Algorithmic Management: Treat the Algorithm as a Managerial Message

Communication · Executive Insight Brief · 234

The score is a management decision

When software assigns, evaluates, recommends, and disciplines work, executives must govern not only the code but what workers can understand, infer, and contest

Premise

Managerial communication no longer requires a manager to be present. Work can be assigned by a matching system, priorities signaled through an app, performance evaluated through ratings or tracked data, and future access to work altered automatically. Workers may experience these actions as decisions made by “the algorithm,” even though the system embeds policies, objectives, data definitions, and design choices created by people.

For executives, this changes the communication problem. Prompts, scores, rankings, warnings, maps, incentives, and access decisions do more than carry information; they can perform managerial functions. What the system measures, displays, rewards, or restricts may communicate organizational priorities more powerfully than formal policy language.

The essential model

Lee, Kusbit, Metsky, and Dabbish introduced the term algorithmic management in 2015 through research on Uber and Lyft, focusing on software algorithms that assume managerial functions together with the institutional arrangements supporting them. Kellogg, Valentine, and Christin later widened the field through six recurring mechanisms of algorithmic control: restricting and recommending to direct work, recording and rating to evaluate it, and replacing and rewarding to discipline it.

The model matters because organizations must translate broad objectives into machine-readable proxies. “Good service,” “safe driving,” or “fair contribution” become variables such as response time, ratings, productivity, errors, or acceptance rates. Those proxies become consequential when they trigger assignments, incentives, rankings, warnings, or exclusion. Opacity then creates an explanation gap: when official logic is unclear, workers develop informal theories from observed patterns and peer discussion.

Why this matters

Consider an illustrative customer-support system. The algorithm rewards short resolution times and high customer ratings, then gives top-ranked agents first access to desirable shifts. Managers still tell employees that difficult cases deserve thorough investigation, but agents learn through consequences that complex cases damage their rankings and begin transferring them whenever possible.

The executive task is to audit the entire message chain: what data are collected, which categories are inferred, what output is shown, what action follows, what explanation is available, and how a worker can challenge an incorrect result. Before deployment, specify which managerial decision is being automated or supported, what each indicator can and cannot represent, what human judgment remains, and which decisions require review rather than automatic execution.

Risks and limits

Algorithmic management is not one uniform technology. A recommendation tool used with professional discretion is not equivalent to a system that can deactivate a worker. Employment status, bargaining power, data quality, legal rights, and the amount of human discretion materially change the relationship. Systems also evolve quickly and can combine several control functions in one interface.

The framework does not determine whether automation is fair, efficient, or legitimate, or how productivity should be traded against autonomy, privacy, explanation, consistency, and due process. The founding 2015 study was qualitative and the 2020 synthesis was a broad review rather than a meta-analysis. The literature documents recurring mechanisms and tensions, not one universal performance effect.

Executive takeaway

Govern algorithmic outputs as consequential organizational communication. Make criteria intelligible at a meaningful level, distinguish recommendations from requirements, disclose the limits of proxies, preserve human accountability, and create contestability that can actually change a result or the system. If workers must reverse-engineer managerial rules from penalties and rankings, the organization has delegated authority without providing adequate explanation.

Key questions

• Which managerial functions does this system actually perform: directing, evaluating, rewarding, restricting, replacing, or some combination?
• What organizational value is each metric or proxy meant to represent, and where can that representation fail?
• What can a worker understand about why an output occurred and what evidence can overturn an incorrect decision?
• Where do worker workarounds or peer explanations reveal a gap between formal policy and the incentives encoded in the system?

Selected sources

• Lee, M. K., Kusbit, D., Metsky, E., & Dabbish, L. (2015). Working with machines: The impact of algorithmic and data-driven management on human workers. Proceedings of the 33rd Annual ACM Conference on Human Factors in Computing Systems, 1603-1612.
• Kellogg, K. C., Valentine, M. A., & Christin, A. (2020). Algorithms at work: The new contested terrain of control. Academy of Management Annals, 14(1), 366-410.