,
46 tweets,
9 min read
Read on Twitter
This is very good and correct, behavioral advertising--especially in its current firm--is, at best, a temporary necessary evil and at worst anti consumer. There's a place and space in which it could exist, but we're not near there right now. techcrunch.com/2019/01/20/don…
I argued back in 2017 (with a starter set of instructions on how!) that publishers should be pushing into contextual targeting hard, to everyone's benefit: digitalcontentnext.org/blog/2017/11/1…
But it's not entirely correct behavioral advertising only exists at the behest of Google and Facebook. They are the strongest beneficiaries, and perhaps the most significant contributors to its invention. But publishers share a significant role, because there was a real need.
See, before programmatic was destroying small and medium publishers, it was thought it would save them. Behavioral targeting's popularity among publishers was originally because it was supposed to help them make money off of one powerful man: Drudge.
A decade ago, Digg & Reddit might send you an occasional glob of traffic but, source of the greatest traffic spikes, was The Drudge Report. Publishers lived in hope that he would send his monster size audience their way. And their engineers lay awake at night in fear of the same
If you look back at caching and hosting products pitched at publishers during the early part of the aughts, you'd see a lot of them would be advertised as "Drudge Proof" but what the hell does that mean or have to do with making money on the web?
Well, during that time small and medium sized publishers got their hosting pretty much like you would, they signed a contract for a single amount of money they would pay for their hosting for a given period of time. But when *you* buy $5 hosting, what do you get?
Basically a promise for a particular amount of computing power. Hosts, especially then, make money by overbooking computing power. You buy $5 of computing, but you receive a promise for a hundred dollars worth and the expectation is that you're really going to use $2 worth
This is a great contract for 90% of people who really aren't likely to have their websites use more than $2 worth of computing power, if they remember to set up anything at all. But if you are a small/med publisher you're likely pushing the maximum of that contract to begin with.
Then Matt Drudge links to you and *everything goes off the rails*. Chances are your website's infrastructure wasn't prepared for somewhere between 20,000 and 10 million+ hits in an hour. There are a lot of places it can go real wrong... thewrap.com/drudge-report-…
A lot of SMB websites would have their database crash when Matt Drudge linked to them. Some would have their top layer (at that time, PHP, Ruby, Java or .NET) that queried the database max out their server capacity and keel over. Even the ones w/caching set up might have issues
Generally, your web host, which is already sort of pissed your taking full advantage of them when they build their business plan around you not doing that, is going to go apocalyptic. So publishers who saw it happen more than once might have to budget their server costs up...
You might have to pay for whole layers of infrastructure, like "Drudge Proof" caching or better "Drudge Proof" hosting or an extra engineer to manage your new "Drudge Proof" object cache. All for an event you couldn't rely on that happened occasionally...
So now your small/medium publishing business is paying for some extra expensive hosting that you don't use 20 days of the month. Your budget is shot and worse still, you can't even sell against the Drudge Bump because who knows if it will happen next month!
This is the space that programmatic advertising and behavioral targeting took advantage of. See when you sell ads as a publisher, even before the era of user tracking, you sell on what you understand to be your audience. But, Drudge likely sends a very different audience your way
Drudge often linked to articles he or his audience disagreed with and the end result is not only could you not sell this unreliable bump in traffic, but even if you did manage to capture that audience regularly the audience wasn't the type you usually sold
In this environment programmatic & behavioral made sense. The idea was that your company could live off the advertising revenue generated by a Drudge Bump for the whole year and support the hosting costs involved in being *able* to regularly generate Drudge, Reddit or Digg bumps
(Also, to give credit where it is due, Fark was pretty good at sending large masses of traffic around this time too)
Exactly! See Programmatic was part of the de-aligning that a lot of publishers were interested in doing with their ads, because they were generating a lot of revenue off of abnormal audience that wasn't anything they usually sold on.
The inventor of the pop up speaks to this in a piece he wrote in 2014. The core idea of the pop up ad is to separate the ad from the content, which didn't work and continues not to work for a lot of different reasons - theatlantic.com/technology/arc…
But in the aughts era of the internet this made a lot of sense. You could pay for outsized hosting costs while surviving a Drudge Bump and then maybe try and build a larger audience to fit your new regular day-to-day capacity for traffic...
But this equation didn't last very long see at the same time small/medium publishers were Drudge Proofing their website two things were starting to happen - The rise of Social Media with Facebook & Twitter. And the launch of AWS in 2006.
