DHH (of Basecamp and Ruby on Rails fame) wrote a blog post in May titled Targeted ads are staggeringly unpopular so we should ban them. The basic premise of the post was that people hate being tracked online and so many people have opted out of tracking that it’s clear the practice is so unpopular that it’s time to be banned. I think DHH is right about tracking being so clearly unpopular something needs to be done. I don’t think banning targeted advertising is the right thing to do though. The fact that I used to work for an email marketing company (whose product has since been shut down) that promoted targeted email marketing messages probably may have influenced that position, but I think the problem is less the act of targeting an ad, and more how we go about targeting.
A brief defense of targeted advertising
Here’s a quick, 1-sentence explanation for why I think targeted advertising has value: There is no valid reason for me to ever see an advertisement for tampons. I don’t need them, and I’m never going to be the person making the decision on what (if any) to purchase. You are, quite frankly, never going to get your money back if you show me tampon ads.
Here’s a quick defense for another aspect of targeted advertising – I think that when a store I frequent sends me an offers or coupons for specific products, those products should be things that I buy. If you’re going to give me offers, there’s no reason not to try to make the offers for stuff I’m clearly interested in. Seeing as how a business that has enough of a relationship with me would clearly have my purchase history, it’s not unreasonable to ask that any deals I’m offered take that into account.
So how does that statement reconcile with the statistics DHH cited? Am I an outlier? Is the rate at which people rejected targeting mechnaisms inflated? I think the answers to all of those questions is “no.” What people are rejecting is the tracking associated with targeted advertising, not the targeted advertising itself.
How is targeted advertising OK but not the tracking that goes with it?
Personally, I think context is key. Specifically who is doing the tracking, where am I being tracked by them, what information are they gathering, and how are they using it? The truth is, I think we all understand that businesses we patronize track us to some extent. A business tracking my activity exclusively within the context of that business, for the purpose of their customizing offers they make to specific customers is probably something we’re all OK with. As an example, let’s say your local grocery store sends periodic flyers with coupons to households that have a loyalty card with them. I think people are OK if those coupons change from household to household based on what those households actually buy. What people don’t like is when they shop at the grocery store exclusively in-person, but see those purchasing habits reflected in online ads on other sites, especially when they’re from other companies.
Another example of where things cross the line is when 1 person looks at something online, only for their spouse to see an ad for it in their Facebook feed (which happened to both my wife and me after looking at sponsored posts). I’m not talking about something that’s likely to be used by everyone in the house (like food, as in my earlier example), but stuff that’s specific to a person (like gifts).
Even if companies aren’t engaging in individualized targeting, targeted ads based on aggregated data can sometimes seem to cross the creepy line, like that time Target sent a teenaged girl ads meant for pregnant women before her father knew she was pregnant (although that urban legend may be more hype than substance).
Another form of targeting (complete with 3rd-party tracking), there’s lookalike targeting, where advertisers use Google and Facebook APIs to upload a list of their customers, and then target ads to people who “look like” them online. The most common example of this would be some variation of “business builds a list of their best customers for a particular product, uploads that list to Facebook/Google, who match that list against the users they have data on as best they can, and then use that data to find another group of people that are similar to the initial set. The business then buys ads to show that other group of people in the hopes they’ll become active customers too.” It’s worth noting, that Facebook and Google never reveal any information about who is in the lookalike audience. You know the list email addresses that it’s building off of (because you provided it), and that’s it.
In this case, it isn’t the business that’s tracking you, but rather Facebook and Google using their existing tracking to do the targeting for other businesses. This seems simultaneously creepy and OK. On the 1 hand, the businesses buying these ads aren’t getting your personal information. On the other hand, this being possible involves a lot of tracking from 3rd parties that are basically selling random businesses access to you. It’s simultaneously creepy and a valuable tool for businesses to gain customers.
