive off-ramp from subjective conflict. Not every disagreement was testable, and measurement did not remove judgment; it helped move the conversation from whose preference should win toward what we had learned, what remained uncertain, and what deserved the next slot.

That is the line I draw from this period. It is not a complete history of PacSun personalization or a claim that testing alone caused it. But the testing practice helped lay the groundwork for a more personalized commerce experience, because it had already changed how I approached experience design, merchandising, measurement, iteration, and disagreement.