How Layout Testing Works in Eevy AI

When you create a section, Eevy AI automatically creates an initial population of layout variants based on your selected section type. For example, if you choose "Review Carousel," the algorithm creates variants with different numbers of visible cards, arrow styles, auto-advance speeds, and card designs. You can influence the initial population by selecting which layout types to include: just carousels, or carousels plus sliders, or all layout types. More layout types in the pool means more diversity for the algorithm to explore, but also more traffic needed to converge on a winner.

Eevy AI uses a multi-armed bandit approach combined with the genetic algorithm for traffic distribution. In early generations, traffic is distributed relatively evenly across variants. As certain variants prove themselves, they receive more traffic (exploitation), while a portion of traffic is always reserved for exploring new variants (exploration). This balance means your store always shows a reasonably good layout to most visitors while still testing new possibilities.

A generation ends when the algorithm has collected enough data to make statistically confident decisions about variant performance. The algorithm then selects top performers, combines their traits (crossover), introduces random changes (mutation), and creates the next generation. You will see a notification in your dashboard when a generation completes. The generation summary shows which variants survived, which were eliminated, and what new variants were created.

Each variant's performance card shows: RPV — revenue per visitor for sessions that saw this variant; Sessions — number of unique sessions; Confidence — how statistically reliable the RPV number is; Configuration — the exact layout parameters (layout type, card count, sort order, etc.). Sort by RPV to see the top performers. But pay attention to confidence — a variant with $2.50 RPV and 50 sessions is less reliable than one with $2.30 RPV and 5,000 sessions.

Eevy AI can test layouts at two levels: Store-wide — the same optimization applies to all product pages. Better for stores with lower traffic where data needs to be pooled. Per-collection — different collections get different optimizations. Better for stores with diverse product types where one layout does not fit all. Configure this under Sections > Testing Scope. Most stores should start with store-wide testing and move to per-collection once they have enough traffic (1,000+ daily sessions).