Chase had ads on 400k sites, then on just 5k, with same results
1–10 of 156 posts
Re: Chase had ads on 400k sites, then on just 5k, with same results
#2Edited for elaboration:
The article doesn't mention the precise methodology for reaching the 5000 sites number, other than some human filtering being involved.
The precise wording of another section is also vague, the 'cost of an impressions' could mean either their total Internet marketing expenditure is the same (thus 395000 sites were largely ineffective and merely exposed them to risk and complexity) OR that they've reduced their spending by around 98+% with no appreciable change in effectiveness.
Re: Chase had ads on 400k sites, then on just 5k, with same results
#3Edited for elaboration: The article doesn't mention the precise methodology for reaching the 5000 sites number, other than some human filtering being involved. The precise wording of another section is also vague, the 'cost of an impressions' could mean either their total Internet marketing expenditure is the same (thus 395000 sites were largely ineffective and merely exposed them to risk and complexity) OR that they…
If the result would have been the same with zero sites, the 5000 can be selected randomly.
Re: Chase had ads on 400k sites, then on just 5k, with same results
#4> Surprisingly, the company is seeing little change in the cost of impressions or the visibility of its ads on the internet, she said.
If/when other companies switch to this strategy, they’ll all be competing for less inventory and cost will go up accordingly.
Re: Chase had ads on 400k sites, then on just 5k, with same results
#5Eh, same would have happened with billboards, newspapers, radio, etc. Bad marketing cannot be fixed by mass deployments.
Re: Chase had ads on 400k sites, then on just 5k, with same results
#6What's the shape of the long tail distribution? Does the top 5000 get 99% of the traffic?
Re: Chase had ads on 400k sites, then on just 5k, with same results
#7> At some point, a human is going to take a look.
I recently saw a talk by Foster Provost, a big ML guy at NYU. The main points of his talk were that using fine-grained behavioral data (like browsing history) is better than demographics (at least in his context: predicting ad lift), and he proposed a way to interpret the model (somewhat). I left feeling disappointed. His system for interpretation was super post-hoc and tenuous, IMHO. It felt like a computer scientist doing social science (because it was).
I think ML is great for lots of things, but there are still lots of problems with using it in systems with humans.
An example: if you do a Google search for "Amazon " then you'll almost always get an ad from Amazon. So Amazon pays for a click that would have happened anyway. Maybe Amazon does this to crowd out other advertisers, but maybe it's just the algorithm being stupid and nobody is watching.
Re: Chase had ads on 400k sites, then on just 5k, with same results
#8"An intern then manually clicked on each of [the 12000] addresses to ensure that the websites were ones the company wanted to advertise on." - true hero of this article
Re: Chase had ads on 400k sites, then on just 5k, with same results
#9it's only been a few days though, I think this fact alone invalidates the data.
Re: Chase had ads on 400k sites, then on just 5k, with same results
#10This is definitely an interesting tactic, but this is a pretty poor article.
#1. Chase is claiming performance hasn't been affected, but it has only been a couple days since they made changes. With display you can't measure performance in only a few days.
#2. The author confuses the number of sites with the number of impressions. Chase is buying the same number of impressions - if everyone else followed this strategy it wouldn't hurt exchanges. It would have weird outcomes, but if the same volume is served the exchange makes the same amount (excluding data costs and how cpm would be affected, etc).