Blog 06 - Performance Management: Can an Algorithm Really Judge an Employee?
Performance Management: Can an Algorithm Really Judge an Employee?
Performance management has always involved judgement. Managers decide whether objectives have been achieved, whether performance is improving, what support an employee needs and, in some cases, whether performance should influence promotion, pay or continued employment. Artificial Intelligence (AI) and algorithmic management are now changing how that judgement is produced.
Digital systems can continuously collect information about sales, response times, productivity, attendance, customer ratings, task completion and many other aspects of work. Compared with a traditional annual appraisal, this appears attractive: the organisation has more data, more frequent measurement and potentially more consistent decisions.
But more measurement does not necessarily mean better management.
Performance Management Is More Than a Score
The CIPD (2026) defines performance management as the set of practices through which organisations maintain and improve employee performance in line with organisational goals. It includes objective setting, performance measurement, feedback, learning and development, reviews and, where appropriate, consequences such as reward or performance improvement.
Importantly, the CIPD argues that effective performance management should not be treated as a once-a-year appraisal. At its best, it is a continuous, two-way process involving regular, open and supportive discussion about progress and improvement.
Figure 1. Performance management as a continuous cycle. Author’s illustration based on CIPD (2026).
This distinction is essential when considering AI. If performance management is reduced to a numerical rating, automation seems relatively simple. If performance management is understood as an ongoing process of direction, feedback, learning, accountability and development, then human interaction becomes much harder to remove.
DeNisi and Pritchard (2006) similarly argue that performance appraisal should ultimately contribute to improving individual performance rather than simply producing evaluations. A performance system that measures employees accurately but fails to help them improve may therefore be technically sophisticated but managerially weak.
Why Organisations Are Using Algorithms to Manage Performance
Algorithmic management refers to software that automates or supports managerial tasks such as directing work, monitoring workers and evaluating performance. The OECD’s 2025 employer survey across France, Germany, Italy, Japan, Spain and the United States found that algorithmic management is already widely used in most of the countries examined. Evaluation tools were especially prevalent in the United States (OECD, 2025).
There are understandable reasons for this development. Digital tools can collect information continuously rather than waiting for a manager to remember examples at the end of the year. They can apply the same criteria to a large workforce and identify patterns that might otherwise be missed. For organisations with thousands of employees, this can provide a level of visibility that traditional supervision cannot easily achieve.
Managers surveyed by the OECD generally perceived algorithmic management as improving the quality of their decisions. However, they also raised concerns about unclear accountability for algorithmic decisions, difficulty understanding the logic of the tools and insufficient protection of workers’ health (OECD, 2025).
AI can provide managers with more timely, consistent and detailed performance information.
Once a number appears on a dashboard, managers may begin treating it as objective truth even when the number measures only part of the employee’s contribution.
What the Data Can See—and What It Cannot
Performance data are often strongest when work produces clear digital outputs. A call centre can measure average handling time. A sales organisation can measure revenue. A delivery platform can measure completed tasks. A customer-service system can track response time and ratings.
But not every valuable contribution appears in a dataset.
Figure 2. Performance metrics and the wider context of employee contribution. Author’s illustration.
Imagine two customer-service employees. Employee A closes 40 cases in a day while Employee B closes 25. An algorithm focused on output may rate Employee A more highly. However, Employee B may have been allocated the most difficult customers, supported two new colleagues and resolved several complex problems that prevented future complaints.
The metric is not necessarily wrong. It is incomplete.
This illustrates a broader issue in performance management: what gets measured can become what gets managed. Narrow performance indicators can encourage employees to optimise the number instead of the underlying purpose of the job.
The CIPD (2026) warns that narrowly designed performance-related measures can create undesirable behaviour when targets do not fully represent good performance. AI can magnify this problem because it allows narrow metrics to be applied continuously and at scale.
Can AI Feedback Help Employees Improve?
