Sure, but in that case, they also have information about whether I've returned the TV or not. If I haven't, it doesn't make nearly as much sense to keep recommending me TVs. Even if your ML algorithm thinks I'm going to return it, it still seems like you'd do better on a percentage basis to keep showing me sound bars, HDMI cables, and other accessories. After all, I bought a TV because I intended to buy a TV. That means I'm probably going to need some of those things to go with it, likely with far greater than 10-15% probability.
I suppose there's a possibility that your ML algorithm could decide that the profit from selling a TV so greatly exceeds that of the accessories that a 10-15% probability of preserving the sale in case of a return might be more worth it, but I don't really know the ins and outs of retail ecommerce to make a judgement there. It still just seems silly to me. ¯\_(ツ)_/¯
I agree, intuitively that makes much more sense. At a minimum, you should get some portion of accessories vs replacement items in your recommendations.
I'd be blown away with surprise to learn that AMZN hasn't optimized for revenue in the calculation though, so it seems likely that we're wrong.
Of course we'd prefer AMZN to optimize for customer happiness/benefit instead. To me, the "smart" approach would be to recommend accessories until and unless the customer shows some signal indicating that they are looking for another TV. Then switch to full replacement strategy (or even return-minimization strategy) mode.
That or similar logical approach may have been abandoned after getting better results (by some metric) from the ML-optimized approach. Assuming there's a human involved at all.
I suppose there's a possibility that your ML algorithm could decide that the profit from selling a TV so greatly exceeds that of the accessories that a 10-15% probability of preserving the sale in case of a return might be more worth it, but I don't really know the ins and outs of retail ecommerce to make a judgement there. It still just seems silly to me. ¯\_(ツ)_/¯