In this assignment I reviewed a paper discussing the effects of predictive marketing on individuals and added some thoughts of my own. Predictive marketing, both in the paper and in this assignment, is considered the toolkit of algorithmic models that use customer (& other) data to more effectively target customers.
Grade I gave myself for this assignment: 92/100 (it’s pretty good but could be better)
Basics of Predictive Marketing
As outlined in the paper, predictive models aim to predict the value of a single target variable given a set of input variable values. By giving the model a “training set” of input variables with corresponding target variable values, the model can adjust the the relationships and preferences it gives to various variables when predicting the target. The model is then tested by running a “testing set” of input variables and comparing the target values it predicts with the actual target values.
In predictive marketing, the target variable can be a wide variety of things: the likelihood of a customer purchasing a product, the likelihood of the occurrence of a relevant event for a customer, etc. The target variable value assigned to a customer by the predictive model can then be used to make a determination about which marketing strategies will be most effective. My understanding (though this is not backed up by any resource) is that the marketing strategy determined using predictive marketing can then be implemented using the marketing customization tools that many platforms offer.
Concerns with Predictive Marketing
There are many possible risks and concerns with predictive marketing; some are concerns about the use of predictive models for making marketing decisions, and others are limitations of the predictive models themselves.
One concern that Kotras raises is that leveraging such powerful predictive models to inform the treatment of customers poses a potential threat to privacy. Predictive models rely on the availability of extensive data on customers in order to effectively make predictions. How much information is actually collected, stored, and used on the average consumer? Is there a big difference between what consumers think gets collected/used and what actually gets collected/used? My guess is way, way more data gets collected and used than most people realize. I am actually so interested in this that I’m going to try to squeeze this into an entire assignment of it’s own. Stay tuned.
Another concern that Kotras brings up in the paper is that the use of predictive modeling can be considered a potential threat to freedom of choice. I interpret this to mean that predictive marketing aims to show optimally relevant content to each user, but an inherent side effect of this is that each user will be limited to only seeing a small subset of all possible content. I can see the argument that people that are not made aware of all possible choices do not truly have freedom of choice. I would also be curious to hear the opposing argument.
A risk of predictive models mentioned in the paper more specific to the technical models themselves is overfitting, which is when the model becomes so well trained the training data that it cannot generalize to new cases and performs poorly on the unseen testing data. There are a variety of ways to combat this that data scientists learn in school that I won’t get into here.
A very, very important limitation of using algorithmic models is that they learn entirely from the input data that you give them and are therefore highly susceptible to perpetuating any errors or biases in that are already present in that data. This is a complex, insidious concern. What are some examples of this? The classic example is facial recognition models that were trained on samples not representative of diversity in skin color, thus resulting in models that did not work properly for people with darker skin tones.
The next point is inspired by a paragraph from the paper: the purpose of predictive models is to use variables to discriminate (mathematically & purposefully) in authorized ways while avoiding discriminating in any unauthorized ways. The example given in the paper is that it is generally “authorized” to discriminate based on income level but generally “unauthorized” to discriminate on something like gender or race. This ends up not being super straight forward, however, because it is possible that combinations of “authorized” fields can act as a proxy for “unauthorized” fields.
Another huge concern with predictive models and AI in general in that they are not normally built in a way that makes it easy to explain or justify their predictions. You may have heard the term “black box,” indicating that the inner workings of the models are not generally intelligible to humans. This also makes it more complicated to address the previous two limitations of models. Based on little changes I’ve seen on apps I use, I think the data science community as a whole is working on this, but there’s still a long way to go.
Impacts of Predictive Marketing
The core of Kotras’ paper is a discussion of the impacts, positive and negative, on individuals from the proliferation of predictive marketing.
First Kotras highlights one assessment of the way powerful predictive algorithms have shifted the way companies view individuals:
“[This] allows algorithms to classify and treat people according to the digital expression of their tastes, thus intertwining profoundly our liking (what we like) and likeness, or who we are like.”
Kotras, 2020
This felt like a very powerful analysis, though I recognize that segmented and targeted marketing that aim to categorize people according to their “likes” have existed long before algorithms became so widely used. Still, companies have never been able to create such a detailed, thorough picture of customers, and I do agree that this further exacerbates interconnectedness between our identity and our interests.
Then, Kotras offers an interpretation on the way that the meaning of individuality in the context of societies has shifted as a result of these models:
“As suggested by Matzner (2019), the algorithmic sorting of people does not necessarily erase the figure of liberal individuality, but rather reshapes it, in a way that we need to analyze in detail”
Kotras, 2020
I agree that the onset of highly personalized marketing shifts the meaning of the individual rather than erase it, as it leans into unique individual aspects of a person to function. However, I can see how through such targeted marketing there could be adverse effects related to individualism. I would love to hear your thoughts on this because to be completely honest this quote kind of goes over my head.
Kotras then explains a more tangible impact that predictive marketing had in one of the cases he studied. He described how the use of predictive modeling to forecast customer needs allowed for higher quality, more personalized customer advising and support. In addition, he found that the ability to predict customer needs more effectively also made the job of the advisors much more pleasant and fulfilling.
Lastly, Kotras emphasizes that through his observations it seemed critical that that data scientists work closely with colleagues with more on-the-ground experience when designing and building models. In light of all of the concerns with predictive marketing I outlined above, I would say predictive marketing should only be leveraged with utmost caution and oversight.
Thanks for reading.
Works Cited
Kotras, Baptiste (2020). “Mass personalization: Predictive marketing algorithms and the reshaping of consumer knowledge.” Big Data & Society. July-December 1-14. https://journals.sagepub.com/doi/pdf/10.1177/2053951720951581.
BYOMSPM 
Leave a comment