MRR = mean of (1 / position of first relevant result) over a set of queries. If the first relevant product is at position 1, that query contributes 1.0; position 2 contributes 0.5; position 5 contributes 0.2. Queries with no relevant result in the cutoff contribute 0.
MRR is the right metric when you only care about the first hit — known-item search (looking up a specific SKU), navigational queries, or re-find scenarios. It’s the wrong metric when shoppers browse multiple results before deciding, where NDCG (which credits relevant items at all positions) is more honest.
In ecommerce, MRR pairs well with branded and SKU-specific queries; NDCG suits broader category and exploratory queries. Many teams track both segmented by query intent.