The Fermi Estimator’s Core Playbook
6 PRINCIPLES
1. Error Cancellation (The Central Engine):
If you break a daunting problem into 4–5 independent sub-variables and estimate each within a factor of 3, errors cancel out symmetrically on a logarithmic scale. Expected total error scales with √N, not N.
2. Supply vs. Demand Decomposition:
Pick between measuring the consumer (demand side) or the capacity/producer (supply side). Choose whichever side has fewer unknown variables.
Example: Estimating taxi trips in a city? Demand = (commuters × ride frequency). Supply = (taxis active × trips per shift).
3. Geometric Mean Bounding:
When completely blind on an assumption, establish an unambiguous lower bound ($L$) and an unambiguous upper bound ($H$). The most statistically defensible guess is the geometric mean:
Estimate = √(L × H).
4. Little’s Law for Flow Systems:
For queues, pipelines, transit networks, or things airborne:
L = λ × W (Units in System = Arrival/Creation Rate × Average Residence Time).
5. Volumetric Packing & Ratio Bounds:
For objects inside spaces:
Count = (Usable Container Volume × Packing Density) / Object Volume.
Random close packing of spheres fills ~64% of volume; rectangular items fill ~75-80%.
6. Strict One Significant Figure:
Round mid-flight ($365 \to 400$; $\pi \to 3$; $68\% \to 70\%$). Extra digits create false precision and slow mental processing.
Universal Anchor Table
3.3 × 10⁸
US Population (~120M households, ~2.7 people/hh)
8.0 × 10⁹
World Population (~8 Billion)
π × 10⁷ s
Seconds per year (3.15 × 10⁷ s ≈ 31.5M s)
1,000 kg / m³
Density of water (1 g/cm³ = 1 metric ton/m³)
100 Watts
Human resting metabolic rate (~2,000 kcal/day)
~280 Million
Registered highway vehicles in the United States
~12,000 miles
Average annual vehicle miles driven in US
0.64 (64%)
Random packing fraction of rigid spheres