📖 Plain-English Guide

How the Analyzer Works

Our tool runs 11 different models on every address you enter. Each one looks at the same location from a different angle. Here's what they do, explained without jargon.

1 Huff Probability Model

Imagine every nearby restaurant is a magnet, and hungry people are iron filings. Bigger restaurants with more to offer pull harder, but distance weakens the pull. This model figures out what share of nearby diners your restaurant would attract compared to the competition.

TL;DR — "What percentage of local diners would pick you over everyone else?"

2 Gravity Model

Like gravity in physics: bigger populations nearby create stronger pull, but the farther away they are, the weaker the pull. This model estimates how many people from each distance zone (5-min walk, 10-min walk, transit, destination) would actually make the trip to your door.

TL;DR — "How many people from each surrounding area would realistically visit?"

3 Catchment Area

This draws invisible circles around your location — a 5-minute walk, a 10-minute walk, and a 15-minute transit trip — then counts how many people live inside each circle and how many competitors you'd be sharing them with. Simple but powerful.

TL;DR — "How many people and competitors are within walking and transit distance?"

4 Distance Decay

People are lazy (in the best way). The farther they have to travel, the less likely they are to bother. This model figures out exactly how sharply interest drops off with distance for your type of restaurant. A delivery kitchen? Distance barely matters. A premium sit-down place? People won't walk far for it when there are closer options.

TL;DR — "How quickly do you lose potential customers as distance increases?"

5 Site Rating Scorecard

Think of it as a report card for the location itself. We grade five things — market size, competition level, transport access, safety, and rent value — each weighted by importance, then combine them into a single score out of 100. It's a quick way to compare different locations.

TL;DR — "If this location sat an exam, what would its grade be?"

6 Revenue Regression

This is the "show me the money" model. Using a formula calibrated to Hong Kong F&B benchmarks, it plugs in your floor area, competition level, local spending power, and transport access to estimate annual revenue. Think of it as a very informed back-of-the-envelope calculation.

TL;DR — "Based on the numbers, roughly how much revenue could this location generate?"

7 Geodemographic Segmentation

Not all neighbourhoods are the same. This model looks at the local population — their incomes, ages, household sizes, and cultural mix — and assigns the area a profile like "Young Professional Hub" or "Family Neighbourhood." Each profile comes with typical dining habits and spending patterns, so you know what kind of customers you'd be serving.

TL;DR — "What kind of neighbourhood is this, and what do the locals eat?"

8 Agent-Based Model (ABM)

We create 100 virtual people — office workers, residents, tourists, foodies, and families — each with their own preferences, budgets, and willingness to walk. Then we let them "decide" whether they'd visit your restaurant. It's like running a tiny focus group, but with maths instead of opinions.

TL;DR — "If 100 different people walked past, how many would come in?"

9 LLM-Enhanced ABM

This takes the agent simulation one step further. Instead of simple maths, it generates a prompt you could feed to an AI language model (like Claude) to get nuanced, written reasoning about whether each persona would visit. Right now we show a structured prompt and placeholder analysis — plug in an API key for the full experience.

TL;DR — "What would an AI think about each type of customer visiting this spot?"

10 Microsimulation

We synthesize 1,000 virtual residents based on the district's real demographics — matching the local age spread, incomes, and household sizes. Then we simulate a full week of their dining decisions: who eats breakfast out, who grabs lunch, who goes for dinner, who's up late. The result is a realistic picture of weekly demand and a range of possible monthly revenues.

TL;DR — "If we modelled 1,000 locals' eating habits, how much demand would there be?"

11 Location-Allocation (P-Median)

This model asks a blunt question: "Is this the best spot, or would somewhere nearby be better?" It distributes demand points across the district, then checks whether your chosen location minimises the average distance to all that demand. If the theoretical optimal spot is right where you are, great. If it's 500 metres away, that's worth knowing.

TL;DR — "Is your address the best possible location in this area, or is somewhere else better?"

What This Tool Cannot Tell You

It can't predict the future. Markets change, new competitors open, pandemics happen. These models use today's data for today's snapshot.
It doesn't know your cooking. Food quality, service, marketing, and brand matter enormously. No model can measure those.
Foot traffic isn't guaranteed revenue. The models estimate potential demand, not how much you'll actually capture. Execution is everything.
Government data has gaps. FEHD restaurant licences don't cover every food outlet (e.g., unlicensed hawkers, food courts inside malls). Our competitor counts are estimates.
Rent negotiations are unique. Our rental data is district-level averages. Your actual rent depends on the landlord, the floor, the lease terms, and your negotiation skills.
It can't replace visiting in person. Walk the street at lunch. Count heads. Talk to neighbouring shopkeepers. Data complements boots-on-the-ground — it doesn't replace it.
Demographic segments are simplified. Real people don't fit neatly into five boxes. These profiles are useful starting points, not definitive portraits.
The LLM model needs an API key. The AI-enhanced persona analysis shows a template without a live API key. Actual LLM responses would be richer and more nuanced.

Ready to try it on your address?

Open the Analyzer →