How Our Supplier Risk Assessment Works
Risk scores are calculated by a deterministic rules engine with published weights. The same inputs always produce the same score. Language models are used only to explain a result in plain language, and they cannot change the number.
Quick answer
We score a supplier across six weighted dimensions: company, quality, compliance, production, supply chain and documentation. Each answer you give maps to one dimension, contributes a weighted risk value, and rolls up into an overall figure out of 100, where higher means more risk. Missing evidence raises the score rather than being treated as neutral. The weights shown below are read directly from the scoring model, not typed by hand.
Scoring framework
Each question belongs to one dimension. Within a dimension, answers carry a risk value between 0 and 100. The dimension score is the weighted mean of its answered questions, and the overall score is the weighted mean of the dimensions below.
22 scored inputs · 6 dimensions · total weight 100.
Risk dimensions and weights
Weights are configurable in our admin and shown here so you can see what actually moves the number.
| Dimension | Weight | Inputs |
|---|---|---|
Company risk Depuis combien de temps ils existent, si l'adresse enregistrée est bien celle de l'usine, et si l'actionnariat est traçable. | 15 | 4 |
Quality risk Si le contrôle qualité est documenté, si un auditeur externe est passé récemment, et si des relevés d'inspection existent. | 20 | 4 |
Compliance risk Registres de conformité sociale, environnementale et produit. Nous demandons ce que vous avez, pas un certificat précis. | 20 | 4 |
Production risk Capacité face à votre volume, recours à la sous-traitance, et expérience sur votre type de produit. | 15 | 3 |
Supply chain risk D'où viennent les matières premières, votre exposition à une source unique, et le réalisme du délai annoncé. | 15 | 3 |
Documentation risk Si la licence commerciale, l'adresse de l'usine, les certificats, les rapports d'audit et les rapports d'essai peuvent vraiment être fournis. | 15 | 4 |
Evidence
The score is only as good as the evidence behind it. Where you tell us you do not know something, we treat it as an open gap and score it accordingly. A supplier can lower its score by producing documents, not by arguing with the model.
Where AI is used, and where it is not
Scores come from the rules engine alone. A language model may summarize or explain a result in plain language, and it may draft the recommended actions. It does not decide the score, and it cannot see the weights as a variable it can adjust.
Limitations
What this score does not tell you.
- · It is not a financial credit assessment. We do not review financial statements unless you ask us to.
- · It does not confirm that equipment exists or that a quality system runs day to day. That needs a site visit or an audit.
- · It is a point-in-time assessment. A score is valid as of the date it was produced.
- · It cannot detect a supplier that deliberately provides false documents, although document cross-checks reduce the gap.
- · Weights are our judgement, not an industry standard. We publish them so you can disagree with us in specific terms.
Update policy
We review weights when we make a material change to the model, and we record the version in use when a score is produced. Methodology changes are listed here with their date.
Changelog
- Current version 1.0 — six dimensions, 22 inputs, weights summing to 100.summing to 100.
Want the checks done rather than scored?
Our verification service confirms registration, site, capability, quality and compliance, and gives you a written report before you pay anything.