I remember the VSOP cognac one (although I’m not interested in it). I see horrible expensive watch ads all the time (I’m a Citizen fan). I don’t remember the appliances, and I’m pretty sure I didn’t see any Michigan tourism ads in mine, but as I’m already here, I suppose your region would get different ads.
Thanks for confirming the bourbon one!
And an embarrassingly large amount of the time, these are just logistic regression models, which are equivalent to the machine coming up with a bunch of clunky rules like, “lots of baby books = pregnant”. The machine is just better at finding these rules.
Raindrop effect?
I don’t mind targeted ads at all (at least once I got over the creepy factor). I’d much rather open my mailbox and get coupons for things I actually use than for a crap I don’t. Likewise online I don’t mind seeing ads for stuff I might actually be interested in.
I don’t mind it either as long as it isn’t both bleeding obvious and brain-dead, like getting served ads for stuff you just searched for and purchased. That’s the opposite of useful.
There was a comic in today’s paper in fact (Ben, I think) in which his wife uses his computer to order Christmas presents for him and he says he will know what she searched for the by the ads that will appear in his browser in the next day to two. So it is well enough understood to be used in a strip. As far targeting magazine ads, I have never seen evidence of it. It sounds to hard to me, although a geographic targeting would be easy enough.
In the past I worked for a consumer research company that did work with Canadian retailers like Loblaws, A&P/Dominion stores (now Metro) and Shopper Drug Mart. Their targeting was not based on simple “machine learning” algorithms, but rather a combination of harvesting their loyalty data combined with detailed consumer interviews. It was not simply buying pregnancy books = pregnant. I’d assume the US retailers have done the same thing.
We’d comb their loyalty data bases and pre-screen customers for a long “loyalty” history, usually at least + 1 year of purchase history. We’d also do a phone screen to make sure they bought at least about 75% or more of their grocery purchases at that retailer. We’d pay around $50 to $100 to complete a very detailed lifestyle survey (demographics, income, kids, aspirations, education, eating habits, competitive shipping behaviour by category etc). We’d then merge that info with their actual purchases and developed based our predictive modelling that data. Normally we did 1000 to 2000 of these detailed surveys across Canada.
Retailers used those models to develop targeted couponing programs, new store locations, new product listing opportunities, new private label product development etc. It was very thorough and sophisticated and developed very detailed shopper purchasing profiles.
That said, the retailers also did do pure data harvesting for couponing programs. A common one was if you were brand loyal, e.g.: only bought coke or pepsi, they’d send you a high value coupon for their own P/L products. I you were coke loyal, pepsi could pay to send you a coupon for their products.
They also used this simple data harvesting for new product launches e.g.: If you’d bought vegetarian meat products in the past, a company launching a new soy burger, would be expected to coupon you when they launched.
Well, yes and no - in my experience, any Fortune 500 company is going to be deploying more sophisticated techniques on any models like this, because when you’re that size, even a 2% lift in accuracy/recall /precision can mean tens or hundreds of millions in additional revenue. But even then, your Random Forest or GBM or Tensorflow model is just a really complicated bunch of conditional rules to classify your outcome.
I don’t see anything in the Wired rate-card about target advertising.
Of course, if they are doing targeted advertising, you’d only see it in the rate card if you…
One of the key points of the company sponsored course was to take us from thinking about the simple basic models to more modern techniques. The professor began by laying the foundation, describing the weakness of many traditional techniques and giving many examples of organizations using those techniques for better and for worse. As the weeks went on, he introduced us to more and more sophisticated techniques, providing many cool examples across various industries that show much more happening than “lots of baby books = pregnant”
The mere fact that a corporation would bring in a professor of the subject from one of the ivy league universities to teach this on company time to hundreds of people should give an idea of the importance.
Sadly, most of this was lost on me, as I am more of a generalist, but it gave me an awareness so I know these things exist and am more prepared to seek them out should I become involved in projects that require this kind of work.
I read through the entire thread again and it doesn’t appear that Balthisar’s original question regarding if publishers can target advertising to specific customers was answered definitively.
I spent the first part of my career in the commercial printing industry with a good bit of time in publication printing and I can affirm that publishers do indeed produce magazines with (different) targeted advertising based upon the recipient’s demographic data provided to the publisher. The process generally applies only to mail copies and not those printed and distributed for news-stand sale. Whether “Wired” magazine does or doesn’t I don’t know, but the technology certainly is available to the publisher.
The trademarked name of the process is call “Selectronic Binding” or Selectronic Gathering" and was originally developed by Chicago based printer R. R. Donnelley in the 1960s. I worked for RRD in the mid 1980s and got to see this process first hand.
Without going into a detailed dissertation of modern magazine production, the binding line machine controller just tells the machine to feed or not feed a certain section of magazine at assembly. These aren’t huge changes in content; it may only be a four or eight page section of the magazine which contains either “advertising package A” or “advertising package B.” The number of different 4 or 8 page sections that might be available for versioning is only limited by the printer’s binding equipment capacity.
The detail or targeting could be as broad as a zip code or as detailed as an individual house.
This article from Mailing Systems Technology magazine is a brief history of print mail technology, mainly focusing on R. R. Donnelley’s contributions: https://mailingsystemstechnology.com/article-2541-A-History-Print-Mail.html
This article from Ag Web Magazine describe how Ag Web customizes content based upon information the farmer provides: https://www.agweb.com/article/this_magazine_is_made_for_you_199098/
I’ve been out of print industry for a few years now, but I don’t doubt that with high speed, high quality wide format inkjet printing that advertisers can offer unlimited versioning with “your” ad copy printed literally seconds before your magazine is assembled - if the technology doesn’t already exist.
2012 – Farm Journal – “The only magazine to use Selectronic binding to create unique and targeted magazines!”
Thanks for this great contribution!