The thing is, unless there’s a great breakthrough, individual behavior is dramatically less predictable than mob behavior. The Target algorithm was a great case study, but they sent those ads/coupons to EVERYONE who fit the algorithm, and much of the time they were EITHER right OR wrong. Target dropped that particular style of campaign because it had too low a match rate (hitting too many non-pregnant folks and missing too many pregnant ones). When it had that “shocking success”, was it truly a great moment for predictive AI’s, or just the chips falling right with the AI simply adjusting odds a bit?
But no, the predictive technology is harder; I say that as someone who has worked in a predictive data science division. The tuning required to make a model work better than control is hard. Ultimately, if you get ads for something you’ve never searched for before the same day you SAY you want that something, it’s the voice recognition. I’ll be clear, most models we’d work on would fail to prove themselves, get thrown out, and be picked up again. If we’d had the abililty to buy voice matches for the word “insurance” from Amazon/Google, we have been in bloody paradise.
“Oh, you wanted me to make a soda fizz, right? We should get St. Germain because that’s good in it.” … starts getting St. Germain commercials every 5 minutes. Didn’t even know they had commercials. Fucking St. Germain.
EDIT: And I have tested it. Came up with stupid things to say in front of speakers a few times, and some have definitely shown up in really bizarre TV ads. I get it, I know why Chewy is advertising to me. But what about the Plant Based Burger commercials I got in a sudden storm for a month? I’m a meat-lover, but decided to talk about plant-based burgers in my living room where the speaker is.
Oh I followed that one, too.
The thing is, unless there’s a great breakthrough, individual behavior is dramatically less predictable than mob behavior. The Target algorithm was a great case study, but they sent those ads/coupons to EVERYONE who fit the algorithm, and much of the time they were EITHER right OR wrong. Target dropped that particular style of campaign because it had too low a match rate (hitting too many non-pregnant folks and missing too many pregnant ones). When it had that “shocking success”, was it truly a great moment for predictive AI’s, or just the chips falling right with the AI simply adjusting odds a bit?
But no, the predictive technology is harder; I say that as someone who has worked in a predictive data science division. The tuning required to make a model work better than control is hard. Ultimately, if you get ads for something you’ve never searched for before the same day you SAY you want that something, it’s the voice recognition. I’ll be clear, most models we’d work on would fail to prove themselves, get thrown out, and be picked up again. If we’d had the abililty to buy voice matches for the word “insurance” from Amazon/Google, we have been in bloody paradise.
“Oh, you wanted me to make a soda fizz, right? We should get St. Germain because that’s good in it.” … starts getting St. Germain commercials every 5 minutes. Didn’t even know they had commercials. Fucking St. Germain.
EDIT: And I have tested it. Came up with stupid things to say in front of speakers a few times, and some have definitely shown up in really bizarre TV ads. I get it, I know why Chewy is advertising to me. But what about the Plant Based Burger commercials I got in a sudden storm for a month? I’m a meat-lover, but decided to talk about plant-based burgers in my living room where the speaker is.