Social media could render traffic bumps that, while often not the size of Drudge's audience (even today!!?!), were decently sized and regular. And AWS changed the hosting equation for many publishers either when they moved on to it directly, or when their hosts did.
See AWS did (and does) put high-level web infrastructure tools into the hands of pretty much anyone. This means load balancers, scaling infrastructure, and attaching direct costs to traffic. It means you could enter into a new hosting approach where you pay for only what you use.
What AWS enabled was for sub-NYT-level operations to minimize their hosting costs while building support (hopefully!) for traffic bumps using tools that let them deploy additional computing power to their site only when needed. This had some interesting effects...
It meant that hosting costs went down, but SMB budgeting for hosting became a lot more complex and inaccurate. Compounding this issue of understanding costs, publishers bought into ad tech that promised them even better profits by matching the user targeting capacity of Google
This promise of enhanced user targeting was matched with promises of enhanced verification of ad views all of which didn't really work and increased cost via vendors while decreasing the value of any given impression. Ad tech blossomed going from 150 to ~7000 products in 7 years
Most of ad verification didn't work, and most user targeting wasn't very effective. As Ellen Pao notes, the promise of a unified user tracking is oft impossible, with users switching devices, clearing cookies, or otherwise making their IDs non-consistent
Google & Facebook drive their user accuracy because users are logged in to a wide variety of services they use through their platforms. This combination of regular logins giving you a steady ID and a variety of input gives them more accurate data, but they still get plenty wrong.
This level of data and capacity for standard ID is not available to SMB publishers, no matter how much Ad Tech promised otherwise. So the actual user targeting ended up general and based on patterns over time on a site.
B/c ad tech firms were promising better user targeting capacity & could deliver their increasing promises on a decreasing amount of the audience the value of traffic on publisher sites started to go down. Instead of going after traffic, publishers had to go after types of readers
This is why companies founded during the recent era are so strongly in pursuit of particular audiences, it keeps their baseline ad value up.
As promises of accuracy w/user targeting went up, traffic costs started exceeding their capacity to deliver profit. The cost of going viral now is two-fold. It costs extra infrastructure that may not have been planned for and it dilutes the value of your audience, decreasing CPMs
This is where this all pulls together in the modern death spiral we see for small/medium publishers. They are paying for infrastructure under the assumptions of one era, and earning money under the assumptions of another. Fatal, but easy mistake, the eras were like, 5 years each.
Smart publishers are trying to fix this by further decreasing infrastructure costs. We see this in convos by engineers at Bustle or Buzzfeed who talk about going "server-less". This is really short hand for increasing the bottlenecks between traffic going up and costs going up.
There's a lot of cool tech in play using stuff like AWS's Lambdas and others that they've talked about that equates to shrinking the amount of time their servers stay awake and costing them money.
But for the average SMB publisher without those smart engineers (and even for BuzzFeed it seems), going viral is more a curse than a reward, even if they don't know it. This is why VC-funded publishers that focused on scale are falling apart. Scale *doesn't* work.
Scale is expensive and the assumption that Ad Tech would deliver user targeted ads on the scale involved and optimize their profits accordingly has been pretty much proven wrong at this point.
No we're back in a weird space. Getting a Drudge Bump is a nightmare for small/medium publishers again. Not because their infrastructure can't handle it, but because their ad tech can't assign value to it. And they are bad at budgeting for it. They're back to losing money.
But this gets at the *core* of the problem. We want publishers to be *able* to go viral when they write something really good. We don't want them to go down and we don't want them to go bankrupt. It breaks the fundamental assumptions of the web.
Direct sales are *important* to a publisher business. But they're not sufficient to cover costs in a world where Matt Drudge might send you millions of pageviews tomorrow & you get hundreds the day after. Programmatic (& user targeting) came b/c there's a need left unfulfilled...
And it has only been filled by vampiric liars who call themselves ad tech. Bad user tracking is *bad*. Its current form is anti-consumer and aggressive and working in an ecosystem filled with fraud. But publishers need some version of this that works and respects their readers.
The programmatic ad system will never disappear because there is an acute need for a working version of it. The cost structure of being an open and functioning web publisher requires it. Solutions that propose to entirely do away with it will never work. We need another way.
Finding a solution to the problem of bad ad tech means being aware of, and understanding, the problems and needs that caused programmatic and user targeting to come into existence in their current form. Hopefully this thread helps.
ugh, *form
My favorite example of programmatic shit going wrong on a viral event: When I worked at Salon, every time Drudge sent a crapload of traffic our way we'd be dealing with pro-gun-related programmatic ads for weeks afterwords, even though the audience that remained hated them.