Lookalike audiences bring up another point in ad targeting, which is businesses using their own customer demographic data to target similar demographics to try to get new customers. This scenario is the whole point of targeted advertising. If we’re going to draw a borderline between what’s allowable in ad targeting and what isn’t, this and lookalike audiences are probably the place where that line gets drawn.
Making targeted advertising work without all the creepy tracking
The benefit of targeted advertising for consumers lies in the ads that we don’t see, but for marketers the power is in knowing the customer well enough to figure out the ads that they should be showing. Right now, the assumption seems to be that you do this by tracking activity and using that to infer interest. Advertising networks that capture this information can (in theory) enable marketers to painstakingly craft hyper-specific criteria for showing particular advertisements, and reap massive financial rewards for their specificity. It sounds good, it makes for a good sales pitch, and you can find some really powerful customer stories to reinforce it. But research on targeted advertising performance shows it’s not that simple.
Let’s take the 2 biggest players in the online advertising duopoly – Google (specifically, their search ads) and Facebook. Google targets ads in their search product based on user activity and intent. Facebook, on the other hand, allows advertisers to target based on interest and web activity. The latter is what enables retargeting ads to follow us around forever – somebody somewhere saw you look at something then they used that information to buy an ad on Facebook.
Google was around for years before Facebook, and users seemed to make their peace with how the targeting worked because it was primarily tied to what we were actively searching for in that moment. I know there’s a lot more to the ad selection than that, and that Google collects as much, if not more, data on me than Facebook and uses that to help tailor which possible ads bid for search placement. But the search query is still the primary input for search ad selection. When users are looking for a good or service, they’re much more receptive to the idea of seeing an ad from someone offering that good or service.
Contrast this with Facebook (or even YouTube ads, if you want to keep it all within Google). There’s no clear intent when it comes time to show those ads, so guessing what you’d want to see in that particular moment is much more important. As a result, companies are more desperate for any basis they can use for the ad targeting criteria. In these scenarios, the context is your feed (for Facebook), which presumably has updates on just about everything you’re interested in, or the video you’re watching right now (for YouTube), which may or may not (and probably not to be honest) be indicative of something you’re looking to buy. At best, advertisers have a set of data about your activity on their sites to feed their targeting criteria. The basic theory here is that the more data you have about a user, the better a guess you can make about their interests, and the better an ad you can show them.
This brings me to my hypothesis that targeted advertising that works primarily off explicit user signals of intent perform better (and are better accepted) than other targeted advertising. Back in my “worked for a email marketing company days,” my team wrote a tool that allowed marketers to let their company’s customers shop on their Instagram page (Instagram since came out with a better implementation). The way our app worked was that marketers would launch a campaign to convince their Instagram followers to sign up for the program by filling out a form linking their email address to their Instagram handle (for a variety of reasons, including a team restriction that our apps be built like a 3rd-party app, we wanted this to be a separate marketing channel without using or updating a customer’s existing contact list). While they were getting sign-ups from their customers, the marketers would open our app and designate a hashtag for the overall Instagram shopping campaign would be, as well mapping secondary hashtags to specific email messages to send when a shopper triggered a message. They could also add any personalization data from their end to the email that they wanted (details from the Instagram post, basic Instagram profile information, etc.). Once they had that data mapped, they just made an Instagram post with the 2 hashtags. Their customers would see it, and if they opted in to the marketing and left a comment that included the campaign hashtag, it’d trigger an email message with additional details and any special offers. For 1 customer, each of these messages worked out to about $2 per email sent in sales, versus about $.08 per message sent out for their regular marketing campaigns.
Adding in the “creepy factor”
I believe the differentiator here is the friction involved in the process. The example application above involved its own opt-in process, so users knew exactly what they were getting into, and explicitly opted-in anyways. The other was that messages had to be explicitly triggered by the consumers who would be receiving the emails (since it required them to leave a comment with the required hashtag). That means if they didn’t want to have the item in the post marketed to them, they could simply not leave a comment, or just leave the hashtag out of any comment they entered. Voila, no unwanted emails sent. This resulted in emails going to the people who were most interested in buying the items in the Instagram post, and people who weren’t interested never saw an email ad.
