Methodology

Four Layers

The system takes open data and your business concept, runs them through mathematical models, then lets AI agents interpret the results. Four layers, each building on the previous one.

Layer	What	Source
1. Open Data	31 datasets: competition, demographics, transport, rent, weather, live feeds	data.gov.hk (3,712 datasets, free)
2. Your Concept	Business type, pricing, target customer, physical constraints	The client
3. Math Models	10 models from retail geography, spatial analysis, and agent simulation	Academic literature (see below)
4. AI Agents	LLM-powered synthetic personas that reason about model outputs	Claude Opus + demographic data

Tip

Layer 2 is what makes this a product, not a report. The same address gets completely different predictions depending on what you plan to open there. A ramen bar and a cha chaan teng on the same block produce completely different numbers.

Data Flow: Datasets → Models

This diagram shows exactly which open datasets feed which mathematical models. Click any node to open its documentation page.

flowchart LR
  subgraph food["Food"]
    FEHD["Restaurant Licences"]
    FF["Food Factories"]
    HM["Hawker Markets"]
  end

  subgraph demo["Demographics"]
    POP["Population"]
    INC["Household Income"]
    HS["Household Size"]
    AGE["Age Distribution"]
    ETH["Ethnicity"]
    RR["Restaurant Receipts"]
  end

  subgraph transport["Transport"]
    MTR["MTR Stations"]
    RIDE["MTR Ridership"]
    FARE["MTR Fares"]
    KMB["KMB Routes"]
    KMBS["KMB Stops"]
    CTB["CTB Routes"]
    TD["Traffic Detectors"]
  end

  subgraph prop["Property"]
    RENT["Rental Indices"]
    PP["Property Prices"]
  end

  subgraph safety["Safety"]
    CS["Crime Stats"]
    CD["Crime by Type"]
  end

  subgraph wx["Weather"]
    WC["Current Weather"]
    WR["Rainfall"]
    AQ["Air Quality"]
  end

  subgraph live["Live Proxies"]
    AE["A and E Wait"]
    BQ["Border Queues"]
    PV["Parking Vacancy"]
  end

  subgraph fac["Facilities"]
    SCH["Schools"]
    BLD["Buildings"]
  end

  subgraph econ["Economy"]
    CPI["CPI"]
    EMP["Employment"]
  end

  subgraph company["Your Concept"]
    BP["Business Profile"]
    PR["Pricing Strategy"]
    TC["Target Customer"]
    CON["Constraints"]
  end

  subgraph models["Math Models"]
    HUFF["Huff Probability"]
    GRAV["Gravity Model"]
    CATCH["Catchment Area"]
    DD["Distance Decay"]
    GEO["Geodemographics"]
    REG["Regression"]
    SR["Site Rating"]
    LA["Location-Allocation"]
    MICRO["Microsimulation"]
    ABM["ABM Theory"]
  end

  subgraph agents["AI Agents"]
    LLM["LLM Agent Simulation"]
  end

  VERDICT["Verdict"]

  FEHD --> HUFF
  FEHD --> GRAV
  FEHD --> LLM
  FEHD --> LA
  FEHD --> REG
  FEHD --> SR
  HM --> HUFF

  POP --> HUFF
  POP --> GRAV
  POP --> CATCH
  POP --> LA
  POP --> MICRO
  POP --> REG
  POP --> ABM
  INC --> HUFF
  INC --> GRAV
  INC --> GEO
  INC --> LLM
  INC --> MICRO
  INC --> REG
  INC --> SR
  HS --> GEO
  HS --> MICRO
  AGE --> GEO
  AGE --> MICRO
  ETH --> GEO
  ETH --> LLM
  RR --> MICRO
  RR --> REG

  MTR --> HUFF
  MTR --> GRAV
  MTR --> CATCH
  MTR --> DD
  MTR --> LA
  RIDE --> GRAV
  RIDE --> ABM
  FARE --> DD
  KMB --> CATCH
  KMB --> DD
  KMBS --> CATCH
  CTB --> CATCH
  TD --> DD
  TD --> ABM
  TD --> SR

  RENT --> LA
  RENT --> REG
  RENT --> SR
  PP --> SR

  CS --> SR
  CD --> GEO

  WC --> ABM
  WR --> ABM
  AQ --> SR

  AE --> ABM
  BQ --> LLM
  PV --> CATCH
  PV --> SR

  SCH --> CATCH
  SCH --> GEO
  SCH --> SR
  BLD --> REG
  BLD --> SR

  CPI --> ABM
  CPI --> MICRO
  CPI --> REG
  EMP --> GRAV
  EMP --> GEO
  EMP --> MICRO
  EMP --> REG

  BP --> HUFF
  BP --> GRAV
  BP --> REG
  BP --> SR
  BP --> MICRO
  BP --> ABM
  BP --> LLM
  PR --> HUFF
  PR --> DD
  PR --> REG
  PR --> MICRO
  PR --> LLM
  TC --> GEO
  TC --> MICRO
  TC --> ABM
  TC --> LLM
  CON --> LA
  CON --> SR
  CON --> REG

