Savings explained
We show savings in two complementary ways.
1) Threshold-based net savings
What it is: We pick a risk threshold and treat everyone above it.
Formula (conceptual):
Net = (True Positives × effectiveness × replacement_cost) - (Treatments × intervention_cost)
- True Positives: flagged people who would have left
- Treatments: everyone flagged (TP + FP)
Use it for: operational reporting (how many flagged, precision/recall).
2) Threshold-free savings (expected value)
What it is: We sum expected value per person over the treated cohort (Top-K or Threshold, with Coverage), without relying on a single cut-off.
Per person Expected Value:
effectiveness × replacement_cost × predicted_risk − intervention_cost
Use it for: comparing scenarios and strategy settings; it reflects how risk actually shifts.
Reading the numbers (example)
| Metric | Value |
|---|---|
| Net savings (base) | $60,217,234 |
| Net savings (scenario) | $60,217,234 |
| Δ Net savings | $0 |
| Threshold-free savings (coverage-adjusted) | −$3,421,600 |
What this means:
As work conditions improve, people's attrition risk decreases, so at some mixture of settings these people will cross the threshold and impact the net savings.
- The threshold-based result: Net savings (scenario) vs Net Savings (base) didn’t change because your what-if didn’t move enough people across the chosen threshold (so flagged counts and TP/FP stayed the same).
- The threshold-free result is negative, meaning the scenario made the treated cohort less profitable in expectation (e.g., you spent money treating lower-risk people or your effectiveness/cost assumptions make the lever uneconomical).
What to try next: - Increase the lever strength (e.g., larger workload reduction) - Target smarter (Top-K with moderate Coverage) - Recheck assumptions: effectiveness ↑ or cost ↓ often flips EV positive