I think what’s important is to understand that these things work because they are at a certain scale. Algorithms are notoriously bad at predicting individual behaviour, hence why recommendation engines are a specialization that is far from solved. But when you have large amounts of traffic, the law of large numbers allows you to predict group behaviour with some accuracy.
So you can’t follow a user around and predict their next move and show them the right ad at the right time. But you can take 50 000 middle-aged males, and bet that at least 10 of them will buy a motorbike if you randomly show them a picture of a guy riding in the sunset. Once you have a good volume of this kind of data you can do some casino math to tilt all your bets slightly in your favour, and start betting 24/7.
It’s really cold reading, like they do in those mentalist shows. It’s a lot dumber than it looks, but it’s way more effective than you think.
I think what’s important is to understand that these things work because they are at a certain scale. Algorithms are notoriously bad at predicting individual behaviour, hence why recommendation engines are a specialization that is far from solved. But when you have large amounts of traffic, the law of large numbers allows you to predict group behaviour with some accuracy.
So you can’t follow a user around and predict their next move and show them the right ad at the right time. But you can take 50 000 middle-aged males, and bet that at least 10 of them will buy a motorbike if you randomly show them a picture of a guy riding in the sunset. Once you have a good volume of this kind of data you can do some casino math to tilt all your bets slightly in your favour, and start betting 24/7.
It’s really cold reading, like they do in those mentalist shows. It’s a lot dumber than it looks, but it’s way more effective than you think.