Cobra Effect

The classic incentive trap: reward the wrong proxy and people optimize the reward while abandoning the actual goal.

The origin

Named for a 1900s Delhi bounty that paid a reward per dead cobra to cut the snake population. People responded rationally to the reward — they bred cobras to claim more bounties — and the cobra population increased. The metric was hit; the goal was lost.

Why it matters

It is a core caution for any leader designing metrics or incentives. A measure becomes a target, the target gets gamed, and the system produces the opposite of what was intended — "humans make systems messy." Within Systems Thinking it is one of the three things that make systems confusing: the others are not knowing which type of system you're in (the Cynefin Framework) and delayed feedback loops.

Live case: Chinese academic fraud (2026)

Per Classmate Geng Shakes Chinese Science (Economist): China ties promotions and funding to publication volume, with explicit cash bonuses + housing allowances for hitting publication targets — and runs mega-labs of hundreds of scholars (10× Western norm) under single PIs. The system optimised exactly as designed: volume up, but ~1 in 10 papers by distinguished scholars are problematic, per Geng Tongxue's estimate. Veteran scientist Rao Yi: China deserves two world records — scientific progress and academic misconduct. Three top-university life-sciences deans lost their leadership posts in April-June 2026.

Source

How To Think SO CLEARLY People Assume You're A Genius (theMITmonk) — the Delhi cobra-bounty example as the canonical incentive trap.
Classmate Geng Shakes Chinese Science (Economist) — Chinese publish-or-perish bonuses as the live case.

Live case: data democratisation as dashboard rollout (2026)

The user's Data Democratisation in Sales — Governed Context Layer, Not Dashboard Access POV reads dashboard-access-as-democratisation as a classic cobra trap: the measure ("dashboards rolled out", "self-service queries answered") becomes the target, and the actual goal (better, shared decisions from consistent meaning) is abandoned. Field access without shared semantics democratises confusion. The corrective primitive — democratise meaning before access via a glossary + skills + learning loop — is from Prukalpa's context-layer essay.

Live case: AI rollout without reallocation mechanisms (2026)

Does AI Adoption Improve Productivity (BOK Issue Note 2026-12) documents the cobra-trap version of corporate AI strategy in a representative national household survey. The measure ("workers using AI") becomes the target — 51.8% of Korean workers now use generative AI for work, ~8× faster diffusion than the internet at the same age — while the actual goal (output growth) is abandoned: at the worker level, the correlation between AI-driven time savings and output growth is zero. The diagnostic is incentive-shaped: in the exception groups (self-employed, professionals, intensive users), performance is directly tied to compensation and time savings do convert to output; in the general population, where the reward for using AI is "use AI" rather than "produce more value," workers cash savings as leisure or stop at "good enough" (Chen et al. 2025). i.e. the trap closes whenever an org optimizes for adoption metrics without designing the reallocation mechanism that routes saved time to high-value activities. See AI Productivity Disconnect.