Paired document-to-structured-record corpus plus content performance data. Brandfuel ingests vendor product files in Excel, CSV, Word, PowerPoint, PDF and proprietary retailer templates, auto-detects structure, maps fields, associates images, translates, then generates and scores publish-ready product content. The latent asset is threefold: (1) raw messy vendor source documents paired with their normalized structured product records - a document-understanding and schema-mapping training set; (2) the accumulated reusable field-mapping graph across vendor formats and retailer data frameworks; (3) AI conversion-optimization and testing results, i.e. generated content variants bound to measured conversion outcomes - preference/reward data for content generation.
Equities in US. Related tickers AMZN, WMT, TGT, HD, LOW, BBY, M, JWN, DG, ORLY.
Multimodal: source side is heterogeneous documents (XLSX, CSV, DOCX, PPTX, PDF, images, multilingual spreadsheets); target side is structured JSON product records with attribute schemas, normalized marketing copy, SEO fields and associated product imagery; plus tabular content-test results (variant x traffic x conversion).
Currently modest and rising: a 6-person vendor serving high-SKU retailers plausibly has normalized on the order of 0.5-5M product records to date, each with a retained source document, mapping decision and generated content. The conversion-testing subset is smaller - thousands to low tens of thousands of variant-to-outcome pairs - but is the scarcest data of the three.
Free labels from the mechanic: source-span-to-attribute alignment labels (which region of which vendor file maps to which schema field) - genuine layout-understanding ground truth; image-to-product association labels; missing-field detection and gap-fill pairs (sparse input to complete record); translation pairs from multilingual vendor sheets; content quality scores per record against retailer standards; approval versus rejection decisions in the review layer with reasons; and the strongest one - content variant to conversion-rate outcome pairs from the optimization and testing loop.
Brandfuel — Paired document-to-structured-record corpus plus content performance data. Brandfuel ingests vendor product files in Excel, CSV, Word, PowerPoint, PDF and proprietary retailer templates, auto-detects structure, maps fields, associates images, translates, then generates and scores publish-ready product content. The latent asset is threefold: (1) raw messy vendor source documents paired with their normalized structured product records - a document-understanding and schema-mapping training set; (2) the accumulated reusable field-mapping graph across vendor formats and retailer data f…
Not yet on file: Sample · Data dictionary · Data frequency · Reporting lag · Point-in-time · License