Soros's Reflexivity: Why Measuring Employees Changes Them (and Usually Not How You Wanted)

60-Second Summary

Reflexivity (Soros): participants' perceptions of a system influence the system, which changes what participants perceive — a two-way feedback loop between cognition and reality.
Related to Goodhart's Law. Reflexivity is the mechanism: the moment employees know a metric matters, their behaviour reshapes to optimise it, and the metric loses its signal.
Every high-stakes HR metric — engagement, NPS, promotion rates, DEI dashboards — is reflexive.
Fixes: separate diagnostic from published metrics, rotate measures, pair leading and lagging, treat any bonus-attached metric as compromised.
The point isn't to stop measuring. It's to design measurement that accounts for its own influence.

Year one: engagement survey shows 62%, honest and useful. Leadership makes 'engagement above 75%' a manager KPI. Year three: engagement shows 81%, everyone smiles. Year four: exit interviews reveal engagement is worse than year one. The measurement caused the divergence.

What Soros meant by reflexivity

“The participants' views influence the course of events, and the course of events influences the participants' views. The influence is continuous and circular; that turns it into a feedback loop.”
— George Soros, The Alchemy of Finance (1987)

Soros developed reflexivity to explain financial markets — where the belief a stock is rising causes buying that causes it to rise. Traditional economic theory treats prices as reflecting an independent 'fundamental value'; Soros argued the two co-create each other. The same logic applies wherever humans observe a system they participate in — which is every HR metric ever collected.

Reflexivity across HR metrics

Metric	Original signal	Reflexive deformation	Result
Engagement score	Employee sentiment	Managers coach teams to score higher; leading questions inserted	Score rises, sentiment doesn't
eNPS	Likelihood to recommend	Timing surveys post-good-events	eNPS becomes theatre
Promotion rate by demographic	Fairness signal	Rushed promotions to hit numbers	Real fairness unchanged
Attrition rate	Retention health	Managers avoid firing to protect number	Retention 'improves', performance decays
Time to hire	Recruiter efficiency	Corners cut on quality; loose bar	Mis-hire rate rises
Training completion	Skill development	Click-through completion at 100%	Metric perfect, capability unchanged

Measurement that survives its own use

Six moves against reflexive decay

1
Split diagnostic from published metrics
The metrics you report / attach to bonuses should not be the metrics you use to understand what's happening. Diagnostic stays quiet; published gets gamed.
2
Rotate measures
A metric loses signal once stable and public. Rotating question sets, changing dashboard emphasis, and refreshing what's 'the KPI' resets the reflexive loop.
3
Pair leading and lagging
Engagement (leading) paired with regretted attrition 12 months later (lagging). If lagging follows, the metric is real. If not, it's been gamed.
4
Assume any bonus-attached metric is compromised
If comp depends on it, it's a compensation number, not a measurement number. Track a shadow version.
5
Qualitative triangulation
Numbers plus exit interviews plus skip-level plus manager 360 plus Glassdoor. The network is harder to fool than any single number.
6
Publish the caveat
'This metric is subject to reflexivity — expect drift after 12 months of publication.' Naming it protects it.

Related but not identical

Goodhart's Law says 'when a measure becomes a target, it ceases to be a good measure'. Reflexivity is the deeper claim — it explains why. Participants' knowledge of the metric changes the system, so the metric's relationship to underlying reality shifts.

FAQ

Frequently asked questions

So stop measuring?

No. Stop over-relying on any single number. Use portfolios, rotate them, keep diagnostic separate from published. Measurement is fine; measurement worship is where the harm starts.

How does this apply to AI HR analytics?

Massively. Once an AI model is used for promotion or firing, employee behaviour reshapes to game it — CV keywords, calendar patterns, Slack style. Model drift is a reflexivity problem.

When is a metric compromised?

Twelve months after it's been public and attached to consequences. If it hasn't drifted, either your workforce hasn't figured out the game — or the metric is uselessly loose.

Takeaways

Every HR metric is reflexive — measuring changes the thing measured.
Goodhart's Law is the pattern; reflexivity is the mechanism.
Split diagnostic from published; rotate measures; pair leading with lagging.
Any comp-attached metric is a comp number, not a measurement number.