The number is only as good as the reasoning behind it.
Alternative-data scoring is only useful if a human can defend the outcome. Every Kreedite score is returned with per-signal contribution, direction of influence, reason codes and a plain-language explanation your underwriting team - or a regulator - can read.
- Attribution granularityPer signal
- Reason codes returnedUp to 5
- Plain-language summaryOn every score
A number, plus everything needed to defend it.
A composite score alone is not a decision-support tool. Explainability is what lets a lender turn a model output into a supervised credit decision.
Each signal returns its sub-score, its weight and the resulting contribution to the composite. A composite of 71 is not a shrug - it is a breakdown that adds up.
Each signal is flagged as pushing the composite up or down relative to the segment average - so an underwriter can see immediately whether the applicant is stronger or weaker where.
Up to five ranked reason codes accompany each score. These map to the lender's own adverse-action framework and remain stable across a model version.
A short, human-readable sentence describing the composite's drivers. Meant for underwriter tools and applicant-facing communication where policy permits.
{
"composite": 71,
"recommendation": "approve_for_review",
"signals": [
{ "id": "txn", "score": 78, "weight": 0.31, "contribution": 24.18, "direction": "up" },
{ "id": "comm", "score": 84, "weight": 0.29, "contribution": 24.36, "direction": "up" },
{ "id": "util", "score": 71, "weight": 0.22, "contribution": 15.62, "direction": "flat" },
{ "id": "mob", "score": 62, "weight": 0.18, "contribution": 11.16, "direction": "down" }
],
"reason_codes": [
"R-101 Community history strong",
"R-206 Inflow stability high",
"R-334 Mobile activity below segment"
],
"narrative": "Approve for review. Composite is driven by strong community repayment history and stable transaction inflow. Mobile signal is below segment average and worth a second look."
}A regulator, a credit committee or a rejected applicant can each ask the same question - why? Explainability is what lets a lender answer that question in the same words the score was built with.
Fair to explain. Fair to be explained to.
- Protected attributes are never used as scoring inputs. The engine reads consented financial behavior, not identity characteristics.
- Reason codes are stable within a model version - two identical signal profiles produce identical explanations.
- Every score retains the model version, weight table and sub-score set used, so a past decision can be reproduced years later.
- Applicant-facing narratives are optional and configurable per market, per lender and per regulatory context.
Bring a decision your committee questioned. We will show what a Kreedite attribution would have said about it.