Another example of targeted advertising that doesn’t cross the “creepy line” is another app from the same “worked for an email marketing company days.” Similar to the aforementioned “grocery store emailing people coupons for things they’ve already bought” example, we wrote a tool that would let marketers send offers based on a individual’s browsing history in their online store. We had 3 levels – general browsing (where the user just looked at {X} different items within a configured time period), category browsing (where the user looked at {Y} items within the same category within a configured time period), and item browsing (where the user looked at the same item {Z} times within a configured time period). For the record, there was a “cooldown” period between automated messages to prevent people from getting spammed. Obviously, item-specific messages did better than category-specific messages which did better than general browsing messages.
While these didn’t have the same explicit trigger as the Instagram shopping program, these messages were still linked to users explicit activity, particularly their apparent intent to purchase something (looking at the same item repeatedly, for example). In both cases, the marketing was limited to the company the consumer was interacting with. Both involved friction (either explicitly triggering a message, or repeated viewing above and beyond general curiosity or mild interest). That, in my clearly biased opinion, is how targeted advertising is done right.
Tracking-based advertising really seems like an evolution of brand advertising. You’re basing your ad based on your customer’s interest, but there’s little to no signal in any of it that indicates an interest in making an imminent purchase. As a result, customer interest is supplemented with recent browsing history, and Facebook likes, and whatever else you managed to scrape off the Internet as an approximation of things people may want to buy at some point. As people showed when given the choice, they didn’t appreciate the constant observation.
Personally, I think the biggest factor in people rejecting tracking is the cross-site nature of it. Most of us probably don’t care that Google tracks the queries we enter, whether search, or maps, or images. Most of us probably don’t care that Google uses its knowledge of what we do on Google to pick the ads it shows us. Google tracking us on non-Google sites to build that ad profile feels like an invasion of privacy.
Then there’s the question of how do marketers get this targeting data in the first place? Some places get it by observing customers on their own site, rather using their own tooling or taking advantage of third-party services like Pendo or Google’s newly-announced audience explorer tool. Again, it’s data on people from your site only, so it doesn’t seem too invasive. It’s when that targeting data is based on all of a user’s online behavior (something that Google built and popularized when building it’s own ad network) that gets people up in arms about tracking.
For the people who are bothered by their behavior being tracked by site owners, it’s worth remembering that some of that tracking is intended to improve the site experience for everyone. This includes things like having an invisible pixel in emails to let marketers know customers are actually reading the emails they send out. Some of it is an attempt to see which emails actually convince someone to click the links. In other words, people are trying to do their jobs better (just because that job is to try to sell stuff to you doesn’t make it horrible). The problem comes when marketers start treating this data less as metrics and approximations meant to guide their behavior and more as KPIs to be optimized. There’s a line between trying to ascertain your customers behavior and adapting to it and trying to manipulate it. Cross-site tracking tends to cross over into the latter territory and turn spammy fast.
I think deeper examination into targeted ad tracking would show that most people are fine with companies tracking users on their site, but opposed to a company tracking them all over the Internet. The real problem is that it’s hard to find good research on this if you don’t have access to scholarly journals. Outside of that, it’s a lot of marketing company whitepapers, which seem a little self-promoting for my tastes. The important thing, is that if you look at customer behavior, they’re choosing to opt-out of cross-site tracking, but they’re also responding well to targeted advertising within retail sites. Particularly, they respond best when there’s a degree of friction that needs to be overcome before triggering the advertising, whether it’s having them explicitly trigger it, or only triggering after a relatively high threshold of behavior on the site (enough that only people who likely intend to buy in the near future would meet). By targeting advertising based on customer intent and behavior (even historical behavior) in your store, you can have highly effective ads without scaring customers