How Audience Extension Hurts Data’s Integrity

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

Today’s column is written by Jeff Sporn, SVP and GM for digital solutions at IXI, an Equifax division.

Data is important for every online advertiser, but the rise in demand has led many to take shortcuts.

One common practice is audience extension, whereby data exchanges or ad networks build lookalike modeled audience segments off of authentic first-party data. This helps keep costs down on the back end, but the data is often passed off to advertisers as if it was true first-party data (and sold at a first-party rate).

When I was growing up, the same type of thinking was applied to an incredible CPG product called Hamburger Helper. To my young brain, Hamburger Helper magically created more food out of a pound of ground beef. What normally fed two to four people now fed eight to 12. And everyone in the commercials seemed delighted.

But let’s be clear: Those happy customers were not eating pure ground beef, and they were most certainly not eating pure prime rib. Everyone equates less filler with higher value, which is exactly the way we should be looking at pure audience data.

Big data’s big pitch is that it improves marketing with small insights, but when marketers pay for a specific audience and actually get a modeled lookalike, it undermines the white-hat nature of the data business. Like Hamburger Helper, audience extension has its place, but in a growing market, advertisers need to be educated about what it is and its drawbacks.

Audience extension, at the most basic level, is a cheap way to get more reach through intuiting similarities across audiences. The entities producing these extensions take first-party or even third-party data and try to use the insights to increase the audience size. First-party data sets are typically smaller, and are generally designed to identify a narrow, very selective audience set. Ad networks and exchanges, on the other hand, are designed to provide advertisers with the biggest, widest audience sets.

Extension based on similar audiences makes the data less effective. But of course there is growing demand and advertisers are willing to pay for the data; why would a purveyor not promise greater audience reach? Truth be told, some data providers are upfront about the fact that they are selling audience extension, but rarely do they define what it is. Others are less forthcoming.

So, if audience extension is less effective, why does it proliferate in the marketplace? Because nearly everyone along the chain benefits. Agencies are under pressure to fulfill campaign requests. There’s no incentive for them to go back to an advertiser and say they can’t hit the target number of consumers, so they pay for extension. Data purveyors benefit because they sell more data products at a much lower back-end cost (extension is almost always less expensive than actual first-party data).

The main problem is that the end client – the advertiser – gets the worst part of the transaction. While extension may cost less to produce, the less scrupulous data providers will sell it at first-party rates. The advertiser therefore has paid for something they are not really getting. Beyond cost, advertisers never see the benefits (increased revenue) with extension that they expect from genuine data. Expectations, set from previous campaigns, are never met.

Meanwhile, the first-party providers get burned when their audience is extended without participation. A data provider may say a segment contains data from a reputable first-party source, but they never reveal how much data is used. If the campaign performs poorly, the source gets blamed – that brand name is now mud.

Another (science) fictional example: Imagine Earth will explode in six months. Scientists can desperately search for another Earth-like planet in one of two ways. Rigorous scientists can study the proper mix of oxygen, hydrogen and nitrogen, the presence of water, proximity to the sun, temperature and the level of gravity. “Audience extension” scientists, looking for planets that look like Earth, will pick planets that have seven large land masses, or a nice white, blue and green coloring. These two approaches will not yield the same results.

Extension is likely to grow as the data market continues to mature, partly because it isn’t the most damaging thing happening in the online ad industry. Advertisers are still purchasing insights based on a kernel of truth, no matter how small or ineffective that kernel may be.

But extension can introduce real problems if media buyers aren’t more aware of the differences between true first-party data and extension. The data exchanges ask about extensions far more than advertiser clients, which tells you that while these are smart people planning campaigns, they’re not completely caught up in the latest options.

Dishonest players will always try to squeeze money out of a confused marketplace, but educated buyers can prevent that from happening. When buyers know what kind of data they’re purchasing for a campaign, they’ll be happier with the results, which is good news for everyone across the ad landscape.

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. Jeff Burkett

    This is the first time I have ever heard the term ‘Audience Extension’ being used in the same context as ‘Look-a-likes’. I too agree the look-a-likes are the seed of the devil, and that data companies who overdo this will find themselves out of business.

    I just hope we don’t confuse buyers with the multiple uses of the term Audience Extension, which is also commonly defined by publishers as a way to extend ad campaigns off-site with high quality 1st party data. The ‘extension’ is the inventory…not the data quality. When done properly, this data would not be watered down as you describe above. It is full of the rich, high quality data that only a savvy publisher with great insights on their loyal audience can provide.

    Love the article, as it does highlight a huge flaw in the data economy…I just hope the headline doesn’t confuse too many.

  2. Insightful article, however, I would counter that God bless any media company that can sell Look-a-Likes at the same rate as first-party data! Advertisers and agencies are savvy enough to know that audience extension is just that and not the same proposition as any true “first-party” data. I would further make the case that “audience segments” provided by IXI, BlueKai, Exelate are not truly “first-party” either since they are in the business of aggregating valuable “first-party” data from online and offline sources and then packaging it up, marking it up and reselling it to whomever has the budget or need for it. I read recently that maybe we need to introduce the concept of “second-party” data??? Big No to that but at the end of the day, we are all trying to achieve that goals of our advertisers and if Look-a-Likes, pre-packaged audience segments or transparent element-level data works and the costs backs out to those goals we all win! Also, let’s hope Jay Rockefeller gets an answer he is satisfied with and moves on to whatever hot button issue his staff tell shim will raise the staus of his own personal brand.

  3. Jeff – The term “audience extensions” applies to several different analytical and data techniques. The effectiveness of a given audience extension technique depends upon the audience seed (what is the basis for the extensions), the data used for the extension, and the algorithm used by the extension. Not all audience extensions are created equal – and it’s important to understand the difference between the various techniques in the market.

    Comparing audience extensions to hamburger helper may be a funny metaphor, but it’s a gross generalization and, frankly, a bit insulting to data scientists. Let’s say a hypothetical 1st party data consortium aggregates their individual data at the zip+4 level to fill in the gaps of coverage and extend their dataset, instead of selling individual level data. Would that also be considered hamburger helper? I think you’d make a strong argument against the above data and analytics technique as “filler”, right?

    I do agree with your sentiment that buyers need to understand how a given audience extension was created in order to gauge how well it fits their business needs…but I don’t think it is correct to broadly generalize all audience extensions as “less effective”. For some business use cases such as new customer prospecting, audience extensions can be more effective than 1st party data because you’re trying to reach outside of your existing customer base. Like most things in our business, the details really do matter and part of our mission should be to educate potential buyers on the options (and pitfalls) in the market.

    -david dowhan
    President, TruSignal