1. What the AI Location Score is
The Locus AI Location Score is a 0–100 rating of how suitable a given address is for a specific business type. It is generated by a large language model that reviews a structured brief of local data — demographics, competition, accessibility, and market saturation — and assigns a score with written rationale.
Scores above 70 suggest strong conditions. Scores between 40 and 70 indicate a workable but competitive or demographically mixed area. Scores below 40 flag material concerns that you should weigh against any qualitative signal you have from the ground.
2. What goes into the score
Every score is produced from a fixed data brief. The model does not browse the web and does not see anything you have not shown it. The brief contains:
- Demographics — local population, primary age group, median household income vs. national average, and age distribution (where available).
- Competitive landscape — direct competitors within the search radius, average competitor rating, quality distribution (strong vs. weak operators), and the people-per-business saturation ratio.
- Local business dynamics — business birth and death rates where the relevant census series is available, combined into a net growth rate.
- Accessibility — nearest public transport stops, Walk Score, and neighbourhood amenity richness from OpenStreetMap where available.
3. Data sources
- Google Places — competitor list, ratings, and review volume within the search radius.
- National census data — UK ONS (MSOA-level) and US Census Bureau ACS (tract-level) for demographics and business demography, matched to the search location.
- OpenStreetMap / Overpass — nearby amenities and transit stops used to compute neighbourhood richness.
- Walk Score API — walkability, transit, and bike scores where the region is covered.
- Mapbox Geocoding — address resolution and place lookup for the search input.
4. How the score is computed
The structured brief is passed to an AI model (currently DeepSeek-Chat with a fallback to OpenAI), which returns a score on a 1–10 internal scale along with a summary, up to three opportunities, and up to three challenges. The score is bounded (clamped to 1–10) and then displayed as 0–100 by multiplying by ten, which keeps the grading coarse and honest — we don't pretend to more precision than we have.
The model is instructed to reference specific numbers from the brief in every insight. The people-per-business ratio and the rating-quality gap are the primary saturation signals; age and income data inform whether the local population matches the target customer profile.
5. Limitations
We are deliberate about what the score is not:
- It is not a prediction of revenue. It is a directional rating of local conditions for a business type.
- It does not include rent, lease terms, or availability. Two addresses a block apart will often score identically and yet have very different take-home economics.
- It does not see foot traffic in real time. Where we cite foot traffic, it comes from Google's popularity signals and is approximate.
- Outside the UK and US, census coverage is uneven. The brief will flag missing data, and the model is instructed to fall back to general knowledge of the area rather than invent numbers.
- Google Places only returns what operators have registered with Google. Quiet independents and very new openings can be missed.
- Large language models can be confidently wrong. Always cross-check the written rationale against your own knowledge of the area before acting on it.
6. Reproducibility
Use the Copy data brief button inside any AI Location Assessment to export the exact structured brief the model was given. You can paste it into any other AI tool to get a second opinion — we publish the inputs rather than hiding them behind the score.
7. Feedback
If a score looks wrong for an address you know well, tell us. We use disputed scores to calibrate the prompt and tighten the data brief. Contact us or use the in-app feedback button.