Why The Number One Issue In Retargeting Is Not Privacy

Data-Driven Thinking“Data Driven Thinking” is written by members of the media community and containing fresh ideas on the digital revolution in media.

Today’s column is written by Chris Zaharias, SVP of Sales at Dapper, an online advertising technology company.

Repeat after me:

The #1 issue in retargeting is not privacy
The #1 issue in retargeting is not privacy
The #1 issue in retargeting is not privacy
The #1 issue in retargeting is not privacy
The #1 issue in retargeting is not privacy

If you read the AdAge, New York Times and WSJ articles, it is exceedingly clear that the issues consumers voice around retargeting do not have to do with privacy, but rather with the effects of excessive frequency and poor targeting.

AdAge (see article):

  • The author says that the ‘weirdness’ comes from Zappos’ recommendations “appearing just about everywhere I’ve been on the web”. That’s excessive frequency, folks, not any concern with privacy. ‘Stalking’, as the author calls it, is the issue, and is defined as “a persistent pattern of conduct that is not wanted by the person to whom it is directed.” (StalkingBehavior.com).

New York Times (see article):

  • Ms. Matlin’s disturbed by Criteo’s retargeting not because she perceives it as an invasion of her privacy, but because the product targeting is poor -as is the targeting towards her specifically. As she states on her blog, after viewing the shoes for a millisecond, she realizes she “could never buy shoes online.” And, as was the case with the AdAge author’s stalked-by-Zappos experience, Ms. Matlin takes issue with seeing “Zappos ads for those shoes” on “almost every site or blog I go to”. There again, the issue is that of excessive frequency of targeting.

As marketing increasingly becomes a technology game, the door is open to both the beauty of cookie-enabled site customization and the beast of poorly-executed retargeting of the type Criteo drags eager but inexperienced advertisers into. I virtually whack the knuckles of journalists who confuse the issues in the retargeting debate, and the vendors who hold themselves blameless because they let users opt out. Let me by very clear on this point – letting users opt out of horrifically inept targeting and excessive ad frequency is no different than a stalker telling the judge that he’ll switch to other victims, and that the solution to his stalking is for his victims to avoid public spaces.

For those less directly involved in the space, excessive frequency and poor targeting generally are a result of three things:

  1. By making no distinction between people who have looked at a product and bought it versus people who looked and didn’t buy. Of course people are going to be annoyed when they see ads following them around the web for the product they just bought. A good retargeting system should be able to understand consumer intent and pick in real-time which product to show, yet most systems simply use custom Javascript to scrape the product(s) you view and store them in your cookie, with no thought given to buyers vs non-buyers. Here’s a wild thought: perhaps the user didn’t convert because they didn’t like the offer on a merchant’s site – so another one would be better, and perhaps one that’s not already stored in the user’s cookie. Technologies that rely on in-cookie product storage are as dumb as a physical retailer assistant following you around the store saying ‘Buy that shirt you just looked at’ ten times in a row.
  2. The non-existence of frequency capping. Frequency capping is the most basic of display ad concepts, but apparently the fact that certain vendors on a CPC basis means that they *knowingly choose* a level of frequency that annoys customers (but makes them lots of money). To avoid the weirdness & annoyance that excessive retargeting ad frequency brings, you have to do two things:
    • Buy inventory on the real-time biddable (RTB) exchanges (Google AdX, Pubmatic, OpenX and soon, Yahoo) where frequency capping is easy to implement;
    • Manage frequency across ad networks and publishers when buying outside the RTB exchanges
  3. A non-aligned pricing model. Criteo takes pride in charging on a cost-per-click (CPC) basis because they think it’s equivalent to the CPC model as it works in search engine marketing (Google AdWords, for example). But, CPC is no better than CPM if there’s no focus on the advertiser’s ROI goal. What does Criteo care if they show the consumer a product they have no intention of buying – or worse, already bought – 20 times, when they get paid for clicks? Eventually that user is going to click, as much out of (curiosity & annoyance) x (frequency) as anything else. Oh, yes, but Criteo will say that $2B in transactions have been driven by their system. Whoa, reality check: 100% of those users have already visited the retailer’s site, and 80% of the $2B in transactions would have happened anyways. Smart advertisers will eventually realize that and demand that retargeting systems prove their net-new revenue value *and* brand-building value, and with that pricing models will have to align with advertisers’ ROI goals. When that happens, much of the creepiness that comes with excessive frequency will go away.

It is no surprise that Criteo’s response to those issues is to draw attention to consumer privacy controls, for those are the only ‘controls’ that appear to exist in their system. Advertisers need to educate themselves on the technology behind retargeting vendors’ systems, and make conscious decisions rather than just going with the vendor that has raised the most money and has the most feet-on-the-street.

That’s my $0.02, and I’m happy to hear from those who agree or disagree.

Follow Chris Zaharias (@searchquant), Dapper (@dapper_net) and AdExchanger.com (@adexchanger) on Twitter.

Enjoying this content?

Sign up to be an AdExchanger Member today and get unlimited access to articles like this, plus proprietary data and research, conference discounts, on-demand access to event content, and more!

Join Today!


