Popular media have warned us of the ability of disreputable marketers and other bad actors to predict and even control our choices using the latest tracking and artificial intelligence technologies.
In the 2019 Netflix documentaryThe big hack, for example, it is argued that data analytics company Cambridge Analytica has been scraping social media for in-depth insight into individuals’ psyches. Using this information, the filmmakers say, this company was able to craft carefully targeted ads to manipulate the 2016 US presidential election in favor of Donald Trump. While discussing the events depicted in the film, the well-known tech investorRoger McNamee turned outthat tech companies “have a data voodoo doll, which is a complete digital representation of our lives. With it, they can manipulate our behavior. “
Likewise, Harvard psychologist Shoshana Zuboffrecently warnedfrom digital marketers, “The idea is not only to know our behavior, but also to shape it in a way that turns forecasts into guarantees… the goal now is to automate ourselves.”
Obviously, the idea that bad people do and are followers of bad things resonates and is consistent with the public tendency towards conspiracy theories. But as Stanford marketing professor Itamar Simonson and I discuss in arecent articleinReview of consumer psychology, closer examination suggests that the claims are grossly exaggerated.
There is no doubt that advancements in AI (mainly machine learning methods) are enabling revolutions in many areas, including image recognition, language translation and many more. However, predicting people’s choices (and human behavior in general) is quite different from tasks where AI shines. Unlike the goals of these other tasks, preferences for specific products and attributes do not exist to be predicted, but tend to form as decisions are made.
To clarify, while people are likely to have general product preferences (for uniqueness, for ease of use, for quality, for a preferred color), people generally do not have preferences. precise and well-defined for specific products, or how they would swap one product attribute for another.
For example, people are unlikely to have a preference before purchasing a toaster for a particular toaster model or configuration. Likewise, they are unlikely to have a clear preference for the extra amount they would be willing to pay for a slightly more attractive toaster, until they are in the process of making a decision about what to do with it. ‘purchase. In other words, these preferences do not exist to be predicted but are “constructed” in the decision-making process on the basis of many largely unpredictable factors.
This is particularly the case in today’s environment of consumer information, where many of the key determinants of choice (e.g. expert and user reviews, product recommendations, new options) are increasingly encountered by the consumer for the first time at or around the time a decision is made and therefore cannot be anticipated in advance. For example, during the buying process, a consumer may come across a product review that highlights the benefits of a seemingly insignificant characteristic that the consumer had not previously considered, and this could drastically affect the choice. of the consumer. The influence of this just-in-time information makes our choices increasingly difficult, not easier, to predict.
Certainly, in some cases, consumersmakehave strong, precise and stable preferences for particular products or attributes. For example, some people prefer to buy a latte every morning. In such a case, making a prediction is relatively easy and requires little sophistication of the data or methods.
Likewise, in some cases, certain variables will predict differences in preferences between groups of consumers. For example, consumers who have purchased an Xbox are likely to be much more receptive to advertisements for Xbox games than consumers who have purchased a PlayStation. As more of what we do (shopping, “likes”, visits, etc.) is tracked today, “easier” predictions can be made.
However, even with detailed consumer data for targeting, the ability to predict who is likely to buy a product in an absolute senseremains low. In onerecent Facebook campaign, for example, when millions of users have seen advertisements for a beauty product targeted to their personality (based on their history of Facebook likes), on average only around 1.5 in 10,000 people who viewed the ads purchased the product.
Admittedly, this result was about 50% higher than for people who saw the ad butdo nottargeted according to their personality. In other words, personality-based targeting increased the likelihood that a person who saw an ad would purchase the advertised product from about 1 in 10,000 to about 1.5 in 10,000. Such a change in the success rate may be economically significant (depending on the cost of the advertisements and the profit margins of the product), but it is far from having a “data voodoo doll” to manipulate consumers or “automate” them.
In other contexts, the use of highly sophisticated machine learning (deep learning) methods has shown limited ability to improve predictions of people’s choices over basic statistical methods. For example,recent researchfound that using more sophisticated models provided only very slight improvements over a simple model in the ability to predict people’s credit card choices – so poor that, given the cost involved, it was probably a waste of effort.
As another example of the limited ability to predict consumer preferences, consider (the lack of) advancements in recommendation engines, such as those used by Netflix or Amazon to direct viewers to new shows or products based on what they have already watched or purchased. Of themrecent Commentsmost of the gains in predictive precision claimed by increasingly sophisticated methods were referred to as “phantom progress”. Simple methods, they found, tended to work as well as more sophisticated methods, with one review concluding that “progress still appears to be limited … despite the increasing computational complexity of the models.”
For consumers and policymakers, the limited ability to predict and therefore influence individual choices should be somewhat comforting. On the other hand, consumers and policy makers need to be vigilant about manipulating reviews and other information that consumers increasingly depend on in today’s information environment to construct their preferences and make choices.
In other words, we should worry less about marketers knowing exactly what we want (or exactly which buttons to press to manipulate us) and more concerned with the integrity of the information we increasingly rely on. to make choices.
David Gal is professor of marketing at the University of Illinois at Chicago. Follow him on Twitter at @realDavidGal.
More opinion of Fortune:
- The congress has just passed the most important anti-corruption reform in decades, but hardly anyone knows
- Trump focused too much on fraud– and it could have cost him the election
- Why a key Georgia county went from red to blue – and what this means for democrats
- After 2020, we all need a gap year
- Investors are starting to demand better companies they own