Huff Probability Model

What It Does

The Huff model answers: given our restaurant’s size and location, what fraction of nearby diners will choose us over competitors?

It’s based on the core idea that a customer’s probability of choosing any restaurant is proportional to its attractiveness and inversely proportional to the distance to it — summed across all options.

Core Formula

P(choose us) = Aᵢ / (Aᵢ + Σ Aⱼ)

Where attraction A = size × distance^(-β). Our attraction is computed at 10 m (effectively zero distance). Competitor attractions use actual distances from the API.

Current Implementation

Attraction Calculation

Our attraction:

sizeScore = min(10, (floorArea / 800) × 7 + 3)
uniqueBonus = 2.5 if cuisines ≤ 1 AND competitors < 5, else 0
ourAttraction = (sizeScore + uniqueBonus) × 10^(-β)

Competitor attraction (nearest 20):

attraction = 5.0 × max(distance, 50)^(-β)

Remaining competitors beyond nearest 20: Modeled at 1.3 × avg distance of nearest 20, each with attraction 5.0.

Capture Probability

captureProb = (ourAttraction / totalAttraction) × 100

Then clamped by maxCapture and floored at 0.5% (previously 5%):

Competitor count	Max capture
≤ 3	55%
≤ 10	40%
≤ 30	25%
≤ 100	10%
≤ 300	5%
300+	3%

Output precision: 2 decimal places (e.g. 12.47%).

Beta (Distance Decay)

Beta is set by calcBeta() before Huff runs:

Service model / price	Beta	Decay function
Delivery	0.3	Exponential
Takeaway	0.5	Exponential
High-end HKD 200+	2.5	Power
Premium HKD 100-200	1.5	Power
Budget / Mid (dine-in)	0.8	Exponential

Data Sources

Floor area: user input (sqft)
Competitors: FEHD licence database, filtered within 800 m, nearest 20 used directly
Beta: derived from service model and price band

Interpretation

captureProb	Meaning
40–55%	Near-monopoly — very few nearby competitors
20–40%	Strong position in a competitive market
10–20%	Average for a busy urban district
< 10%	Highly competitive environment — differentiation critical

Known Limitations

Uses FEHD F&B licences as a proxy for competitor count — excludes unlicensed stalls and hawkers
All competitors assigned the same baseline attraction (5.0) regardless of size or quality
Nearest 20 used directly; remainder modeled at a flat 1.3× average distance
No time-of-day variation (lunch competition may differ from dinner)

Changelog

Date	Change
2026-03-25	Floor lowered from 5% → 0.5%; precision changed to 2 decimal places; granular maxCapture tiers added (55/40/25/10/5/3 based on competitor count)
2026-02-10	Initial implementation — single 50% cap, 5% floor