Pearch probes cross-retailer outcome infrastructure

This lands on a real gap: most retail ML stacks optimize on clicks and sessions because the true preference signal only shows up after checkout. If Pearch can normalize messy catalog data and make email-derived purchase memory trustworthy, it could become useful infrastructure for shopping personalization.

• –Most recommender research still leans on single-platform logs, ratings, or session events; cross-domain recommendation papers exist, but neutral cross-retailer outcome data remains unusually scarce.
• –Email receipts are a smart ingestion wedge because they avoid one-off retailer integrations, but they create hard problems around identity resolution, SKU matching, missing return states, and privacy.
• –The durable moat here is the ontology layer: consistently mapping heterogeneous products and labeling outcomes like keep, return, replacement, and repurchase across hundreds of merchants.
• –If the data gets good enough, the payoff is bigger than recommendations alone: return prediction, sizing intelligence, repurchase timing, and shopping agents all benefit from longitudinal outcome memory.