Performance management used to be a fairly familiar mix: objectives, supervision, feedback, appraisals, and (sometimes) formal capability processes. Increasingly, though, the first thing shaping performance conversations isn’t a person at all. It’s a system.
In many workplaces, software now tracks output, flags risk, allocates tasks, ranks performance, or predicts demand and schedules labour accordingly. Sometimes it’s obvious, like a dashboard with real-time targets. Sometimes it’s built into routine tools, like ticketing systems, call analytics, or shift allocation platforms. Either way, the effect can be the same: data starts to define what “good” looks like, and the people doing the work feel it in their day-to-day reality.
This is often called algorithmic management. It can genuinely improve consistency and reduce admin. But used badly, it creates quiet unfairness, defensiveness, and stress — and that can undermine the very productivity gains organisations are chasing.
What algorithmic management looks like in practice
Algorithmic management doesn’t always mean a single “AI tool”. It’s usually a series of automated or semi-automated decisions shaped by data. Common examples include:
- performance scoring based on speed, volume, error rates, or customer ratings
- automated flags for “underperformance”, “attendance risk”, or “disengagement”
- shift allocation based on predicted demand (often with limited notice)
- ranking systems that compare workers to each other, not to a standard
- nudges and alerts designed to increase pace or reduce breaks
- recruitment filtering that shortlists or rejects applicants automatically
In plenty of roles, the system is positioned as “support” rather than “decision-maker”. But in reality, once a score or flag exists, it starts to steer outcomes. Managers begin a conversation with a pre-set narrative (“the data says you’re slipping”). Employees feel judged before they’ve been heard. Over time, trust erodes.
A useful test is this: Is the system helping people do better work, or simply making it easier to police them? The answer usually predicts how staff will respond.
Why these systems can make people feel worse at work
People can accept measurement when it feels proportionate, understandable, and connected to meaningful outcomes. Stress tends to rise when monitoring is constant, opaque, or disconnected from how work actually happens.
The biggest pressure points are:
Opacity: If employees do not know what is being measured, how it is weighted, or how it will be used, they fill the gap with worst-case assumptions.
Over-simplified metrics: What is easy to measure is not always what matters. Speed can beat quality. Volume can beat judgement. “Response time” can beat thoughtful problem-solving.
Loss of context: A system rarely sees complexity: a difficult case, a vulnerable customer, a technical fault, a new starter learning the ropes, or a colleague needing support.
Fear of consequences: When data feels punitive, people adapt defensively: they game the metric, avoid difficult tasks, take fewer breaks, and stop asking for help. You might get “better numbers” and worse outcomes at the same time.
This is where culture shifts without anyone explicitly choosing it. People become less creative, less honest, and less willing to take sensible risks. That is not a performance win.
The fairness problem: disability, neurodiversity, and adjustments
One of the most serious risks of data-driven performance systems is that they quietly enforce a narrow idea of “normal” work. That can disadvantage people who work differently, even when they work well.
For example, an employee may produce high-quality output but not in a steady linear pattern. Another may communicate most clearly in writing. Someone managing chronic health issues may need flexible pacing. If the system equates “good” with a single standard of speed, availability, or output rhythm, difference can be misread as deficiency.
This is why adjustments cannot be treated as a side issue. They have to be compatible with the way performance is monitored and assessed. If the monitoring system makes it hard to apply adjustments in practice, the organisation is building legal and reputational risk into its daily operations.
A helpful deeper read on what this looks like across recruitment and performance is: How do employers handle neurodiversity adjustments in recruitment and performance processes?
When AI rollouts happen without consultation
Even a well-designed system can fail if it is introduced in the wrong way. Where new monitoring, scoring, or scheduling tools appear without meaningful explanation, staff often experience it as a shift in power rather than a productivity improvement.
The common signs of a poor rollout include:
- unclear communication about what data is collected and why
- a “trust us” approach to scoring or ranking
- new targets introduced without resourcing changes
- managers relying on dashboards instead of supervision
- limited routes to challenge inaccurate data or unfair inferences
In these environments, the psychological contract changes. People feel watched rather than supported. Small issues become big grievances because employees feel they are not being treated as credible witnesses to their own work.
This is especially acute where there is no formal mechanism for consultation, feedback, or collective input on how the system is used. A useful perspective on the lived experience of this is: How do workers experience AI rollouts when there is no formal consultation or collective bargaining?
The hidden business costs of getting it wrong
Organisations often adopt algorithmic management to improve efficiency. Ironically, misusing it can generate costs that are harder to see on a dashboard but very real in the long run.
Attrition and recruitment churn: If high performers feel mistrusted or unfairly judged, they leave. Replacing them costs far more than improving the system.
Sickness absence and burnout: Constant measurement can push people into sustained stress responses, particularly where targets are high and autonomy is low.
Quality failures: If speed is rewarded, errors rise. If difficult cases are avoided because they hurt metrics, risk increases.
Manager capability erosion: If managers stop coaching and start “reading the numbers”, their people skills decline. That makes problems harder to fix later.
The real goal is not more data. It is better judgement. Data should support that judgement, not replace it.
What good governance looks like
You do not need to reject modern tools to use them responsibly. The difference is governance: clarity, accountability, and safeguards that protect fairness.
Make the system explainable: Employees should be able to understand what is measured, how it is used, and what decisions it does not make. If it cannot be explained clearly, it is not ready to influence someone’s job security or progression.
Treat metrics as prompts, not verdicts: A dashboard should trigger investigation and conversation, not automatic conclusions. Context should always matter.
Build in routes to challenge and correct: Data can be wrong. Systems can misattribute time, misread activity, or punish work that is valuable but less visible. Employees need a clear way to raise issues and have them addressed.
Audit outcomes, not intentions: If one group is disproportionately flagged, ranked lower, or scheduled less favourably, investigate. Sometimes the problem is not bias in the software, but bias in the assumptions the organisation encoded into “good performance”.
Keep humans accountable for decisions: “Computer says no” is not a defensible management position. Someone senior should own the policy and the outcomes, not just the procurement.
Avoid metric overload: Too many measures create anxiety and gaming. A smaller set of meaningful measures, balanced with qualitative judgement, usually improves performance more.
What employees can do if they feel unfairly measured
You do not need to be technical to respond effectively. The best approach is calm, specific, and evidence-led.
- Ask what is being measured, over what period, and how it is weighted.
- Ask what “good” looks like in plain terms, not slogans.
- Bring context that the system may miss: complexity, constraints, system faults, and the quality of outcomes.
- Keep records of work quality and decision-making where metrics do not capture it.
- If you need adjustments, request them early and link them to outcomes and quality, not just comfort.
- Most importantly: treat a metric as a representation, not a truth. You can respect data without being defined by it.
Where this is heading next
As AI systems develop, workplaces are likely to use them not only to measure performance but to recommend actions: who should be placed on a plan, who should be promoted, who is “high potential”, who is “at risk”, and what training is “needed”.
That raises the stakes. The more automated the recommendation, the more important it becomes to insist on transparency, accountability, and human judgement. People do their best work when they believe the system is fundamentally fair. When they believe outcomes are decided by a black box, motivation tends to collapse into compliance.
Algorithmic management can support good work, but only if it is implemented with a clear human standard: proportionate monitoring, explainable decision-making, and genuine room for context and difference.
