No More Microtargeting

August 20, 2020
Michael Taylor
Facebook Twitter WhatsApp LinkedIn

Micro-targeting is the practice of running tailored ads to niche audiences at scale, using unique information about each group to increase the effectiveness of your ads.

For example if I have 10,000 people on my website browsing an inventory of 100 unique products, I can automatically generate a unique ad for every single individual person, and measure what they purchase after seeing or clicking on an ad.

Everything you buy feeds back into prediction models, which also use 1,000s of other variables like what pages you viewed online, the contents of your social posts, chat messages and emails, the location of your phone… everything you do online is used to show you ads.

"The best minds of my generation are thinking about how to make people click ads"
- Jeffrey Hammerbacker
, Early Facebook Employee

The unreasonable effectiveness of micro-targeting

Micro-targeting is so effective it became a ‘threat to democracy’, responsible for getting Trump elected, russian hacking and the Brexit vote. It’s possible with this method to predict you’re pregnant (before your parents know), know if you’re falling in love and predicting your personality traits better than friends and family.

Two of the most valuable companies in the history of the World, Google and Facebook, with a combined $1.7 trillion market cap, owe their 70% market share to this technology, which they provide freely to advertisers. Beginner-level developers are able to use off-the-shelf, open-source algorithms to implement the basics of this technique — you can even replicate it in Microsoft Excel.

“I was able to go from not doing machine learning to knowing what I need to know in one week. That’s all it took.”
- Aleksandr Kogan, Cambridge Analytica researcher

The surveillance economy is unravelling

Micro-targeting is under threat — as part of the fallout of the above listed scandals, wide reaching consumer privacy protections have been enacted, such as GDPR in Europe and CCPA in California. This has made data collection far harder, and riskier.

Consumers have been fighting back too — annoyed at ‘creepy’ ads and worried about the implications of surveillance capitalism, there has been a backlash against targeted ads. About 1/3rd of the internet use AdBlockers and companies are starting to take notice.

Apple, one of the few tech companies that doesn’t rely on targeted advertising revenue, has made moves to ban 3rd-party cookies (used for tracking user behavior) from Safari, and is killing the IDFA, the unique identifier that let’s advertisers track behavior across mobile apps. Even Google plans to kill 3rd-party cookies in Chrome.

The impact of this privacy backlash is predicted to be catastrophic — untracked ads are worth so little Facebook doesn’t even show app install ads to the 30% of US iPhone users who have already opted out of IDFA. Google conducted a study that found a 52% drop in revenue from removing cookie tracking. The ICO found that they lost the ability to track 90% of users when implementing their own best practice consent form for GDPR compliance. 

“He’s making a list
He’s checking it twice
He’s gonna find out who’s naughty or nice
Santa Claus is in contravention of article 4 of the General Data Protection Regulation (EU) 2016/679”

- joe (@mutablejoe) May 20, 2018

Advertising doesn’t work that way

There will be winners and losers from this massive shift in technology and behavior, but marketing will be broadly fine. Before the rise of the internet and micro-targeting, you bought advertising on TV, the radio or in the newspaper. Guess what — from then to now, despite all the changes, advertising has continued to be a stable 1% of GDP since the 1920s.

That’s because advertising works primarily by building long term brands and reinforcing memory structures. If you’re walking down the grocery aisle, and you recognize one brand over another, making you slightly more likely to pick it up and buy it, the marketers did their job. 

Despite common objections, we don’t buy brands because we’re brainwashed into a decision we didn’t want to make. Instead we typically care very little about the brands we buy, and by eliminating other choices the decision is made easier, helping us move on with our lives. 

Advertising does it’s best work in public, via cultural imprinting where costly signalling indicates that a product is reliable (otherwise how could they afford to invest in ads?). This is the same mechanism as how a peacock finds a mate it’s ability to divert resources to growing elaborate feathers is an impossible to fake signal that it’s healthy.

“A flower is a weed with an advertising budget.”
- Rory Sutherland, Ogilvy

Does micro-targeting actually work? 

