Business Case Studies & Business Publications

Bias in AI Advertising Models

Biggs, Max;Freeman...

Case

Bias in AI Advertising Models

Biggs, Max; Freeman, Rupert; Parmar, Bidhan L.; Kalogeropoulos, Demitri; Klopfenstein, Amy

E-0518 | Published August 7, 2026 | 5 Pages Case

Collection: Darden School of Business

Product Details

JobHive, a professional social networking platform, must develop an algorithm for targeting employment advertisements while avoiding discriminatory outcomes. Students take the role of a data scientist using historical advertising data to determine which users should receive a job ad. They must consider potential proxy variables for protected characteristics and balance competing measures of algorithmic fairness against the financial costs of ineffective targeting. The case challenges students to recognize that quantitative analysis is not value neutral and to examine how ethical judgments are embedded in decisions about predictive performance, fairness, and business objectives. At the University of Virginia Darden School of Business, the case has been jointly taught by business ethics and decision analysis (DA) faculty in the core business ethics course. Requiring only rudimentary spreadsheet analysis, it is suitable for undergraduate or MBA courses in business ethics, AI ethics, or quantitative analysis and can be taught by one or multiple faculty members. The calculations can also be performed using standard AI tools without compromising the case’s central teaching message. An excellent companion to the case is "Quantitative Analysis and Ethics" (UVA-E-0527), a technical note that provides a broader framework for examining the ethical judgments embedded in quantitative analysis.

- Understand that ethical judgments are embedded into quantitative analysis. - Evaluate and defend specific trade-offs between desirable but mutually incompatible features of a solution. - Appreciate that solutions to a technical problem may lie outside of the specific quantitative model used to frame the problem.