  1. Chris — couldn’t agree more with your assessment that the frequency of the Zappos ads were what led to user discontent and have written as much here: http://tellapart.tumblr.com/post/1038401457/retargeting-respect-for-the-consumer

    And your comment about the mis-aligned pricing model is also spot-on. The conversions from a retargeting campaign — more than any other type of targeting — need to 1) be based on a click, and 2) be proven to be incremental by way of a strict A:B test. A CPC-based model is a good first step, but the incremental conversion is what the retailer really cares about.

  2. Irene Lopez

    I could not agree more. And what’s unfortunate is that John and Jane Doe are being lead to think that this so called “creepy” advertising is bad by media that doesn’t really understand it.

  3. @Josh McFarland- Definitely in agreement with you as well on the frequency problem. Once users have to opt out, the damage is done.

    And getting to the real incremental lift IS the quest I think we’re both in alignment on – yet I think we need to get away from limiting our tools to analyzing click behavior. There seems to be an ‘all-or-nothing’ approach to attribution on view vs. click, and the truth is consumers respond to display advertising often without ever clicking, but so many vendors try to ram that metric through without any data to back it up.

    I’ll quote Jeff Glueck (former CMO of Travelocity and current CEO of Skyfire) from our podcast in saying that you have to run a test that measures *both* click and non-click based conversions to get at the true lift. Travelocity learned a lot from these studies – our customers do too.

    • Yes, views can be shown to matter. But view-throughs are opaque to the advertiser, prone to gaming (cookie stuffing), and are almost always responsible for far less direct response lift than for which they’re given credit.

      The simplest way for an advertiser to account for view through lift without getting scammed is to run a solid A:B test, figure out the ratio of view:click incrementality, and then add that “bonus value” for views onto the value of the click-driven metric.

      Ex. if for every 10 click driven conversions, there is one view-driven conversion, raise the amount you’re willing to pay for click conversions by 10%, thus incentivizing the provider to seek even higher click-based conversion volumes, knowing that the extra view-based conversions will continue to roll in as well.

      More here: http://bit.ly/c08r0d

  4. Totally disagree. The whole point of privacy concerns is that, once entities get some fragmentary piece of information about you, they run off and make all sorts of assumptions. E.g., “visited diaper website” = “wants to buy diapers” and “Looked up ‘herpes medication'” = “has herpes”. Whoever just gathered that data fragment about you is now going to start tailoring their interactions with you according to this new fragment — and you as the consumer don’t know who knows what or why or what they’re doing with that information. You can’t solve this with a slightly different targeting algorithm – it’s a much bigger issue.

    In order for consumers not to get freaked out by privacy, the records need to be made public, like a credit report, so that I can look at them and correct inaccuracies. “Looked up ‘herpes medication'” was for a school project. And “visited diapers website” was a gift for my new nephew. I don’t want anyone knowing about or using my herpes surfing history unless I know who those people are, what conclusions they’re trying to draw based on that data fragment, and unless I have the ability to dispute their information’s accuracy.

  5. Very close to the #1 issue in retargeting IS in fact privacy. If privacy means “transparency” and honorable advertising, then it is #1.

    Totally agree with your point about ad serving systems not being able to deal with frequency capping, or with having no other tactic besides showing the last SKU browsed is definitely generating the angst about remarketing campaigns.

    One comment –

    “Technologies that rely on in-cookie product storage are as dumb as a physical retailer assistant following you around the store saying ‘Buy that shirt you just looked at’ ten times in a row.”

    Using cookies as browse behavior storage is NOT the issue. Cookies are good for a variety of reasons, not the least of which is providing the ability for the consumer to exercise control over what is being observed about them. Server side cookie stores are rightfully viewed suspiciously – see Phorm or NebuAd. Cookie utilization is not the issue, intent determination is. “Intent determination” must be done on the server on the fly, given sufficient hints from the cookie or browse context or whatever is known demographically about this consumer.

    Based on running many retargeting campaigns, with the Tumri platform, for large advertiser and agency clients, we converged on the following as the requirements of an intelligent re-marketing creative ad server:

    a) Ability to do frequency capping, at the creative experience, SKU and category levels

    b) Understand the real intent of the shopper or the abandoner, and construct the experience on the fly to be in concert with that intent (this is the stuff that requires sophistication in semantics, and scalability in ad serving and analytics, with data mining feedback loops into the ad server for decision making). False positives are hard, as brian above points out with the “herpes” example or the “diapers” search. But these kind of issues exist with search as well, and I don’t see people up in arms about privacy in search. One doesn’t have enough transparency about search engines and what they do with your query history, but that’s another issue that we ought to debate some other day.

    c) Understand when fatigue levels have been reached, and engage the user differently with different tactics of up-sell, cross-sell, branding, interactions, surveys etc. All of this is possible with a dynamic ad generation system.

    d) Has the ability to tell the user exactly what has been observed about their behavior and allow them to remove these entries so as to address:
    – transparency
    – user control
    – and an acceptance that all software systems of this nature can result in false positives.

    Thanks for reading.

    –Pradeep Javangula
    Founder & CTO, Tumri Inc.