GeoInsightsSite Selection
Riyadh · M3 alpha

Site Selection · M3 alpha

Drop a pin anywhere in Riyadh and get a scored, explained, verifiable result.

A candidate site is measured against a composed H3 res-9 baseline of the city, over a real drive-time catchment, under a weight profile you can read. What comes back is a Tier 1 suitability index on 0-100 with its whole derivation attached — and, where the data does not exist, an explicit refusal instead of a number.

The baseline

Baseline cells
42,682
H3 res-9, the only resolution scored over.
Usable factors
14
Columns with real spatial variation, weightable today.
Flat, not weightable
5
Values exist and are identical everywhere. A weight on one is refused.
No data for this region
4
Excluded from the score; the surviving weights are renormalised.

Two things this engine will not do

Both are visible in the product, on purpose. A tool that quietly papered over either would produce prettier screens and worse decisions.

It will not weight a factor that does not vary

The age-structure columns are a national age pyramid applied through a uniform share: every hex in Riyadh carries the same composition. Weighting one ranks candidates on a value that is identical everywhere — the ranking sums correctly, renders beautifully, and means nothing. Ask for it and the engine returns a written refusal naming the factor, why it is flat, and what would unlock it. It reads the per-cell flag at request time rather than holding a list, so the day a district layer lands the factor becomes weightable with no code change.

It will not read a missing value as zero

Affluence, spend potential, land value, retail floorspace, households and daytime population have no source for Riyadh yet. Scoring them as zero would assert that a district has no wealth — indistinguishable from a measurement, and wrong in a way no aggregate check catches. They are excluded, the surviving weights are renormalised so scores stay on 0-100, and every scored result reports what share of its intended weight actually reached it. A score computed from 63 % of its profile says so, next to the number.

The path, end to end

  1. 1

    Drop a pin

    Arm placement in the Map Studio and click anywhere in Riyadh, or add a site by address or bulk upload in Candidates.

    Open the Map Studio
  2. 2

    The trade area routes

    Valhalla returns 5 / 10 / 15 / 20-minute drive-time contours from the snapped point. If the router cannot answer, nothing is drawn — there is no circle fallback, because a disc of the same reach overstates a Riyadh catchment four- to six-fold.

    Open the Map Studio
  3. 3

    The score computes

    The baseline cells under the contour are rolled up into one feature vector, area-weighted, and run through the weighted MCDA under the chosen profile. Tier 1 of the model ladder — an index on 0-100, never a currency.

    Score from Candidates
  4. 4

    The explanation renders

    Every factor with its raw value, its rank in the regional reference, the weight as authored, the weight as applied and the points that came out. The arithmetic reconstructs without the server.

    See a breakdown
  5. 5

    Compare

    Up to six candidates aligned factor by factor from their stored scores. Nothing is recomputed — comparing means holding the conditions still — and the view refuses to rank candidates scored under different weights, vintages or contours.

    Open Compare

The screens

Get set up

Where you are, what to do next, and what each step unlocks — with the locked rungs of the model ladder and their exact conditions on screen. Also states which weight profiles this region can and cannot score.

Map Studio

Drop a pin anywhere in Riyadh. The drive-time trade area routes over the res-9 baseline, and the score lands on the pin with its weight coverage.

Candidates

Add sites by map click, address or bulk upload, move them through the pipeline, and score them in batches.

Competitors

Confirm the competitors open data can see, correct the ones it got wrong, and add the ones it is missing. Corrections stay private to your tenant.

Your data

Upload your store network and sales history — the inputs every model above the suitability index needs. Customer home locations are aggregated to H3 res 8 with k-anonymity enforced before anything is stored.

Compare

Up to six candidates aligned factor by factor from their stored scores, with every unavailable factor marked in every column.

Methodology

public

Every source, its licence, what it was checked against, and what this product will not claim. Public — no sign-in.