Stretched Inferences

A recent article in HBR called “Buying Consumer Data? Tread Carefully” by Professor Catherine Tucker at MIT is clarifying for anyone who buys audience data for targeting or personalization of marketing. 

You should read it yourself but here’s the gist … 

Businesses spend an estimated $200B per year on data from brokers to help them improve marketing experiences with audience segments like "fashion interested" or "men 25-54”.

But there’s a problem. 

The process by which data brokers create these segments is kept secret for competitive reasons, so marketers don’t always know if they can trust the data or how well it aligns with their own segment definitions. 

This is exacerbated by the fact that 3P data purchases typically impact large portions of media budgets. So if the data is bad, the risk extends beyond just the cost of the data.

Oftentimes segments are built using probabilistic methods, leading to stretched inferences about individual audience members that aren’t really there. 

This point echoes in the snarky question from New York Times CEO Mark Thompson - “When we say a member of the audience is a female fashionista aged 20 to 30, what’s the probability that that’s actually true?”

Professor Tucker and her project team tried to answer that question, by cross-validating the accuracy of a range of audiences and brokers with external sources like Facebook where data is declared rather than inferred.

What did they find? 

First, demographic data is particularly bad.  

For example, the average accuracy of gender segments classifying individuals as males was only 42.5% accurate — lower than the 50% natural chance of just guessing.

There are several others you can read about in the experiment write-up.

These findings suggest that marketers should be cautious of buying audience data unless the sources and methods of segmentation are open and visible.

And that marketers should limit their exposure to narrow segments, to avoid campaign inefficiency.

Read more here … 

https://hbr.org/2020/05/buying-consumer-data-tread-carefully