United StatesRetail and e-commerce2024 – ongoingThe RealReal
Machine-learning triage and scoring of items for counterfeit risk
The RealReal uses two machine-learning models to route risky items to expert authenticators and score photos for fakes
Problem
The luxury resale site receives about 20,000 items a day, and each one must be checked for authenticity. Human authenticators had to review everything, with no automatic way to tell which items carried the most risk.
Approach
The in-house engineering team built TRR Shield, a machine-learning model that decides which items go first to human authenticators and sends the riskiest ones to the most experienced staff. It also built TRR Vision, which examines item photos and predicts the likelihood that an item is counterfeit. Vision is used on 50% of handbags uploaded, per the company. The tools work alongside the existing authenticators and copywriters.
Results
Takeaway
The models rank and score items, and people still make the final authenticity call. Risk scoring let the company direct scarce expert time to the items most likely to be fake.
Categories
- Industry
- Retail and e-commerce
- Business function
- Production and quality
- Innovation type
- Process innovation
- AI technology
- Predictive analytics and MLComputer vision
- Adoption
- Built in-house
- Evidence
- Company's own claim
Sources
Source summaries
Business Insider describes how The RealReal uses TRR Shield to prioritize items for human authenticators and TRR Vision to predict counterfeit likelihood from photos. It gives the company's figures: Vision on 50% of handbags, about 20,000 daily uploads and more than 200,000 counterfeits found since 2011. The accuracy of the models is not reported.
Full details are in the original sources.