AI Auto-optimize
Ideal for maximizing recommendation performance without manually selecting a fixed strategy. AI Auto-optimize leverages a multi-armed bandit approach (Thompson Sampling) to intelligently explore various logics and allocate traffic to those currently yielding the highest engagement.
How it works
- Monitors the click-through performance of all candidate recommendation logics in real-time.
- Employs Thompson Sampling to optimally balance exploring new logic possibilities and exploiting the currently highest-performing strategies.
- Continuously adapts to seasonal trends, campaign changes, and catalog updates, eliminating the need for static A/B test resets.
Best placements
Storefront-wide pages
Homepages, collection pages, search results, and blogs—anywhere you'd typically hard-code a single recommendation logic.
Product Detail Pages (PDPs)
Allows RecoBoost to dynamically switch between product-context logics such as collaborative, similar items, or Complete the Look, based on performance.
Cart page
Especially useful when uncertain which logic—rankings, cart recovery, or cart cross-sell—will perform best on the cart page.
Things to know
- Sufficient impression volume is required for effective learning. Initial traffic may appear 'noisy' during the exploration phase.
- Standard widget settings for filters, product pinning, and stock priority remain fully configurable.
- The dashboard provides detailed breakdowns by logic, allowing you to observe which strategies AI is currently prioritizing.