See all Press

EDITED’s Retail AI Knows Style Better than Fashion Pros

As the leader in retail data analytics, EDITED confirms its machine learning classifier for retail outperforms humans at every category...
As the leader in retail data analytics, EDITED confirms its machine learning classifier for retail outperforms humans at every category it tested against.  NEW YORK - September 5, 2018 - EDITED, the world’s biggest source of real-time data for brands and retailers, today revealed its machines are more reliable at analyzing garments, along with classifying footwear styles and footwear subcategories than apparel professionals.  Based on an internal classifier benchmarking study, EDITED’s classifiers outperform retail experts across a range of classification tasks. The machines made 2.5 fewer pp errors when identifying garment types, climbing to over 9 fewer pp errors when determining subcategories, and around 6.5 fewer pp errors when classifying specific footwear styles. Machines showcase more efficiency at performing redundant tasks such as identifying products, especially at retail’s current state when new styles are rolling out at a rapid pace.  “The difference between humans and machines is that we’re abstract thinkers--and you can’t replace that,” said Geoff Watts, Co-founder and CEO of EDITED. “Our Classifier Benchmarking is a prime example of places where machines should do the simple laborious tasks, so people can focus their time and energy into developing creative strategies.”   The Classifier Benchmarking was an online test conducted where respondents identified nearly 1,286 products randomly selected from EDITED’s database. Their answers along with the classifier predictions were compared to each product’s category in order to determine accuracy.  Additional findings include: 
  • Classifiers outperform humans on even trivial tasks of identifying garment types at 97.8% accuracy against 95.4% for humans, and 96.7% accuracy against 87.4% for humans on identifying footwear subcategories.
  • Classifiers and humans both struggled to accurately identify footwear styles at 69% to 63% proving the apparel industry’s complications around categorization. Overall, the footwear styles classifier was still better than humans despite a more conservative approach. 
  • Classifiers are able to classify all of the products in the sample dataset in a matter of seconds, while it took on average six and half minutes for a human respondent to classify 57 random products--the equivalent to nearly 2.5 hours for the whole dataset. 
Task fatigue is evident that human errors increase over time, whereas the machine learning model actually improves with time. In general, the reliability of machines to automate this process and classify large datasets is valuable in contributing to a more cost-effective internal retail strategy, especially when retailers currently dedicate tremendous time to manually track markets across regions and languages.   Since 2009, EDITED’s software uses artificial intelligence, sophisticated analytics and image and text recognition to understand pricing, discounts, assortments and trends in real-time. EDITED is built and updated in-house by data scientists and engineers working alongside former buyers and merchandisers from global brands and retailers.   About EDITED  EDITED helps the world’s best retailers drive sales by eliminating guesswork. The Retail Decision Platform uses A.I. to optimize buying and merchandising decisions, ensuring retailers get their product and prices right every time. This is how the world’s most innovative retailers stay ahead of the competition; including PUMA, Tommy Hilfiger, Diesel, Ann Taylor and the Arcadia Group. For more information, visit Media Contact Daphne Duong  EDITED  [email protected] Phone: +1 646-606-2790