How Location Data Helps Businesses Pick the Right Spot: Lessons from Building PetaAI

Ask a founder how they chose their first outlet and you will often hear some version of the same story: "We drove past it, the street looked busy, and the rent was manageable." Sometimes that works. Often it does not, and the cost of a wrong location is rarely just the rent. It is the fit-out, the staff you hired, the inventory you stocked and the months you spent waiting for customers who never came.

At CodeWorks we built PetaAI's Nemu, a location intelligence tool that scores and ranks candidate sites for franchise rollout, retail expansion and rental-property investment. This post shares what we learned along the way: why gut feel is not enough, which data signals actually matter, what makes this hard in Indonesia, and how any business can start applying the same thinking.

Why gut-feel site selection fails

Intuition is not useless. Experienced operators notice things a spreadsheet misses. The problem is that intuition is hard to compare, hard to repeat and easy to bias.

• A single visit is a single sample. A street that is packed on Saturday afternoon may be empty on Tuesday morning.
• Competition is easy to undercount. You see the two coffee shops across the road, not the five within a ten-minute ride.
• It does not scale. Comparing three spots by feel is possible. Comparing dozens across several cities is not.
• Decisions are hard to explain. When an investor or franchise partner asks "why here?", "it felt right" is not a convincing answer.

This is the gap PetaAI set out to close. Franchise operators, retail chains and property investors had no fast, data-driven way to compare candidate sites across factors like demand, competition, demographics and accessibility.

The data signals that matter

No single number tells you whether a location will work. Good site selection combines several signals and weighs them according to the type of business.

1. Demand and foot traffic proxies

Direct foot traffic counts are rarely available for every street you are considering. In practice you look for proxies: the density of offices, schools, campuses, residential areas and transit points nearby. A kos-kosan near a university and a lunch spot near an office cluster rely on very different kinds of demand.

2. Competitor density

Competition cuts both ways. A cluster of food stalls can signal a proven dining destination, or a saturated market. What matters is the balance between demand and how many players are already serving it.

3. Demographics

Who lives and works in the area? Age profile, household type and spending power shape what people buy and how much they pay.

4. Accessibility

Can customers actually reach you? Road access, parking, proximity to main roads and public transport, and how easy the spot is for ride-hailing and delivery drivers all affect daily performance.

5. Points of interest

Malls, markets, mosques and churches, hospitals, gyms and campuses act as anchors that create regular reasons for people to be nearby. The mix of nearby points of interest often says more about an area than any single statistic.

What makes location intelligence hard in Indonesia

Building a tool like Nemu is not only a modelling problem. It is first and foremost a data problem, and Indonesia adds its own layer of difficulty.

• Inconsistent addresses. The same place can be written many ways, with references to gang, RT/RW, landmarks or old street names. Turning that into reliable coordinates (geocoding) takes real effort.
• Uneven map coverage. Data is usually richer in big cities than in second- and third-tier cities, which are exactly where many franchises want to expand next.
• Fast-changing areas. New housing, toll roads and commercial strips change a neighbourhood quickly, so data ages fast.
• Mixed sources. Useful signals come from different places in different formats, and they must be cleaned and aligned before they can be compared.

For PetaAI, the data sources and the approach to cleaning and geocoding them were [CONFIRM: describe PetaAI's data sources and geocoding/cleaning approach that can be shared publicly].

Product and engineering lessons from building PetaAI

Rank, do not just map

A map full of colourful layers looks impressive but still leaves the user to do the thinking. Nemu takes a different approach: it ingests location inputs and returns a scored shortlist. In its site-ranking module, Nemu scores 100 candidate sites and gives each one a tier label, so users can focus on the top options instead of staring at a heatmap.

Explain every score

A number without a reason is hard to trust. That is why every ranked site in Nemu comes with a written justification. Business owners and investors need to understand why a site scored well before they put money on the line, and explanations make the result something they can discuss with partners.

Different businesses need different weights

What makes a good spot for a franchise outlet is not what makes a good kos-kosan. Nemu uses a weighted ML model with adjustable algorithm parameters per use case, and its analysis is tuned specifically for retail expansion and rental-property investment. One fixed formula would have served nobody well.

Design for the people making the decision

PetaAI serves franchise investors, property developers and franchise business owners. Each asks slightly different questions, so the results are presented through an interactive dashboard rather than a raw data export.

Make it easy to plug in

Larger organisations often want scores inside their own systems, not in yet another app. PetaAI's enterprise plan exposes a public REST API so the scoring engine can be integrated elsewhere.

The goal of location intelligence is not to replace the decision-maker. It is to give them a shorter, better-explained list to decide from.

How your business can apply location data

You do not need a full platform on day one to make better location decisions. Here is a practical way to start.

1. Define what success looks like. For F&B it may be weekday lunch traffic; for retail, weekend family visits; for a franchise, consistent performance across many outlets.
2. Study your best and worst existing locations. List what is around them: competitors, anchors, access, the type of neighbourhood. Patterns usually appear quickly.
3. Pick a handful of signals and weight them. Start with demand proxies, competitor density, demographics, accessibility and nearby points of interest, then adjust weights for your business model.
4. Score candidates consistently. Use the same criteria for every site so comparisons are fair, and write down why each one scored the way it did.
5. Close the loop. After opening, compare actual performance against the score and refine your weights.

F&B brands can use this to avoid oversaturated food streets. Retailers can find underserved neighbourhoods. Franchise owners can give partners a transparent, repeatable reason for every new outlet. [CONFIRM: optional short example of how a PetaAI user applied Nemu, if the team can share one publicly]

---

Building a data or AI product?

PetaAI showed us that the hardest part of a data product is rarely the algorithm. It is turning messy, real-world data into answers people trust and act on. If you are exploring a location intelligence tool, a scoring engine or another data or AI product, talk to the CodeWorks team.