Policy overlay
The signal enters as a layer on your eligibility framework. You set the weight and the cut-off.
Risk methodology
AI does not replace the funder’s credit policy. It structures thin-file evidence, estimates a probability of performance and delivers reason codes a Risk team can defend.
Principle
In unbanked populations the problem is not absence of risk: it is absence of a file. We work so underwriting starts higher up the stack — less noise, fewer gaps, less manual reconstruction.
The signal enters as a layer on your eligibility framework. You set the weight and the cut-off.
Knock-outs and an eligibility band before analyst or bureau capacity is consumed.
Disciplined comparison against your live scorecard once vintages allow it.
Vectors
We do not publish features or weights. We do declare the conceptual family a credit committee recognises.
Subject consistency and file quality. Without usable identity there is no origination.
Territorial stability and lower operational flight incentive. A discipline proxy, not a real guarantee.
Value-generation patterns. Attested activity, not banked cash flow alone.
Punctuality on micro-obligations. Evidence of willingness-to-pay, not an income statement.
Analytical cycle
Supervised machine learning on performance outcomes. The institution’s hard rules always sit above the model.
We do not force a single score. Rooted rural context is not read like high urban mobility. Clustering before estimation.
Supervised models estimate P(C) / a working PD from consistency and conduct. The output is a trajectory, not a snapshot. It is not an IRB rating or a loss promise.
The estimate is crossed with the institution’s exclusions, ticket and threshold. Puelo Cloud delivers the file. The institution decides.
With live outcomes we ask what drove performance: capacity, commitment or shock. Champion–challenger when the sample allows it.
Risk pack
A standard package, with an origination log and minimised PII. Ready for policy. Not a sales lead.
Risk band. High-level reason codes. Eligibility flags (proceed / stop / review). Non-binding product-fit suggestion. Log of consents and rules applied.
Every pre-qualification carries reasons defensible before Model Audit and before the desk analyst. AI prioritises; it does not hide.
Governance
Deep identity remains encapsulated. The marketplace operates on an anonymised profile and synthetic risk variables. Model detail is discussed under NDA, not on a landing page.