Skip to main content

Scoring System

The initial scoring system should be rule-based. Machine learning can be considered later after enough data, consent coverage, and legal review exist.

Initial rule-based formula​

Example:

score =
  60 * on_time_payment_ratio
  + 20 * low_delay_score
  + 10 * contract_completion_score
  + 10 * data_completeness_score

The result is normalized to a 0-100 range.

Example inputs​

MetricMeaning
on_time_payment_ratioShare of installments paid on or before due date.
low_delay_scorePenalty based on average and maximum delay.
contract_completion_scorePositive signal for completed contracts without unresolved debt.
data_completeness_scoreConfidence based on months of available data.

Confidence score​

The score should include confidence based on:

  • number of installments observed
  • length of rental history
  • recency of data
  • percentage of reconciled payments
  • number of missing or manually corrected records

Shadow scoring​

At first, scores should be calculated in shadow mode:

  • not visible to tenants
  • not visible to landlords
  • used only for internal validation
  • compared against manual expectations
  • monitored for bias and unexpected outcomes

Future ML evolution​

Machine learning should only be introduced after:

  • clear consent is collected
  • explainability requirements are defined
  • sufficient training data exists
  • bias and fairness checks are implemented
  • a manual appeal or correction workflow exists