Grocery and pet discovery for the weekly basket

Grocery and pet search must connect shopping intent with the correct product, pack and fulfilment context. Evaluate exact brands and descriptive queries alongside availability, required attributes and repeat-shopping needs. Plan substitutions and promotions as explicit decisions so broader discovery does not hide a mismatch in the basket.

## Keep the store context explicit Define how the storefront identifies the selected store, fulfilment method and shopping time. Follow that context through search, recommendations and product detail pages. A catalogue item being available somewhere does not establish availability for this order.

## Separate preferences from verified constraints Maintain structured dietary, ingredient and allergen information from an accountable source. Use those attributes to express constraints and preserve unknown values. A semantic match to a phrase such as “dairy-free” is not a substitute for verified product data.

## Plan unavailable-product behaviour Decide when to suppress an item, explain its status or offer an eligible substitute. Test pack sizes, unit pricing and variant distinctions so a substitute does not silently change the shopper’s requirements.

## Bring these examples to an evaluation Include a store change during a session, a stale stock record, an ambiguous dietary term and a product with incomplete attributes. Judge the result against the selected store and the documented product facts.

Match the product the shopper actually means

Build a query set around the work of filling a basket: a familiar brand, a generic staple, a particular pack size, an ingredient or a product with specified attributes. Include ambiguous terms and combinations that are easy to misunderstand.

Separate brand preference from a hard request, and distinguish product type from related accessories or complementary products. Pet ranges add their own structured attributes, such as species, stated life stage and pack weight; verify these against authoritative product information rather than inferring them from an image.

Make fulfilment context part of acceptance

The right item can still be unusable if it is unavailable for the selected location or fulfilment option. Map which system supplies product availability, how current it is and how unknown or delayed values should behave.

PA’s Products API includes a product stock flag. That alone does not establish a store-specific inventory contract. Agree the retailer’s location and offer model, then test that search results, product pages and basket checks remain consistent.

Treat substitutions as a separate choice

Keep required product attributes and customer selections visible when offering alternatives. Explain a change in brand, quantity or pack size. If the information needed to establish suitability is missing, do not silently treat a similar-looking item as equivalent.

For a hypothetical search for a particular one-litre oat drink, a related two-litre product may be a useful alternative, but it should be labelled with its actual size and price. A lower zero-result rate is not a success if shoppers repeatedly remove the suggested products.

Review relevance and commercial rules together

Check promotions and private-label preferences against the same query panel used for organic results. Retain clear eligibility boundaries for sponsored placements. Where repeat-shopping features are in scope, evaluate how prior purchases interact with explicit current intent rather than assuming a historical preference always wins.

In PA’s published grocery study, a 705-query offline comparison reported recall@120 increasing from 0.485 to 0.800. That result describes the evaluated catalogue and judgment method; it is not a sales uplift or a guarantee for another retailer.

Start with a representative basket mission

Bring real query examples, pack and attribute definitions, and fulfilment requirements to the review. Judge search relevance, valid substitutions and product availability separately, then measure the agreed live outcome across the complete shopping journey.

Frequently asked questions

Does semantic similarity establish dietary suitability?

No. Use verified product information and explicit requirements. Missing or ambiguous evidence should remain unresolved; a semantically related product must not be presented as satisfying a mandatory attribute without support.

Does a product stock flag prove local availability?

No. A location-specific assortment or fulfilment model needs an agreed mapping and freshness contract. Test the selected store or fulfilment context through search, product page and basket.

Should private-label products always rank first?

Define the commercial preference and its scope, then test it against the shopper’s stated requirements. Inspect exact-brand and generic queries separately so a promotion does not conceal a mismatch.