3 ms·
The RTB industry is IMO missing a trick by using properties of people for targeting rather than "info vectors" and "bloom lists". The info vectors would be gen
by londons_explore 2y ago
The RTB industry is IMO missing a trick by using properties of people for targeting rather than "info vectors" and "bloom lists".
The info vectors would be generated by a neural net based on all information known about a customer. The vectors would specifically not be usable to identify a customer, and therefore not count as PII under GDPR rules. The vectors would be trained to predict the probability of clicking an ad (the pCTR). Using that training metric, everything an advertiser cares about will end up encoded in the vector, whilst no individual private piece of info (eg. sexual orientation) is extractable with any certainty.
Notably, these vectors will be addable, allowing multiple companies to add their vectors on one visitor, creating a new vector with even greater predictive strength, but no company shares their private customer data to competitors.
The 'bloom lists' will be able to identify customers, but only probabilistically. They will encode data such as "user bob@mail.com has an account at Walmart, Lowes and Target". However, the data they hold is only right ~99% of the time, since it is implemented with a bloom filter. That allows businesses to do remarketing to specific users, for example past customers. No individual user could ever be sure they are or are not part of such a filter, and hence it again doesn't meet the EU's PII definition.
By using the above two, I believe the total CTR can be increased (since users get better targeted), whilst also giving users the privacy they expect.
- staunton 2y agoNo inference made on data is ever guaranteed 100% to be right. So I'd expect the legal standard in this case to be "are companies able to identify individuals from the data" and if that's possible in 99% of cases, any court will rule against you.
- AlexandrB 2y agoI think a lot of people think that bad targeting is a technology problem, but isn't it also an economic one? It doesn't matter how good your categories are if ftx.com pays for a massive ad buy across all demographics and annoys the shit out of everyone. The opposite is also true - products I might actually be interested in probably don't get enough budget for me to see an ad for them more than once.
- londons_explore 2y agoThis is solvable too. If an ad is 'annoying' and makes a customer install an adblock or stop using a website, then the website should be compensated for all the future lost revenue. Thats easy for an ad exchange or publisher to enforce by simply subtracting a factor from all bids from an advertiser to 'pay' for the chance of future loss of revenue caused by said annoyance. That factor would be decided based on data of past customers who stopped using a site or installed an adblocker. There are plenty of other similar factors already in use - they're typically aggregated into a "quality factor". They also encompass ads which make a promise but don't deliver ("Everyone wins an iphone!"), or otherwise generally leave the user unhappy (which, in aggregate is quite easy to estimate using bounce rate, dwell time, conversion rate, etc)