This article from The Intercept has been doing the rounds recently. It follows a case lodged by Investor Village, a small financial news message board, against Facebook. The case alleges that Facebook is knowingly mis-selling advertisers on its ad targeting capabilities.
The original complaint, which you can find here, cites internal Facebook emails in which employees describe Facebook's ad targeting as "crap" and "abysmal".
The most damning line, taken from an internal Facebook email, is:
“interest precision in the US is only 41%—that means more than half the time we’re showing ads to someone other than the advertisers’ intended audience. And it is even worse internationally. We don’t feel we’re meeting advertisers’ interest accuracy expectations".
The compliant goes on to compare this to the sorts of benchmarks which Facebook presents to advertisers; where they claim the Facebook Ads platform allows you to target audiences with 89% accuracy. The true accuracy figure, according to the complaint, was as low as 9%.
These sorts of horror stories are likely not a surprise to many who actually advertise on Facebook's platform. I suspect that most advertisers treat Facebook's targeting claims with a pinch of salt, and certainly don't expect that 100% of people within an audience will actually belong to said audience.
If this is the case though, does it really matter if Facebook isn't 100% accurate with it's targeting? Is 41% really that bad, or is it even good? The answer, I believe, is a little more complex than it first seems.
I don't think it really makes sense to give a single answer to this question. The reasons why advertisers use a platform like Facebook Ads are many, and so to give one answer that covers all advertisers is to oversimplify the matter.
One important distinction between ways that advertisers use the platform, and which is relevant to whether accuracy matters (for them), is whether they're a brand or direct response advertiser.
A brand advertiser advertises to build their brand. They're not (at least directly) measuring sales that result from their ads. Instead, they're more interested in building so-called brand metrics (brand awareness, brand favorability, et cetera) in the minds of their target audience.
This last point, of building brand metrics amongst a target audience, is important. If you run a brand selling pet food, then your brand metrics only really matter amongst people who own pets. It doesn't matter if non-pet-owners know your brand or not; they're highly unlikely to ever give you money.
Brand advertisers are in a precarious position therefore. It's critical for them to have accurate targeting data in order to reach their target audience, but they have to rely on Facebook for this targeting data. Facebook might mis-sell brand advertisers on the ability to target a particular audience on Facebook Ads, yet the brand advertiser has no alternative but to trust Facebook.
To make matters worse, there's no real ability for the brand advertiser to validate the audience data that they're paying for. Because the brand advertiser isn't measuring the success of their advertising through direct sales, or anything that would give you a sense of how accurate Facebook's audience data really is, it's difficult for them to question the data they're given.
All of this means that brand advertisers are hurt particularly hard by inaccurate targeting data within Facebook Ads.
The other main category of advertiser, and the one with which most are more familiar, is the direct response (DR) advertiser.
A DR advertiser runs ads solely for the purpose of generating a response. For the majority, this is some form of website or app-based conversion; perhaps a registration or a purchase from the advertiser's site.
Unlike with brand advertisers, DR advertisers have a tight feedback loop between spend and ROI. A DR advertiser will have their ad campaigns hooked up to their conversion event (a purchase on their site, for instance) and will only spend on ads so long as they're able to drive conversion events efficiently.
If a DR advertiser isn't able to generate conversions efficiently (i.e. where their cost per conversion is below their product's LTV), they'll simply pull spend.
In this way, the DR advertiser relies on faith in Facebook Ads to a much smaller extent than the brand advertiser. If a particular audience on Facebook Ads had horrific accuracy, the DR advertiser would likely notice this in the form of a low ROI, and simply choose to put her budget elsewhere. In this sense, low-accuracy audience data hurts DR advertisers far less than their brand counterparts.
The other thing to note, particularly when it comes to DR advertisers, is about how prices are set on Facebook Ads.
Prices are set using a (second price) auction, which means you effectively pay a (constantly changing) market rate to show your ads.
This auction pricing means that the cost to reach a particular audience is relative to how good it is deemed to be by other advertisers (where good essentially means profitable). If a particular audience on Facebook has low accuracy, then this will likely be reflected in advertisers bidding less on that audience, and prices dropping to the point where it's just as efficient to bid on that audience as a more accurate one.
In this sense, auction-based pricing tends to level out and account for differences in audience quality.
One last thing that's worth considering, is that most (DR) advertisers nowadays don't take a particularly audience-centric approach to advertising. Gone are the day where you'd pick out a couple of key audiences, and funnel your spend through them.
Even if Facebook's pre-packaged audience data is lousy, the machine learning that powers its delivery algorithms certainly isn't. Because these algorithms are so efficient at picking out high-value users within any audience, it often doesn't really matter what audience your start with.
As a result, it's become increasingly common for advertisers to target broader audiences (perhaps a whole country) or use lookalike targeting to effectively create their own audiences.
This trend is particularly noticeable on larger accounts, where the ads are generating so much data that it makes more sense to let the delivery algorithms run unconstrained by specific audience targeting.
In this sense, bad audience data is really more of a concern for smaller advertisers. These advertisers may not have enough data to build robust lookalikes from, or to let a campaign run wild across a whole country, and that's why they're pushed towards using Facebook's pre-packaged audiences.
There are really two segments of advertiser hit hardest by Facebook pushing advertisers to bid on inaccurate audience data:
As we saw from data emerging from the 2020 Facebook ad boycott however, the segments above aren't bringing in the lion's share of Facebook's profits. Firstly, direct response advertising likely makes up a far larger chunk of revenue than brand advertising. Secondly, advertisers so small that they're wholly reliant on Facebook's audience data aren't (even in aggregate) going to be worth a huge amount to Facebook's bottom line.
In this sense, I can't see there being much incentive for Facebook to respond to this wave of criticism by upping their focus on accuracy.