Was the advance the cause, or did the approval process shape what we observed?
When repayment shortfalls rise with advance size, that relationship alone does not tell a risk team what would happen if it changed an advance policy. Recipients may differ, and records may omit workers whose repayment outcome is not yet settled.
Each question calls for different evidence.
A risk team considering segment-specific advance caps needs to distinguish the amount received from the rule that determines who receives an advance.
Illustrative synthetic evaluation: These estimates come from a closed-model simulation, not real worker or customer outcomes. They are not validated portfolio results.
| Question | Example | Evaluation result | How to read it |
|---|---|---|---|
| Association | Do larger advances coincide with more repayment shortfalls? | Unadjusted estimate: +0.00064 shortfall probability per additional dollar. | This combines the advance relationship with differences in who receives advances and whose repayment outcomes are observed. It is a signal to investigate, not an answer about policy. |
| Intervention | Among observed advance recipients, how does repayment shortfall change as advance amount changes? | Backdoor-adjusted estimate: +0.00049 per dollar (95% CI: 0.00036–0.00062), with a DML cross-check. | A modeled dose effect over the observed advance range in a synthetic cohort. Unobserved requested demand and the adjustment set remain limitations; this is not validated for real customers. |
| Counterfactual / policy | What would happen if the cap changed for a segment? | Not identified by the advance-dose estimate. | Changing the amount received and changing the eligibility or maximum-amount rule are different interventions. The cap question needs its own evaluation. |
Bring the actual policy lever into the question.
CausalGraphOS helps teams define the decision, organize evidence and assumptions, and estimate only what the data can support. For earned wage access, that means keeping requested amount, approval rules, advance amount, and repayment-outcome observation distinct.
It also means surfacing unresolved demand and unsettled repayment records as limitations, rather than treating an amount estimate as proof that a cap, pricing, or payout-policy change will work.
Bring us the advance-policy question your team needs to answer.
We’re shaping CausalGraphOS with design partners working on consequential decisions.