AI does not have to be used only for control. It can also provide feedback.
A system might identify recurring errors, suggest better ways to complete a task or give employees immediate information that would otherwise arrive weeks later. CIPD research has examined the potential for AI-generated feedback to support employee performance, while also highlighting the need to understand how workers respond to feedback delivered by technology (CIPD, 2022).
The distinction between developmental AI and disciplinary AI is therefore important.
Developmental use: “Here is a pattern in your work that may help you improve.”
Disciplinary use: “The system has scored you below the threshold; therefore you will be penalised.”
Employees are likely to experience these two uses very differently. The first positions data as a resource for learning. The second positions the algorithm as an authority over the employee.
The Problem of Algorithmic Bias
An algorithm does not automatically remove human bias. Its outcome depends on what is measured, how the system is designed and what data are used.
For example, an organisation might train a performance model using historical ratings given by managers. If those historical ratings contain patterns of unfairness, the model may reproduce them. Alternatively, a system may indirectly disadvantage employees who work differently because of disability, caring responsibilities, language or job location.
The U.S. Equal Employment Opportunity Commission has repeatedly warned that employers remain responsible for complying with employment discrimination law when they use AI and automated systems in employment decisions. Automated tools are not outside ordinary employment responsibilities simply because a vendor designed them (EEOC, 2023).
The EU Treats Employment AI as a High-Risk Area
The seriousness of this issue is also reflected in the European Union’s AI Act. AI systems used for employment and worker management—including systems used to monitor or evaluate workers—can fall within the Act’s high-risk category because of their potential impact on career prospects, livelihoods and workers’ rights.
Under the current EU implementation timeline, rules for high-risk systems in employment are scheduled to apply from 2 December 2027. The Act includes requirements concerning human oversight, relevant input data, monitoring, documentation and worker information. Employers using relevant high-risk AI systems are also required to inform affected workers and worker representatives that such systems are being used.
Figure 3. Key governance principles for AI-supported employee evaluation. Author’s illustration informed by OECD (2025) and the EU AI Act.
This regulatory direction supports an important HR principle: the more serious the consequence of a performance decision, the stronger the need for explanation, review and human accountability.
Employee Voice Can Improve Algorithmic Management
The OECD (2025) argues that worker consultation can reduce risks associated with algorithmic management while increasing employee engagement with and acceptance of the technology.
This is logical because employees often know where a performance metric becomes misleading. A delivery driver understands which routes are unusually difficult. A lecturer knows that class outcomes depend on student circumstances as well as teaching effort. A salesperson knows that markets and customer portfolios vary. A support agent understands that not every customer case has the same complexity.
Performance systems should therefore not be designed only by software vendors and senior management. Employees and line managers should help identify which indicators are meaningful, which contextual factors should be considered and how employees can challenge incorrect data or decisions.
A Better Human + AI Performance Model
The strongest approach is therefore not to reject performance analytics. Data can improve management when it is used appropriately. The objective should be to use AI as an additional source of evidence while preserving human responsibility.
A responsible model should include:
- Clear objectives: Employees understand what good performance means before it is measured.
- Relevant measures: Metrics reflect the real purpose of the role rather than only what is easy to count.
- Multiple sources of evidence: Quantitative data are combined with quality, context, feedback and managerial discussion.
- Human oversight: Important career decisions are reviewed by accountable people.
- Transparency: Employees know what data are collected and how they influence decisions.
- Right to challenge: Incorrect or misleading performance information can be questioned.
- Development focus: Data are used to help employees improve, not only to rank and punish them.
- Regular review: HR checks whether the system produces unintended bias or unhealthy work pressure.
A Global HRM Perspective
Global organisations face an additional challenge because attitudes towards monitoring, privacy, employee voice and managerial authority differ across countries. Brewster et al. (2017) emphasise that International HRM operates within different cultural, legal and institutional environments.
