Course contentsShow
← Machine Learning and Deep Learning
Lesson 3486 of 3,53876. Ethics and Societal ImpactFree lesson

Case Studies in Stakeholder Engagement Failures and Successes

Examine real-world examples where stakeholder involvement prevented harms or where its absence led to system failure or backlash.

Case Studies in Stakeholder Engagement Failures and Successes

What you'll learn: How real-world examples demonstrate the critical impact of involving—or excluding—stakeholders when deploying ML systems.

Why Case Studies Matter

Understanding stakeholder engagement isn't just theoretical—real projects have succeeded or failed based on how well they practiced the principles you've learned. Examining these examples reveals patterns that can guide your own work.

Notable Failures

Amazon's Hiring Algorithm (2014-2018): Amazon developed an ML recruiting tool that showed bias against women. The system was trained on historical resumes—predominantly from male candidates—without involving HR diversity teams or women in tech during design. When discovered, the damage to trust was severe. Failure point: No participatory design with affected stakeholders; power dynamics ignored.

Google's Gorilla Image Classification Incident (2015): The image classifier tagged photos of Black individuals as "gorillas." The company hadn't adequately engaged diverse user communities during testing. Failure point: Homogeneous development team; insufficient community review before deployment.

Healthcare Risk Prediction Bias (Obermeyer et al., 2019): A widely-used hospital algorithm systematically under-referred Black patients for care programs. Medical professionals and patient advocacy groups weren't consulted on proxy metrics. Failure point: Stakeholder values not translated to technical requirements; feedback loops absent.

Notable Successes

Detroit Community Technology Project: When deploying facial recognition, Detroit established community review boards with residents, civil rights advocates, and technologists. This led to stronger oversight policies and usage restrictions reflecting community values. Success factor: Advisory panels with meaningful power.

ProPublica's COMPAS Investigation: While not a deployment success, this investigation showed how external stakeholders (journalists, affected defendants) can create accountability through transparency demands.

The Pattern

Failures typically stem from treating stakeholders as afterthoughts. Successes involve them from problem definition through deployment, creating iterative feedback loops and managing conflicting interests transparently.

Key Takeaway: Stakeholder engagement isn't a checkbox—case studies show that meaningful, early, and ongoing involvement prevents catastrophic failures and builds systems that serve all affected communities.