The Map is not the territory
Models, metrics and narratives are representations. They are useful abstractions, not reality. Treat them as tools, validate them against the world, and update when they drift.
author
Popularised by Alfred Korzybski
Model type
About
Korzybski’s line reminds us that all models omit detail. Abstraction lets us reason and decide, but the omissions and measurement protocols create model error. Confusing the model with the thing leads to brittle strategies and perverse incentives.
How it works
Abstraction – maps simplify to highlight what matters; every simplification hides something.
Fit to purpose – usefulness beats realism; a subway map distorts geography to aid routing.
Drift – territories change faster than maps; assumptions decay without remeasurement.
Reflexivity – when people optimise to the map (metric/OKR), behaviour changes the territory.
Fit to purpose – usefulness beats realism; a subway map distorts geography to aid routing.
Drift – territories change faster than maps; assumptions decay without remeasurement.
Reflexivity – when people optimise to the map (metric/OKR), behaviour changes the territory.
Use cases
Strategy – distinguish narrative from operating reality; validate theses with ground checks.
Metrics & incentives – design guardrails against Goodhart effects when a metric becomes a target.
Forecasting & models – stress test model sensitivity to omitted variables and data revisions.
Commercial due diligence – reconcile deck KPIs with customer behaviour, cohort curves and cash evidence.
Data & AI – monitor dataset shift and re-train when production data diverge from training assumptions.
Metrics & incentives – design guardrails against Goodhart effects when a metric becomes a target.
Forecasting & models – stress test model sensitivity to omitted variables and data revisions.
Commercial due diligence – reconcile deck KPIs with customer behaviour, cohort curves and cash evidence.
Data & AI – monitor dataset shift and re-train when production data diverge from training assumptions.
How to apply
Name the map – what representation are you using (metric, heatmap, LTV model, narrative)?
List key omissions – what does it ignore (data cuts, edge cases, latency, incentives)?
Ground-truth – sample real cases; compare predictions to observed outcomes.
Instrument drift – set triggers to review when inputs/behaviours/regulations change.
Design safeguards – pair metrics, add qualitative checks, and cap incentives to reduce gaming.
List key omissions – what does it ignore (data cuts, edge cases, latency, incentives)?
Ground-truth – sample real cases; compare predictions to observed outcomes.
Instrument drift – set triggers to review when inputs/behaviours/regulations change.
Design safeguards – pair metrics, add qualitative checks, and cap incentives to reduce gaming.
pitfalls and cautions
Reification – treating the model as the thing; arguing from the spreadsheet, not the operations.
Over-precision – spurious decimals signal false certainty; show error bars and ranges.
Goodhart’s law – when a metric becomes a target, it stops measuring what it did.
One-map tyranny – use multiple representations (customer stories, cohorts, unit economics) before deciding.
Over-precision – spurious decimals signal false certainty; show error bars and ranges.
Goodhart’s law – when a metric becomes a target, it stops measuring what it did.
One-map tyranny – use multiple representations (customer stories, cohorts, unit economics) before deciding.