Hwang is once again correct that behavioral targeting ads aren't as accurate as many vendors and platforms lead advertisers to believe. This is not surprising to anyone in the Banner Design digital advertising world or anyone who has received a poorly targeted ad. This is especially true given the mountains of data fed into platform bidding algorithms, both from: Platform-enabled trackers (cookies, pixels, beacons, application data and others that capture online activity). Third parties and other advertisers (transferring data from CRM systems + augmenting with other offline datasets, etc.). This data is of varying quality, accuracy and relevance – and mixing it up can have disastrous consequences on the overall accuracy of the data. By analogy, consider what would happen Banner Design if you mixed perfume with smelly excrement - the result would still be quite unpleasant. The same thing happens with high quality data and low quality data. Advertising Continue reading below And when that poor data gets into other ecosystems, it can pollute them too. Making matters more complex is the fact that this process is often automated, with many machine learning algorithms by default trusting the validity of input data and assuming that the view is complete (i.e. all relevant data are included in the set).
Both of these things are almost never true - so most of the data that passes through these algorithms is somewhat flawed. As a result, many algorithms Banner Design are pretty good at signaling when a user's purchase intent is increasing, but terrible at determining when it's decreasing. This leads to sticky situations where a user makes a relevant purchase (e.g. a kettlebell or microwave) on a website, but is continually tracked by ads for more of the same item for Banner Design days, weeks, or years. months after purchase. Although this is a specific example, it illustrates the larger problem: Advertising Continue reading below Advertising money is wasted (directly or indirectly) on users who are simply no longer relevant to the brand in question and are not going to buy. The cynic in me thinks it's intentional. After all, having larger “in-market” audiences increases auction participation, and in Banner Design doing so, appears to make platforms more money (more bidders = higher bids = more money for advertising platforms).
The technologist in me believes that (i) merging massive datasets in near real-time, each of them generated by sources of varying quality and (ii) then Banner Design drawing on those datasets to understand the intent and user behavior is an incredibly difficult problem. to solve in a vacuum - and virtually impossible on a large scale and without perfect information. The reality is probably somewhere in between, with some platforms trying their best to solve the puzzle, and others just sitting back and collecting checks from customers/advertisers. That's a long way of saying: yes, some of the data used to target programmatic ads is in varying degrees of rot - but that's not necessarily a problem. Advertising Continue reading below All data decays – there is no Banner Design escape – the question is how good are platforms at removing bad data before it further pollutes the data set. For me, the more pertinent question is: is this really such a serious problem as Hwang suggests? After all, in Hwang's argument, faulty data is like bad coordinates fed to a Tesla rocket - a small mistake is enough to cause a big boom.