This is the sort of thing machine learning algorithms are pretty good at at.
Coupled with however many millions of interactions a day, you would have no problem correlating changes to your algorithm against increases in revenue.
But. It’s often not that impressive. Humans are equally good at noticing patterns.
All it takes is for one person at FB to see their wife or daughter delete a post, ask them “why did you delete that post” and take away from the response of “It made me look fat” to go “there’s a new targeted ad that’ll get me a bonus”.
In a similar vein, 80% of your banks anti-fraud systems isn’t deep learning models that detect fraudulent behaviour. Instead it’s “if the user is based in Russia, add 80 points, and if the account is at a branch in 10km of Heinersdorf Berlin, add another 50…. We’re pretty sure a Russian scammer goes on holiday every 6 months and opens a bunch of accounts there, we just don’t know which ones”.
This is the sort of thing machine learning algorithms are pretty good at at.
Coupled with however many millions of interactions a day, you would have no problem correlating changes to your algorithm against increases in revenue.
But. It’s often not that impressive. Humans are equally good at noticing patterns.
All it takes is for one person at FB to see their wife or daughter delete a post, ask them “why did you delete that post” and take away from the response of “It made me look fat” to go “there’s a new targeted ad that’ll get me a bonus”.
In a similar vein, 80% of your banks anti-fraud systems isn’t deep learning models that detect fraudulent behaviour. Instead it’s “if the user is based in Russia, add 80 points, and if the account is at a branch in 10km of Heinersdorf Berlin, add another 50…. We’re pretty sure a Russian scammer goes on holiday every 6 months and opens a bunch of accounts there, we just don’t know which ones”.