DenmarkRetail and e-commerce2026 – ongoingBestseller
Return predictions for advertising bids
Bestseller reported 50% higher Dutch advertising returns after factoring predicted returns into bids
Problem
Advertising systems valued orders before accounting for returned clothing. Excluding frequent returners was ineffective for new or anonymous shoppers.
Approach
Bestseller and Refyne built a model using Vertex AI and transaction data. It scored orders immediately after purchase and sent Google Ads a value adjusted for predicted returns. The company tested the approach in the Netherlands and Sweden.
Results
Takeaway
The model changed the sales value used for bidding. The reported advertising gains did not establish a reduction in customers' return rates.
Categories
- Size
- Large enterprise
- Industry
- Retail and e-commerce
- Business function
- Marketing and sales
- Innovation type
- Process innovation
- AI technology
- Predictive analytics and ML
- Adoption
- Partner co-development
- Evidence
- Company's own claim
Sources
Source summaries
Bestseller's digital and media manager described the return-prediction system. He reported lower acquisition costs and higher Dutch advertising returns after predicted returns were incorporated into conversion values.
Full details are in the original sources.
