Criterica Intelligence — production models trained on real court records, not synthetic data
The Economics of Accuracy

In a probabilistic business, accuracy is a P&L line.

Every institution that touches legal risk runs on an implicit accuracy number — a selection rate, a reserve band, a settle-or-litigate call. Small changes in that number compound into large changes in financial outcome. Set your own assumptions below and see what a percentage point is worth on your book. Every output follows arithmetically from your inputs; nothing here is a Criterica performance claim.

CAPITAL DEPLOYMENT — MODELED FROM YOUR INPUTS
Cases in portfolio100
Average deployment per case$500K
Your baseline favorable-resolution rate55%
Selection lift you want to test5pp
Gross multiple on favorable resolution2.2×
Your average case duration26 mo
Duration reduction to test4 mo
$50.0M COMMITTED · EXPECTED GROSS RECOVERY
YOUR BASELINE
$60.5M
WITH YOUR HYPOTHESIZED LIFT
$66.0M
What 5 points of selection precision is worth on this book
+$5.5M
Computed from your inputs above: committed capital × your hypothesized rate change × your gross multiple. The lift is your hypothesis to test — not a Criterica performance claim.
Duration is capital velocity
26mo · 43.9%22mo · 53.7%CASE DURATION (MONTHS) → ANNUALIZED GROSS RETURN AT 2.2× MULTIPLE
Same case, same multiple: resolving 4 months earlier moves annualized gross return by +9.8%. Duration models price this dimension on every scored case.
Why the lift is your hypothesis

We do not publish a universal lift number, because no honest one exists: incremental accuracy over your process depends on how good your process already is. A disciplined underwriting team starts from a higher baseline than a volume intake desk, and the same models add different value to each.

What we can state precisely is how our numbers behave: calibrated probabilities, tested on real outcomes the models never saw, at the jurisdiction level — so a 70 percent prediction is right about 70 percent of the time. What that is worth on your book is exactly what the audit measures.

How you find your real number
01
Portfolio Intelligence Audit
We score your historical book — cases with known outcomes — and measure model performance against what actually happened. Your data, your baseline, measured lift.
02
Side-by-side period
Run model outputs next to your existing process on live intake, without changing a single decision. Compare after the fact.
03
Then decide
The calculator above told you what a point is worth. The audit tells you how many points are on the table. Both numbers are yours.
Start with the audit →