There’s something I’d never really considered: Now that Meta, Google, ChatGPT, etc. are starting to roll out AI Ads, what are the chances advertisers end up paying to reach users who would’ve bought the service anyway – without an ad? I started thinking about this after reading this study on the “ghost attribution” trap. It looked at how user targeting works on Retail Media Networks (RMNs) like Amazon Ads, Walmart Connect, and Instacart Ads – and found that the models were biased. Automated RMN bidding models naturally target users who are already highly likely to buy a product (e.g., people who’ve bought the brand 5 times this month). Then the AI dashboard shows a massive Return on Ad Spend (ROAS) and advertisers keep spending. But randomized controlled trials showed that up to 70% of these sales were non-incremental. The consumer would’ve bought the item anyway. The AI was essentially taking credit for organic behavior, masking a massive waste of ad budget. Scary.