🔬 Research summary by Shannon Egan, our Quantitative Research Methods Intern.
[Original paper by Peter Seele, Claus Dierksmeier, Reto Hofstetter, Mario D. Schultz]
Overview: Pricing algorithms can predict an individual’s willingness to buy and adjust the price in real-time to maximize overall revenue. Both dynamic pricing (based on market factors like supply and demand), and personalized pricing (based on individual behaviour) pose significant ethical challenges, especially around consumer privacy.
Is it moral for Uber’s surge pricing algorithm to charge exorbitant prices during terror attacks? Using automated processes to decide prices in real-time (i.e. algorithmic pricing) is now commonplace, but we lack frameworks with which to assess the ethics of this practice. In this paper, Seele et al. seek to fill this gap.
The authors performed an open literature search of Social Science and Business Ethics literature on algorithmic pricing to identify key ethical issues. These ideas were filtered into an ethics assessment – categorizing the outcomes of algorithmic pricing practices by the level of society which they impact. “Micro” for the individuals, “meso” for intermediate entities like consumer groups, or industries, and “macro” for the aggregated population. The outcomes were further sorted as morally “Good”, “Bad”, or “Ambivalent” from an ethics standpoint, forming a 3×3 table that can be used to generalize the ethics assessment.
For all levels of the economy, the authors identify good, bad, and ambivalent outcomes that are likely common to most implementations of algorithmic pricing. Personalized pricing presents additional ethical challenges, however, as it requires more invasive data collection on consumer behaviour.
What is algorithmic pricing?
The idea of dynamic pricing has been popular since the 1980s, but the practice is increasingly powerful and profitable in the age of the internet economy. While it is easy to define algorithmic pricing in technical terms, the authors identified the need for a definition that is useful in a business ethics context. The result is as follows:
“Algorithmic pricing is a pricing mechanism, based on data analytics, which allows firms to automatically generate dynamic and customer-specific prices in real-time. Algorithmic pricing can go along with different forms of price discrimination (in both a technical and moral sense) between individuals and/or groups. As such, it may be perceived as unethical by consumers and the public, which in turn can adversely affect the firm.”
While there are a variety of approaches, all pricing algorithms typically have the same mandate: predict the consumer’s willingness to buy at a given price, either on aggregate (dynamic pricing) or at the individual level (personalized pricing) in order to maximize profit. The use of cutting-edge algorithms like neural networks and reinforcement learning, as well as increased tracking capabilities via browser cookies, enable companies to do this in an increasingly sophisticated way.
Although algorithmic pricing primarily alters the individual consumer’s experience with a merchant (micro), the ripple effects of this practice extend upwards to the organization level (meso), and further to society and the entire economic system (macro).
Evidently, firms benefit from the increased sales that algorithmic pricing facilitates, but could this practice also contribute to the common good? The answer is yes, with some caveats. Algorithmic pricing doubles as a real-time inventory management mechanism. Good inventory management can lead to reduced waste in the product supply chain, thereby decreasing both costs to the firm and the carbon footprint of the production process. The firm will enjoy increased profits, which can also be considered a moral “good” if prosperity gains are shared with the rest of society; either through innovation, increased product quality, or wealth distribution.
The major ethical dilemma of algorithmic pricing comes from the collection of fine-grained consumer behaviour data, as well as the lack of transparency around that data collection. The driving force for algorithmic pricing models, especially personalized pricing, is tracking cookies. This data, which includes browsing activity such as clicks, and past page visits, as well as personal information entered on the site, can be used to finely segment consumers according to tastes, income, health etc. in order to display the most advantageous price for the merchant.
Many consumers are unaware that this information is even being collected, and merchants do not have to ask explicit consent to use tracking cookies for pricing purposes. The onus is left to the consumer to protect themselves if they do not want to be targeted for personalized marketing. This data collection creates a massive informational advantage for companies, which may offset the price advantage that a consumer can gain due to the ease of searching online. It also limits a consumer’s ability to plan future purchases, as prices constantly fluctuate. These ethically “bad” features of algorithmic pricing may limit the economic freedom of the consumer, even if the moral “goods” tend to enable more choices.
Other outcomes of algorithmic pricing fall into a grey area. Surge pricing is effective at managing demand to ensure that services/goods can be delivered to all who need them, but this method occasionally creates a trap of choosing between forgoing an essential service or paying an exorbitant price.
Ultimately, any application of algorithmic pricing requires a case-by-case treatment to ensure that the good outcomes are enhanced, and the bad are appropriately mitigated.
Between the lines
Algorithmic pricing is already transforming our economy, and we need to adapt our understanding of economic systems to one where price reflects not only the value of the good but the consumer’s perception of that value.
The problem of ensuring ethical algorithmic pricing has intersections with many existing domains: privacy, data sovereignty, competition law, micro and macroeconomics. In some cases, existing laws and principles from these related fields can be extended to address algorithmic pricing. However brand new incentives and regulations specific to algorithmic pricing are also needed. For example, policymakers should investigate limiting what time frame the information can be stored for, as well as granting consumers “the right to be forgotten” so that their ever more detailed consumer profile can occasionally be erased.