Credit decisioning

A recommendation your officer can defend

Khepee scores an application and recommends an outcome. A person makes the decision. That boundary is deliberate — an automated rejection nobody can explain is a regulatory and reputational liability.

recommendation

The scorecard

Rules you set, applied identically every time

The default engine is a transparent rule-based scorecard. You configure the factors and weights; it applies them consistently to every application, which is precisely what a manual process cannot promise.

  • Income multiple against the requested instalment
  • Existing obligations, where the borrower declares or your data shows them
  • KYC quality and document consistency
  • Repayment history with your institution on the platform

Explainability

The score arrives with its reasons

A number on its own tells an officer nothing they can act on or explain. Khepee surfaces the factors that moved the score and how far each one moved it.

  • Factor-level contributions, not an opaque total
  • Thresholds visible, so a borderline case is recognisable as one
  • The recommendation is advisory and labelled as such
  • What the audit trail records is the human decision and its stated reason

Extensibility

Room for bureau data and models, when you want them

The scoring engine sits behind an interface. Adding a credit bureau feed or your own model means implementing that interface — the origination workflow, the audit trail and the officer console do not change.

  • A single `ScoringEngine` seam, versioned so old decisions stay interpretable
  • Bureau integration behind the same vendor-interface discipline as KYC and payments
  • A model can run in shadow mode against live applications before it influences anything
  • The score version is recorded with the decision, so history remains explainable

Guardrails

What the engine is not permitted to do

  • Approve on its own

    The engine recommends. Approval requires an officer with the permission to give it.

  • Exceed a regulatory limit

    A recommendation above the NRB ceiling or the 36-month cap is rejected before it reaches an officer.

  • Use data outside its purpose

    Scoring reads borrower data under the credit_assessment purpose, and that read is logged.

Why it matters

Consistency is the point

Indicative of what a codified credit policy changes. Actual outcomes depend on the policy you configure.

  • 1

    Credit policy
    Applied identically regardless of branch, officer or day of the week.

  • 100%

    Decisions with a reason
    Adverse outcomes cannot be saved without one.

  • 0

    Automated rejections
    A human is accountable for every decline.

  • Full

    Reconstructability
    Score, factors, officer and ruleset version, retained together.

Fair treatment

Designed so a borrower can be told why

A declined borrower who receives no explanation escalates — to your grievance channel, then sometimes to the regulator. The platform is built so a clear, accurate reason is always available.

  • A declined application always carries a written reason
  • The reason is retained for the full retention period
  • Your grievance process can retrieve the full decision record
  • Protected characteristics are not collected, so they cannot enter a score

Coming soon

Model your credit policy with us

Bring your current underwriting rules and we will show you how they configure — and where the platform would stop you.