Log in Legacy app
Compare · Predict vs Verdict Search

A library, vs a model built on top of one.

Verdict Search and the related verdict reporters are the legal database the bar has used for 30 years. They're a library of comparable matters; Predict is a model that ingests that kind of data and outputs a confidence-banded number. The two are not the same instrument.

Predict is a web tool your intake team runs at the desk: type the case facts, get a confidence-banded value and its cited comparable cohort in about 30 seconds. Plaintiff-only, by contract.

Time to a defensible number

The lookup is where the workflow currently breaks — 1.5 to 2 hours of paralegal time per case, then another pass to weight the comps.

Predict
30 sec
Verdict Search
~2 hrs

The 30 seconds produces the same kind of banded number a 1.5 to 2 hour paralegal comp-pull produces, cited to a comparable cohort you can open and read. The speed comes from the model doing the cohort selection, not from skipping it.

Predict
Verdict Search
What it is
Predictive model with confidence bands
Searchable database of reported verdicts and settlements
Cost
$499 / mo, unlimited predictions
~$200 / mo subscription + 1.5–2 hrs of paralegal time per case
Output
A number with a ± band and cited comparables
A list of matters to read and weight manually
Jurisdiction-tuning
Stratified state + county folds, automatic
Manual filter by court — you do the cohort selection
Side-of-table
Plaintiff-only, contractually
Both sides have access — carriers subscribe too
In the workflow
At intake, before the retainer
Pre-demand, after the retainer (the time cost is the reason)
Original verdict text
Cited cohort with links — not a replacement for reading them
Full reporter text, searchable
Specific named case
Yes, if it's in the comparable cohort
Yes — that's the core use case
vs your own estimate
A starting number with a stated 90% band and the cohort it came from, so your judgment adjusts a defensible baseline instead of replacing it
Still your manual read of the matters you pull

Predict is built on top of the same kind of data.

312K verdicts and reported settlements. The training corpus overlaps substantially with what Verdict Search indexes — published verdicts, reported settlements, court records. The difference isn't the data; it's what the instrument does with it.

Verdict Search hands you the cohort and asks you to value the case. Predict values the case and hands you the cohort. Same underlying material, different outputs, different point in the workflow. If you're researching a specific named case or pulling original verdict text, the lookup is still the right tool — and we'll cite to it from inside the Predict output.

On a held-out test set, Predict's valuations are 90 to 92% accurate against the realized settlement (median absolute percentage error) for MVA and premises cases. Every number also carries a 90% confidence band. The band is the methodology, not a hedge: the number is the headline, the band shows how sure the model is.

See how accuracy is measured →
$185,000
± $28,000 · 90% CI
CA · LOS ANGELES · MVA
The Verdict Search workflow on this case: 1.5 hours to pull comparables, another 45 minutes to weight them, then a memo. Predict outputs the band and cites the same kind of cohort in 30 seconds.