Programmatic advertising, a summary
3rd February 2021
Recently I read Subprime Attention Crisis by Tim Hwang. I was hoping to find out if advertising was the reason for social media changing from chronological posts in your timeline to a grab-bag of algorithmic bullshit. I didn't get an answer to that question, but I did get a scare about how fragile the digital advertising economy really is.
Most of the information in this post comes directly from Hwang's book or articles I looked up from his bibliography. If this topic is relevant to your work, I strongly suggest you read the book instead! If you're local to me, you can borrow my copy. Or if you prefer audio, there's a great interview with Hwang on the Freakonomics podcast.
What's the problem?
These days, most digital advertising is programmatic. This means that people with things to advertise buy slots on websites via a trading system that runs a lot like the stock market. Hwang explains how that came about, then suggests that the current programmatic ad market has a lot of similarities to how the stock market was running just before the Global Financial Crisis. The features he reckons they have in common are:
- opacity - no-one knows exactly what they're buying, because individual ads have to be sold through a series of third parties who all charge a fee, and are often re-packaged into giant bundles before being algorithmically deployed by machine learning. In spite of vast quantities of data-tracking, a business has no way of knowing who saw their ad, or where.
- falling value - the quality of what's being sold is dropping but the prices are staying the same. Ad blockers reduce the value of an ad space, and Google's own research suggests that at least 50% of ads are never seen anyway (PDF).
- rampant fraud - when no-one knows what's being sold anyway, it's easy for
money launderers and scammersto join in undetected for some ad fraud. Plus Google and Facebook have also been lying and cheating their clients (I've linked one example where they colluded, but there's plenty of examples of them lying independently as well).
One factor that Hwang touches on a little, but I've seen discussed more in web developer chats, is the performance cost of all the data-gathering. Most big-name websites have dozens of advertising trackers to find out when a page is loaded, who's loading it, what else they've looked at today, and what demographics they're part of. That information then gets fed into the stock-market of advertising to help all the third-parties bundle and re-sell your attention. There's no evidence that this improves ad targeting (PDF), either in the relevance of the ads you get shown or the likelihood of you buying the product. But it creates a lot of jobs and opportunities for fraud so hey, why not take an extra 7 seconds to load the page?
So what can we do?
Alternative 1: Get rid of online advertising
This is kind of the scorched earth option. If current advertising sucks, why not just ditch it altogether? The problem with this is that someone has to pay for content authors and design and hosting, etc. If advertising isn't paying the bills, who will? It's not a huge stretch to guess that a lot of sites would move to a paid subscriber model.
Given that so many sites are dependent on Facebook and Google for traffic, they have a lot of control over how this plays out. They've never given half a damn about democracy or equal access before, and they're always looking for another way to make a buck. So a two-tier internet for the Haves and the Have-nots seems like a real possibility here. And as soon as that happens in any other market sector, you get predatory businesses trying to scam or take advantage of the Have-nots (see: pay-day loans, pawn shops, US health insurance, etc).
So I'm not keen on letting go of advertising altogether.
Alternative 2: let the industry sort it out
Hwang reviews the current attempts by Google, Facebook and the advertising industry to fix this. The short version is: hah! as if!
They're all getting rich from programmatic advertising and will survive just fine even if there is a market crash. They're just tinkering at the edges to look like they're trying. Their efforts aren't designed to make any real, lasting change.
I don't think we can afford to wait for the advertising industry to grow a conscience.
Alternative 3: Gently deflate the market bubble
Hwang says that instead of waiting for the financial bubble to burst, we could attempt a controlled demolition instead. His suggestions seem pretty good to me:
- independent research - currently all the research into programmatic advertising is done by Google, Facebook and marketing companies, usually via astroturf "industry associations" staffed with their mates and full of lobbyists. We need to fund truly independent research into the value of digital advertising, the impact of ad blockers, the amount of fraud, and so on. Without an accurate picture of the situation we can't do much at all.
- whistle-blowing by employees who know their bosses are committing fraud or covering up inconvenient facts. For this to have an impact, we (the general public) have to follow up on scandals and demand punishment and prevention.
- regulation to prevent a crash. As with the stock market, large financial transactions should be monitored for fraud and have regular health-checks.
We're seeing some efforts in the US and the EU towards regulation. Usually politicians aren't up-to-date on technology, but with solid support by the tech industry (the parts of it who care about ethics and long-term stability, at least) they could really help. I'd certainly trust politicians like Elizabeth Warren or Katie Porter in the US, or people like Scott Ludlam and Senator Jordon Steele-John here in Australia, to do their homework on this and make some good changes.
I'd add one more suggestion: that people buying advertising space do an audit of what they're spending on. There's been a few examples lately of brands reducing their daily ad spend by checking conversions and blacklisting sites that aren't brand-friendly (i.e. racist blogs, Covid disinformation, etc). And getting rid of fraudulent vendors didn't reduce effectiveness for Uber (podcast without a transcript, but the page gives a summary). At least this way the clients aren't contributing to the inflation of value.
Alternative 4: Get rid of the "programmatic" aspect
Hwang also suggests that the programmatic stock-market model of advertising isn't necessary, and we could go back to category advertising. I was already persuaded of this idea. Category advertising is where you say "I have a website for people who geek out about productivity software" and so you sell ads to businesses that sell productivity software and related things like cool notebooks and pens, office furniture or all those everyday carry tools. You don't need to know my age, gender, postcode or eye colour to know that as a productivity geek I will always be interested in buying far too many pretty notebooks.
I know that the only two ads I've ever clicked on deliberately were from The Deck, an advertising platform which is now sadly closed. They did the sort of category style advertising which isn't common anymore, and the ads were always simple yet effective. Maybe they had a style guide or constraints on what their advertising partners could do? Anyway, I deliberately whitelisted sites that used The Deck so my ad-blocker would let them through. And one of those two times I clicked an ad, I bought a really great product that I'm still a customer of today.
A more significant example is NPO, a Dutch broadcaster (like the BBC or the ABC) who got rid of advertising cookies to avoid the hassle of complying with the GDPR. They still auctioned off their ad space, but buyers got information about the page being viewed instead of the person doing the viewing. Their digital revenue is up.
This also seems to be backed up by a detail that Hwang briefly mentions: when HotWired started putting banner ads on their sites in 1994, click-through rates were 44%. Nowadays clickthrough rates are more like 0.5%. Not all of this can be put down to the change in advertising techniques, but surely some of it can?
So what next?
I'm a big fan of options 3 and 4 above. But I think we have to move quickly on it, and I don't actually have any control over any sites with advertising or products which need marketing. This makes my support a bit theoretical :)
I figured I'd write up a blog post though and see if I could persuade other people in the tech industry to start looking into how advertising is working for them. Is your current setup helping or harming your work?