eBay - The Science of Making Better Predictions
•
1 like•565 views
eBay joined NMPi's 2016 Digital Seminar on "Transforming Data-Driven Marketing." Everything happens for a reason; but on some occasions that reason is clearer than others. Phuong Nguyen explores how data can reveal the logic behind apparently serendipitous events, drawing on insights from eBay’s 19 million UK monthly users to explain seemingly unexplainable trends in consumer behaviour.
Report
Share
eBay - The Science of Making Better Predictions
- 1. The Science of Making Better Predictions Phuong Nguyen Director of Advertising
- 2. eBay’s credentials • One of the world’s biggest repositories of data • One of the UK’s biggest media publishers: 19 million unique users per month. 6 billion ad impressions per month. 41% reach of UK adult population. Source: eBay internal data 2015.
- 3. 20 minutes on ebay.co.uk By the time I finish, the following will have been bought on eBay: 12pushchairs 100pairs of jeans 10TVs 28iPads/tablets 20cars Source: eBay internal data 2015.
- 4. Data drives insights on eBay Our data isn’t limited only to purchases. Shopper insights This forms an enviable list of observed not inferred shopping behaviours few if any can match. Buy Bid Search View
- 8. 3 things we couldn’t have predicted
- 9. 3 things we couldn’t have predicted
- 10. 3 things we couldn’t have predicted
- 11. How can we deliver better predictions?
- 12. 1. Validate efficacy of data Inferred vs observed
- 13. 2. Understand the impact of data decay
- 14. Understand the impact of data decay Example: Length of shopping windows 50,000,000 transactions on eBay UK every month DaysMonths versus Source: eBay internal data 2015.
- 15. Consideration window varies markedly between products and categories eBay Advertising Christmas Tracker is based on search data from ebay.co.uk in 2014. Peak windows of influence correspond to protracted peaks for searches within each category. Home Furniture & DIY Peak window of influence: 2nd November - 6th December CATEGORIES THE COUNTDOWNTO CHRISTMAS September October November December September October November December Christmas Day Cameras & Photography Peak window of influence: 2nd November - 13th December Clothes, Shoes & Accessories Peak window of influence: 12th October- 6th December Computers,Tablets & Networking Peak window of influence: 2nd November - 6th December DVDs, Films & TV Peak window of influence: 16th November - 24th December Mobile Phones & Communication Peak window of influence: 21st September - 15th November Sound & Vision Peak window of influence: 9th November - 6th December Toys & Games Peak window of influence: 5th October - 6th December Video Games & Consoles Peak window of influence: 16th November - 13th December
- 16. 3. Use quality data to explore ‘hypothesis’ driven trends
- 17. Use quality data to explore ‘hypothesis’ driven trends Example: SanDisk SanDisk wanted to sell more memory products but avoid wastage. eBay specifically targeted only those shoppers who: • Were currently in-market for memory products • Had recently bought a host device Results • Sales tripled • Net promoter scores increased across all metrics studied • Broadcast level of audience reached (2.2m unique individuals) but all qualified leads
- 18. 4. Dig deep into the data to reveal ‘serendipitous’ trends
- 19. Insight driven advertising solutions Example: The Co-operative Electrical Shoppers of bakeware are incredibly diverse. The Co-operative Electrical wanted to target them with a campaign for kitchen appliances. Bespoke home bakers segment targeted to coincide with the final of cult TV show The Great British Bake Off.
- 21. Predictive targeting case study: New Mums High Street branch of chemists Campaign performance (CTR) Baby category targeting: 0.05% Baby keyword targeting: 0.07% New Mums targeting: 0.11%
- 22. Summary 1. Validate efficacy of data (Inferred vs observed) 2. Understand the impact of data decay 3. Use quality data to explore ‘hypothesis’ driven trends 4. Dig deep into the data to reveal ‘serendipitous’ trends 5. Productise your predictions
- 23. Let’s hope predictive analytics can help us all make the right choice in the future
- 24. Thank you for your time @phuongster