Solutions · By practice

Predict for premises liability practices.

The harder PI case. Smaller cohorts, more variability by sub-type (retail, commercial, residential), and notice/duty issues that can move the band by an order of magnitude. Predict tunes the model on every one of those dimensions.

Predict your case Free PL calculator Free · 60 seconds · no signup to start
Inside the workspace · a premises case
Illustrative example. A sample Miami-Dade retail slip-and-fall in the Predict workspace. $72K predicted, ± $34K (90% band). The band is wide because the matching cohort is 41 cases, not 312, and notice status is only partly documented. Sample data, shown to illustrate the view. Run your own case to see your numbers.

PL is where the gut is least reliable.

An MVA practice sees enough rear-end soft-tissue cases to develop a defensible instinct. A PL practice sees a slip in a grocery aisle, then a commercial landlord stair fall, then a residential pool drowning — three cohorts that share almost nothing. The default instinct is rarely calibrated to the case in front of you.

Today

One cohort for every PL case

"Slip-and-falls go for $40–80K." That cohort exists, but the case in front of you — commercial premises, prior notice, lost-wages claim — sits inside a sub-distribution the gut average flattens away.

With Predict

Sub-categorized comparables

Cohorts split by premise type (retail / commercial / residential), notice status, injury severity, and jurisdiction. The number is built on the slice that actually matches the case.

Today

Notice issues hidden in the number

A PL case with documented prior notice is a different distribution from one without — sometimes 2–3x. The single-number gut estimate hides which side of that line the case is on.

With Predict

Notice flagged in the band

Prior-notice / no-notice is a top-three feature in the PL model. The band reflects which side of the duty line the case sits on — and tells you when the answer depends on discovery you haven't done yet.

The PL dataset

Trained on premises liability specifically — and honest about the harder cases.

72K
PL records in the active training set
Premises cases from the 2018–2025 verdict-and-settlement set, after deduplication and the 4-year rolling cut-off. Smaller than the MVA fold, in line with the underlying market.
91%
Held-out PL accuracy
MdAPE on cases the model never saw during training (test set 2024–2025). One point below MVA. See exactly how we test it →
6
Sub-categories
Retail, commercial, residential, public sector, hospitality, parking. Each modeled separately before the cross-jurisdiction tune.

PL is where the confidence band earns its keep. When the matching cohort is thin or the notice status is ambiguous, the band widens, and that's the signal to slow down on case selection, not to ignore the model. Read the full methodology →

A wider band is still better than a wrong gut.

A PL case with 18 comparables instead of 312 gets a wider 90% confidence interval — the methodology forces honesty about uncertainty. That is the input to the case-selection decision. A $40K–$110K band tells you the case can pay if worked, and it tells you the upside isn't a guarantee.

The default tool — gut + a senior partner's recall — produces a single number with no uncertainty surface. The number feels more confident; it isn't.

How the confidence band works →
$72,000
± $34,000 · 90% CI
FL · MIAMI-DADE · PL
Wider band than a typical MVA case — fewer comparables, notice status only partially documented. The band is the answer.

Where Predict fits in a PL workflow.

With Predict
Without
Versus a verdict database
One jurisdiction-tuned number with a 90% band, built only on the sub-cohort that matches premise type and notice status
A list of past verdicts you weight and average by hand. No uncertainty surface, no notice-status filter
Case categorization
Premise type, notice status, injury severity, jurisdiction
"It's a slip-and-fall" — one category, broad cohort
Intake decision
You set a value your firm needs a case to clear before you take it. Predict shows whether the case clears it, and flags a wide band as "run discovery before you sign the retainer"
Take it if the senior partner has time to look at it
Notice status
Top-three feature in the model; band widens if undocumented
Carried in the attorney's head; rarely changes the number
Demand letter
Valuation block: number, band, sub-cohort, exportable PDF/Word
One number with no methodology — defense counsel pushes the carrier to ignore it
Wide-band cases
Flagged for senior review before retainer
Discovered at depositions, 6 months and $40K of paralegal hours later