A global company may therefore want one consistent performance platform while facing very different expectations and regulations in each country. The EU’s approach to high-risk employment AI is one example of how local regulation can shape the use of global HR technology.
This means multinational organisations should avoid assuming that a performance-management system designed in one country can simply be deployed everywhere without adaptation. Global consistency must be balanced with legal compliance, cultural understanding and employee trust.
🎥 Recommended Video
CIPD Senior Adviser Jonny Gifford discusses why modern performance management is moving away from isolated annual appraisals towards more continuous performance conversations.
▶ Watch on YouTubeMy Reflection
Before examining this topic, I believed that having more performance data would naturally make performance management more objective. I now think that this is only partly true.
More data can reduce reliance on memory and personal impressions, but data still depend on what the organisation chooses to measure. If the wrong behaviour is measured, technology can make the wrong judgement more consistent rather than more accurate.
I also realised that performance management should not be treated only as an evaluation exercise. Employees need feedback, coaching, resources and opportunities to improve. An algorithm can identify a pattern, but it may not understand the full reason behind that pattern or know what support the employee requires.
For me, the most appropriate future model is therefore not manager versus algorithm. It is a model where AI provides evidence and human managers remain responsible for interpretation, discussion and decisions.
💬 Join the Discussion
Would you accept an AI-generated performance score if it affected your promotion, salary or job security?
What information should an algorithm be allowed to use when evaluating an employee, and which parts of performance should always remain a human judgement? Share your view in the comments.
References
Brewster, C., Sparrow, P., Vernon, G. and Houldsworth, E. (2017) International Human Resource Management. 4th edn. London: CIPD.
CIPD (2022) Can AI feedback improve employee performance? Available at: https://www.cipd.org/en/knowledge/bitesize-research/ai-feedback-employee-performance/ (Accessed: 12 August 2026).
CIPD (2026) Performance management: An introduction. 29 January. Available at: https://www.cipd.org/en/knowledge/factsheets/performance-factsheet/ (Accessed: 12 August 2026).
DeNisi, A.S. and Pritchard, R.D. (2006) ‘Performance appraisal, performance management and improving individual performance: A motivational framework’, Management and Organization Review, 2(2), pp. 253–277.
European Commission (2026) Guidelines for providers and deployers of AI high-risk systems. Available at: https://digital-strategy.ec.europa.eu/en/policies/guidelines-ai-high-risk-systems (Accessed: 12 August 2026).
European Union (2024) Regulation (EU) 2024/1689 laying down harmonised rules on artificial intelligence (Artificial Intelligence Act). AI Act Service Desk, Article 26. Available at: https://ai-act-service-desk.ec.europa.eu/en/ai-act/article-26 (Accessed: 12 August 2026).
Marchington, M. and Wilkinson, A. (2020) Human Resource Management at Work. 7th edn. London: CIPD.
OECD (2025) How widespread is algorithmic management in workplaces? Paris: OECD Publishing. doi:10.1787/cda7a114-en. Available at: https://www.oecd.org/en/publications/how-widespread-is-algorithmic-management-in-workplaces_cda7a114-en.html (Accessed: 12 August 2026).
U.S. Equal Employment Opportunity Commission (2023) EEOC Hearing Explores Potential Benefits and Harms of Artificial Intelligence and Other Automated Systems in Employment Decisions. Available at: https://www.eeoc.gov/newsroom/eeoc-hearing-explores-potential-benefits-and-harms-artificial-intelligence-and-other (Accessed: 12 August 2026).
Considering your MBA background, a insightful comment would be that while algorithms can provide objective data and patterns, true performance management still requires human empathy and judgment to understand the full context of an employee's contributions.
ReplyDeleteYour two customer-service employees say it perfectly. The one closing 25 cases might be the one everyone leans on, and none of that shows up on the dashboard. The line about the metric being incomplete rather than wrong is a fair way to put it. And once a number is on a screen, people do start treating it as the truth.
ReplyDelete