Recommendations
A vs B has a built-in recommendation engine. It reads your Live Catalog (the single, always-fresh copy of your products), plus your order and product-view history. Each night it ranks your products. It saves the results as a dataset it manages for you. You do not need to pay for a separate service, run a separate data pipeline, or set up a catalog per recipe.
Each rule is called a recipe. You pick an algorithm, set a few options, and turn it on. A vs B does the rest. It reads your data, ranks your products, and keeps the results current. Then it serves them from the edge: a server near your visitor, so cards load fast.
Because recipes read the Live Catalog, two things stay true without any extra work from you:
- Out-of-stock products are never recommended. The engine checks stock live, right when it serves a card. A product that sells out stops appearing within about a minute. No rebuild needed.
- Prices on these cards are always current. Price, compare-at price, and stock come from the Live Catalog at serve time, not from last night's snapshot.
Want to know which recipe actually makes you more money? Run an ordinary experiment. Give each variation (each version you're testing) a different recipe to call through the recs API. Add a holdout (a variation with no recipe at all) too, so you can measure what recommendations are worth. Then compare revenue per visitor across variations to find the winner.