Originally published in CMS Wire.
I have worked for more than 20 years in the advertising and marketing industry and I finally have the courage to confess my true feelings. “I am a data skeptic.”
Through the years we were promised an impressive leap forward in marketing proficiency through the ability to target the right people, at the right time, in the right place.
As the years sent on the promise of data became more and more fragmented — i.e. advertisers looking for women, 25-34-years-old in a high income bracket that bought at least 3 cosmetic products in the past two months and have an interest in health. Honestly, this was a request I received. This kind of data-driven targeting is a marketers dream.
If you know there is a 40 percent increase in buying membership to your gym brand for example, with women fitting the description above, targeting them is the most effective way to manage your campaign.
So as good as all this sounds, why and how did I become a data skeptic? There are four reasons, which I’ll outline here:
Uniqueness (of the User)
Let’s examine the following scenario: You are over 35 with two kids, and open your home computer to the YouTube home page. There is an extremely high probability that most of the videos will be related to your kids preferences and viewing habits. YouTube recommendation is one of the longest running and most developed algorithms with huge amounts of data, yet even this algorithm is often mistaken.
The main challenge with data is not the concept but the quality and relevance of the data.
The Youtube example relates to cookie/attributed users. Evidently, the problem is that in many cases several people often use the same computer; actually in most cases several people use the same computer and or same IP address.
In mobile, you would think the situation is much better since it is a personal device, but most of the applications we use request a login to the app and if you have children like in our example above, very often the children login through the parents’ log-in (naturally, this is the appropriate protocol so the parents could see all their kids activities).
In summary, targeting data should be unique and personalized, but currently there is no way to define someone in this manner across desktop or in most mobile apps.
Have you ever wondered where DMP’s acquire their data from? How often do they update it? How they define person as 35+ person? Is it a questionnaire? Behavioral assets? and so on …
The fact is, currently there are no data quality certifications, only self-regulation. Actually you do not have any method for evaluating your vendors’ data quality.
If you are a professional in this field, try to target a campaign using one DMP and then audit It with a second one. What was the discrepancy? Probably higher than 20 percent.
If you want to see how Google defines you just follow this link.
Apparently I like SUVs and construction too. But in reality, I don’t own a car or an SUV, in fact I’ve never thought of buying an SUV and never fixed anything in my house.
Bottom line is you have no way to know or verify the quality of the data. As such, it’s about time this area of our industry be regulated.
As discussed in the two previous points, consumer device usage is almost evenly divided between mobile and desktop. However, many consumers also will have a tablet. How do I connect the right user information between all these platforms and devices?
Currently, you have two options: having an application signed in on those different platforms (i.e. Facebook) or using WiFi, calculating where you are located and then assuming that if a PC and a mobile device are using the same WiFi they must be the same person.
The WiFi calculation has its drawbacks, it often calculates several people in the home or office as the same person, and doesn’t give you the granularity of people online, just households, etc.
Which leads us to the last and most important point of my data skepticism
Audience targeting promises to reduce your waste calculation, so you will present relevant ads only to the most relevant person.
But, on the other hand, it means for the publisher that there will be inventory which might not be sold.
If everyone is only buying female audiences, what should the publisher do with the male segment of its audience? When we get to hyper-segmentation, there is a valid possibility of not buying a user that we do not have relevant details about. The easy solution from the supply side is increasing pricing.
In the gender example, I will request +25 percent on female audiences as it is oversold and need to do so balance the unsold male audience. On this you have ~30 percent commission on the data company.
Simple calculation shows that it is better to buy all demo and have 50 percent waste….
The hyper-segmentation issue is apparent in all publishers. P&G recently decided to stop hyper-segmentation even on Facebook (which in all other points was the better alternative).
Big data targeting in theory is the best way to manage your campaign. However, in practice, in most cases, the data quality and accuracy are not nearly as accurate as you would expect. The pricing too makes it much less effective. And, this is why I am such a data skeptic.
We need to push the industry toward data certification and better auditing of data. Until then, “spray and pray” models (buying all audiences at low cost) remain a valid and good option for all the reasons I outline above.