  HUFF --> VERDICT
  GRAV --> VERDICT
  CATCH --> VERDICT
  DD --> VERDICT
  GEO --> VERDICT
  REG --> VERDICT
  SR --> VERDICT
  LA --> VERDICT
  MICRO --> VERDICT
  ABM --> VERDICT

  HUFF --> LLM
  GRAV --> LLM
  CATCH --> LLM
  SR --> LLM
  MICRO --> LLM

  LLM --> VERDICT

  click FEHD "/data/food-restaurant-licences/"
  click FF "/data/food-factory-licences/"
  click HM "/data/food-hawker-markets/"
  click POP "/data/demo-population/"
  click INC "/data/demo-household-income/"
  click HS "/data/demo-household-size/"
  click AGE "/data/demo-age-distribution/"
  click ETH "/data/demo-ethnicity/"
  click RR "/data/demo-restaurant-receipts/"
  click MTR "/data/transport-mtr-stations/"
  click RIDE "/data/transport-mtr-ridership/"
  click FARE "/data/transport-mtr-fares/"
  click KMB "/data/transport-kmb-routes/"
  click KMBS "/data/transport-kmb-stops/"
  click CTB "/data/transport-ctb-routes/"
  click TD "/data/transport-traffic-detectors/"
  click RENT "/data/prop-rental-indices/"
  click PP "/data/prop-property-prices/"
  click CS "/data/safety-crime-stats/"
  click CD "/data/safety-crime-detail/"
  click WC "/data/weather-current/"
  click WR "/data/weather-rainfall/"
  click AQ "/data/weather-air-quality/"
  click AE "/data/live-aed-wait/"
  click BQ "/data/live-border-queues/"
  click PV "/data/live-parking-vacancy/"
  click SCH "/data/facility-schools/"
  click BLD "/data/facility-buildings/"
  click CPI "/data/econ-cpi/"
  click EMP "/data/econ-employment/"
  click BP "/company/business-profile/"
  click PR "/company/pricing/"
  click TC "/company/target-customer/"
  click CON "/company/constraints/"
  click HUFF "/models/huff/"
  click GRAV "/models/gravity/"
  click CATCH "/models/catchment/"
  click DD "/models/distance-decay/"
  click GEO "/models/geodemographics/"
  click REG "/models/regression/"
  click SR "/models/site-rating/"
  click LA "/models/location-allocation/"
  click MICRO "/models/microsimulation/"
  click ABM "/models/abm-theory/"
  click LLM "/models/llm-abm/"

Note

Every node in the diagram is clickable. Tap a dataset to see its API schema, or a model to see its formula and data requirements.

Academic Foundation

Every model in this system comes from peer-reviewed academic work. No proprietary algorithms, no black boxes.

Book	Authors	Year	Models
Retail Geography and Intelligent Network Planning	Birkin, Clarke, Clarke and Wilson	2017	Huff, Gravity, Regression, Site Rating, Location-Allocation, Microsimulation
Geospatial Analysis (6th ed.)	de Smith, Goodchild and Longley	2021	Catchment, Distance Decay, Geodemographics
An Introduction to Agent-Based Modeling	Wilensky and Rand (MIT Press)	2015	ABM Theory
Original extension	This project	2026	LLM-Powered Agents

The LLM agent extension replaces classical ABM rule-sets with Claude Opus reasoning. Same spatial data, same model outputs, but agents that can explain why they chose a restaurant in natural language.

Layer Interaction

The four layers form a pipeline where each step builds on the previous:

1. Open Data (31 datasets) provides the objective reality of Hong Kong: where people live, work, travel, eat, and spend. We can approximate the demographic mix at any location at any hour
2. Your Concept (4 dimensions) defines the specific business: what you’re opening, at what price, for whom. This calibrates every formula differently
3. Math Models (10 models) take both inputs and produce quantitative predictions: market share, catchment population, site score, revenue estimate
4. AI Agents (10 synthetic personas) receive model outputs and simulate actual customer decisions: would they walk in, or keep walking? They surface barriers and opportunities that pure math misses

The final Verdict synthesizes all model outputs and agent consensus into a risk assessment with confidence intervals.

Reproducibility

Every data point comes from a public API. Every formula is textbook. Every agent prompt is documented. Anyone can reproduce this analysis for any address in Hong Kong.

What Makes This Different

Traditional Approach	This System
Consultant picks 3-5 “key factors” subjectively	31 datasets, 11 models, exhaustive coverage
Excel model with hardcoded assumptions	Parameterized models that adapt to company input
One-time report, stale in months	Live data feeds update in real-time
”Trust me, I have been in the industry”	Every number traceable to a public API endpoint
Generic advice	10 AI agents simulate actual customer decision-making