How Case Studies Work

What a Case Study Is

A case study applies all 11 models to a specific address with a specific business concept. It answers: should you open here?

The output is not a vague “it depends” — it’s a structured verdict with confidence levels, key risks, and the quantitative reasoning behind each model’s conclusion.

The Pipeline

1. Define the site — address, floor area, proposed lease terms
2. Define the concept — what you’re opening, pricing, target customer, constraints
3. Pull open data — all 31 datasets queried for the specific location and surrounding area
4. Run math models — 11 models execute in parallel, each producing a quantitative output
5. Run AI agent simulation — 10 synthetic personas reason about whether they’d visit
6. Synthesize verdict — model outputs + agent consensus → go/no-go recommendation

What Each Layer Contributes

Layer	Output	Example
Huff Model	Market share probability	”You’d capture 4.2% of the 400m catchment”
Gravity Model	Customer flow by origin zone	”60% of customers will come from within 800m”
Catchment Analysis	Geographic boundary of reach	”Primary catchment: 8,000 people”
Regression	Revenue benchmark	”Comparable sites do ~HKD 1.2M/month”
Site Rating	Composite score 0–100	”Site scores 70/100 — above average”
Geodemographics	Customer segment profile	”Young professional + elderly mix”
Agent Simulation	Qualitative reasoning	”7/10 agents would visit; main barrier: price”
Verdict	Go/No-go with confidence	”Conditional go — with concept adjustment”

First Case Study: In Development

The first full case study — 14 Wa In Fong East, Sheung Wan — is currently in development. It will include:

• Complete model runs for the specific address
• Synthetic population for the Sheung Wan TPU
• Agent simulation results (10 personas, Claude Opus reasoning)
• Sensitivity analysis: how the verdict changes with different business concepts
• Comparison against alternative sites in Central & Western District

Note

Case studies are the “show your work” layer of this system. The models are generic — they apply to any address in Hong Kong. A case study shows exactly what happens when you point all 11 models at one specific location with one specific business concept.

Reproducibility

Every case study is fully reproducible:

• All open data sourced from public APIs (documented in the Data section)
• All model formulas from academic literature (documented in the Models section)
• All agent prompts documented and versioned
• Anyone with the address and business concept can re-run the same analysis