The Streetlight Effect for HR: Why We Measure What's Easy and Ignore What Matters

60-Second Summary

Streetlight Effect: bias toward measuring what's easy, not what's important.
HR is structurally exposed: it inherits HRIS-easy numbers and ignores latent, important ones.
Result: dashboards full of vanity metrics; the questions the CEO actually asks have no answer.
Fix: start with the question, not the data. Build the metric backwards from decision.
For tech: it's the analytics anti-pattern of optimizing what's logged instead of what's valuable.

An HR team built a 14-tab dashboard: headcount, time-to-fill, training hours, eNPS, attrition, diversity ratios. The CEO opened it once a quarter and asked a question it couldn't answer: 'Which of the people I most don't want to lose are most likely to leave in the next 6 months?' That question requires regrettable-attrition prediction. The dashboard had none of it — because it's hard.

Where the bias comes from

“A man searches for his keys under a streetlight, though he lost them in the dark. Asked why, he says: 'because this is where the light is.'”
— Behavioral-science parable, sometimes attributed to Nasruddin

HR inherits an HRIS schema designed for compliance and payroll. The easy metrics — headcount, tenure, completion, time-to-X — are pre-built. The hard metrics — skill transfer, belonging, regrettable attrition, network health, decision quality — require new instrumentation. Under deadline pressure, dashboards default to the streetlight.

HR's brightest streetlights (and what's actually in the dark)

What's measured (lit)	What matters (dark)
Training hours	Skill transfer 30/60/90 days later
eNPS / engagement	Belonging, voice, psychological safety
Time-to-fill	Quality + 12-mo retention of hire
Aggregate attrition %	Regrettable attrition by performance tier
Diversity headcount ratios	Inclusion experience by intersection
Manager 1:1 completion	Manager coaching quality (sampled)

Working in the dark

Start with the decision: 'What will we do differently if this number is high vs. low?' If no answer, don't build the metric.
Inventory the top 5 questions the CEO actually asks. Build instrumentation for those, in priority order.
Use proxies cautiously: 'manager coaching quality' is hard, but skip-level sentiment + IC growth velocity is a defensible composite.
Sampling beats census for hard metrics: 80 high-quality interviews on belonging beat 4,000 survey responses.
Set a 'dark/light ratio': at least 30% of your reported metrics should be ones that didn't exist 12 months ago.

Diagnostic

Audit your last quarterly HR report. Count metrics. How many directly answer a real executive decision? In most companies the answer is <30%. The rest is streetlight light.

Tech analog

Analytics teams know this trap: you instrument what's easy in your event pipeline, then optimize what you instrumented, even when the lever that matters lives off-platform. The discipline of 'metrics from decision, not from data' is what separates senior analytics work from junior. The same applies to People Analytics.

Takeaways

Easy metrics dominate because they're easy, not because they matter.
Build from the decision backwards. If no decision changes, kill the metric.
Spend 30% of People Analytics effort instrumenting things that didn't exist a year ago.

References

Freedman — Why Scientific Studies Are So Often Wrong: The Streetlight Effect — Discover Magazine, 2010
Kahneman — Thinking, Fast and Slow (availability heuristic) — FSG, 2011
Google re:Work — People Analytics — Google

Written by Pawan Joshi.Sources cited inline.

First published 9 Jun 2026See site changelog →