CMO @ Marketing Architects | Marketing Effectiveness Student & TV Advertising Enthusiast
Third-party data promises precision. In practice, it's sometimes too expensive and inaccurate to justify it. In our latest episode of The Marketing Architects Podcast, we discuss the study “How Effective Is Third-Party Consumer Profiling and Audience Delivery?” by Nico Neumann, Catherine Tucker, and Timothy Whitfield. The researchers ran three large-scale field tests across 19 data brokers and more than 90 audience segments to test how well third-party targeting actually works. Study One simulated a typical programmatic campaign targeting men aged 25-54. Using DSPs and Nielsen Digital Ad Ratings, they found that on average only 59% of impressions reached the intended audience. That’s better than random targeting (which would have hit 26.5% of the time), but it came at a high price and accuracy across DSPs varied widely. Study Two removed the DSP layer and looked at the raw accuracy of the audience data itself. The results were worse. Across 11 data brokers, the average accuracy for identifying men 25 to 54 dropped to just 24%. Gender classification averaged 42%. Age accuracy ranged from 10% to 32%. The study found in homes with kids, targeting accuracy declined significantly. This is likely due to shared devices. Study Three tested interest-based segments like “sports,” “fitness,” and “travel.” These performed slightly better, with accuracy between 72.8% and 87.4%. Still, there was wide variation by provider. The authors also ran a cost-benefit analysis. For high-CPM placements, the cost of targeting might be justified. But for lower-cost placements (like display), third-party data often adds significant cost without enough improvement in performance to justify it. This line sums it up perfectly: third-party audiences are often economically unattractive. So if you're investing heavily in hyper-targeted media plans, it's worth asking: Are you actually reaching the people you think you are? Or just paying for the illusion of precision? Links in the comments to the podcast episode + full study. 👇
An important distinction should be made between "personal targeting" vs. "audience targeting". The lure of the data brokers is that you can personally target (and "personalize") ads down to the level of unique individuals. This is clearly a colossal failure! Audience targeting is estimating the concentration of a given ICP within a given channel. You don't need to know the personal identity of a SINGLE member of that audience to be able to do this! --- Personal targeting is about fetishizing "ad wastage" (we don't want to pay for anyone to see our ad who's not ready to buy). Instead of optimizing for maximum ICP reach at the lowest effective-CPM rate. A channel audience that's only 1% your ICP at $5 CPM is fully equivalent to a perfectly targeted channel with 100% ICP (which isn't reality) at $500 CPM It's all about effective-CPM, not personal targeting, which is largely a fantasy to actually achieve.
I need to actually read the entire paper, but from your synopsis it doesn't necessarily sound like the targeting wasn't effective, just not worth the ROI (except study 2 which says a lot more about the data brokers than anything else). Am I reading it right? Assuming that's correct, isn't a viable question: why don't DSP's right-size the business model to make it a larger part of the conversation? If targeting is more effective for a brand, it likely means it's also a better experience for the consumer. Seems like a win-win, no? Same dollars spent, same # of ads, just to a more defined audience. In the US at least, RMN's and other 1P providers are getting better and better at executing against it. If the broader industry doesn't jump on board, they shouldn't be surprised when these dollars move out of traditional channels...
Elena Jasper Really insightful! I am dating myself😊, but I remember working with third-party data in the '90s and early 2000s. It certainly was far less sophisticated, but I’d have to say working with it and testing, it was much cleaner. Fewer overlaps, fewer platform accounts per person, and fewer privacy workarounds. Today’s ecosystem is flooded with fragmented, duplicated, and often outdated profiles. Add in shared devices, multiple emails, incognito browsing, and increasing consumer reluctance to share real info due to privacy. I think people are much more sophisticated now. I’ve found that data brokers are really hit and miss, and the point of this post – pretty darn expensive when calculated into the total costs. It seem like we’ve prioritized scale and automation, but at the cost of reliability. Makes you wonder how much of our media “precision” is just a mirage. Appreciate you elevating this conversation.
Interesting. Poor ROI for sure! What’s wild is how far off some of the basic demographics are - you’d expect at least gender or age to be directionally right. It appears almost "random".
right? "Arguably this isn’t a very significant problem, considering the accuracy of that third-party data. An analysis by researchers from MIT in 2019 of over 90 third-party audiences across 19 data brokers revealed that the cookies identified gender correctly only 46% of the time on average. Since that is less accurate than guessing randomly it seems hard to justify the additional budget." https://www.warc.com/newsandopinion/opinion/advertisings-future-past/en-gb/4126
They should also involve Augustine Fou to track how many of them were actually real people
If “waste” is cheaper than targeting, it’s not waste.
Some of the numbers on this are brutal. I've read that up to half of these 3rd party providers mess up the gender of the person, and even interest-based targets miss more than a quarter of the time. I think the solution is for us to stop marketing to the outside of the person -- their immutable traits -- and market to the inside of the person -- how they think, act, and perform. The study didn’t test these, but I have employed approaches that are far better than 3rd party (and often outperform first-party data), like: 1. Understanding customer Big Five OCEAN Personality Trait Targeting (Big 5 traits > demographics) 2. Semantic Identity (Targeting why people buy vs. who they are by understanding the language they prefer and priming them with it) 3. Neo-Geo Targeting (Urban/coastal/rural + water proximity > ZIP codes. Not 'where' a person lives, but 'how' they live) I've seen far better results and understood total addressable markets much better using the inside of the person than just seeing them from the outside.
Really? "Targeting men aged 25-54"? Don't experts these days call out targeting by age as a dismal marketing incompetency? For decades, a target segment has been recognised as a group of people with similar of identical, needs, wants and behaviors... How can contemporary business executives STILL fall into the demographic trap?
CMO @ Marketing Architects | Marketing Effectiveness Student & TV Advertising Enthusiast
2moPodcast episode: https://podcasts.apple.com/us/podcast/nerd-alert-why-your-hyper-targeted-ads-might-be-a-waste/id1679048313?i=1000721022674 Study: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3203131