Big advertisers like P&G, Mars and Unilever have all started to reduce reliance on targeting online ads and shift back to traditional methods and measurement methods like Econometrics, where performance can be independently attributed without reliance on potentially biased measurements coming from Facebook and Google themselves. Even performance marketers are following suit, with even famously data-driven Zynga shifting strategies to broad audience targeting rather than pursuing a handful of ‘whales’ or high spenders. 

There are challenges to claims that Cambridge Analytica’s methods even worked, and only ~30% of conversions claimed by retargeting ads are incremental (i.e, wouldn’t have happened anyway, had you not run the ad). Despite seemingly better measurement methods compared to traditional channels like TV, online attribution methods can be gamed. Facebook claims all conversions that happen within 28 days of an ad being clicked and 24 hours of it being shown — show an ad to everyone in the World and Facebook will implausibly claim to be responsible for 100% of your sales. When 56% of display ads aren’t even seen by a human, keeping micro-targeting looks less important.

I’m very sure there is a lot of real effectiveness in these methods — I’ve personally worked with hundreds of small startups who owe at least some of their growth to methods like these. I’ve seen major quantitative differences in the cost per acquisition between Facebook’s ad platform and it’s less advanced rivals. Despite issues with incrementality, even by the strictest attribution models, retargeting often performs.

Despite my economics background, in the past 10 years I've found myself shifting focus away from targeting and optimization towards creative testing. On clients I typically get results not by micro-targeting, but by regularly testing hundreds of ad variations until I find one that works. When I do find the best performing ad for a niche audience, it usually turns out to be the best ad for every audience I add it to.

The truth is that by far the biggest impact on performance in my experience is creative testing — creative quality is the biggest factor responsible for 49% of campaign performance compared to just 9% for targeting. There I think lies the culprit — the performance marketers who were doing micro-targeting were also simultaneously running thousands of creative variations, which may have been what boosted performance.

“Obviously, it is not big data analytics that wins the election,” he wrote back. “Candidates do. We don’t know how much his victory was helped by big data analytics.”
- Michal Kosinski
, a Stanford professor

Marketing is going back to broadcast 

Contextual targeting, which is showing ads based on the context of where they’re placed (rather than the behavior of the user, never really went away, but is now due for a rise in prominence. If we take the average CPM on Google Search (where ads are placed based on what keyword the user searched for), vs Facebook Ads (primarily based on micro-targeting), the CPMs (how much it costs to show 1,000 ads) are still almost double — $22.94 on Google vs $12.79 on Facebook. So it’s clearly possible that what the user is doing right now (searching for a product) can be worth more than who the user is (demographics, behavioral). 

With the onset of GDPR, the main Dutch public broadcaster dropped personalized ads altogether and saw revenue go up! They built their own ad server for contextual ads, cutting out the more than 50% of ad revenue that goes to the middlemen. Contrary to Google’s findings on the topic, an independent study found that removing cookies only decreased publisher revenue by 4%

Buying digital ads will start to resemble how TV ads are bought — against content that’s most relevant to their products. Shoe brands advertising on sports sites; fitness apps during Youtube workout videos; and airlines on travel blogs. Data-hungry marketers will migrate to subscription businesses, where enforced logging in will ensure users opt into tracking. Some marketers will pivot away from paid ads all together, towards organic channels like Virality. Other marketers will compete on creative — like super bowl ads, segmentation will be done with the content of the ad, rather than the targeting of the placement. New innovations in AI will enable mass generation of creative ideas for large scale testing. If the right customer sees the right message at the right time, if the targeting fits with the content, and privacy is respected, ads have a chance to become more impactful than ever.

“Facebook now talks less about targeting, preferring to emphasize the sheer numbers of consumers that it can help brands to reach. It is investing in video capabilities and is encouraging its clients to make short films. After years of telling clients TV is wasteful, it is now doing a good job of imitating it.”
- Ian Leslie
, Advertising Strategist

September 16, 2020
October 16, 2020

More to read