Solutions · By practice

Predict for MVA attorneys.

A jurisdiction-tuned settlement value for motor vehicle accident cases — rear-end, soft-tissue, intersection, commercial vehicle, wrongful death. Calibrated specifically to an MVA hold-out test set at 92% accuracy.

Predict your case Enter one MVA case, get a confidence-banded value in about a minute. No signup to start.

Prefer to explore first? Try the free MVA calculator →

Inside the workspace · an MVA case
Illustrative example. A Texas Harris County rear-end case in the Predict workspace — $485K predicted, ± $42K confidence band, cohort of 312 comparable Harris County MVA settlements. The artifact attorneys export under the demand letter.

The MVA case is where the carrier playbook is sharpest.

Soft-tissue rear-end with mid-five-figure med-specials is the case carriers value against an internal model every day. The plaintiff attorney working from gut + PD photos is bringing instinct to a fight that's been quantified for a decade.

Today

Anchored to property damage

Crumpled bumper → "soft case." Light contact → "lowball." The PD photo is a poor proxy for the soft-tissue cohort that actually drives settlement value in your jurisdiction.

With Predict

Anchored to comparable verdicts

5–10 comparable MVA settlements from the same county and same injury severity tier. The band is wide when the cohort is thin and tight when it's deep — the methodology is the answer to "how do you know?"

Today

One number, no range

"Probably worth around $80K." The carrier hears a single number with no methodology; the demand letter reads as a guess. The first offer comes in at $32K and the negotiation starts in the wrong half of the distribution.

With Predict

A defensible band

$148,000 ± $22,000 at 90% confidence, cited to 312 Harris County rear-end settlements at moderate injury severity. The demand-letter export is shipped with the band and the cohort, not the gut.

The MVA dataset

Trained on motor vehicle accidents specifically.

198K
MVA verdicts & settlements
The training set is filtered for MVA before the model is calibrated. No premises, no products, no malpractice noise.
92%
Held-out MVA accuracy (MdAPE)
On the held-out MVA fold, predictions came within 92% of the realized settlement (median absolute percentage error). The train/test split is strictly temporal: trained through 2023, tested on 2024 to 2025. See the held-out results and methodology →
312
Comparable cohort, typical case
Median MVA prediction is grounded in 312 same-jurisdiction, same-severity comparables. The band widens when fewer are available — never silently.
How we know

Where the numbers come from, and how you can check them.

The settlement and verdict data is plaintiff-only. Every record is drawn from publicly available verdict and settlement filings, plaintiff-side reported settlements, and PACER docket extracts, then filtered to MVA before the model is calibrated. No claims-side or carrier data is used in training, and none ever will be.

The held-out results are published, not asserted. The methodology page shows the training data, the model architecture, the strictly temporal train/test split, and the held-out accuracy on the MVA fold, including how the band behaves when the comparable cohort is thin. Read how the data is sourced →

Built for the intake call, not the demand letter.

The case-selection decision on an MVA case happens at minute four of the intake call. Predict produces a confidence-banded value in 30 seconds — before the retainer goes out, not after the file has been worked.

The output is a number, a band, a jurisdiction tag, and the comparable cohort. The artifact that survives a partner second-opinion or a paralegal handoff: same number, same methodology, every time.

Read the full methodology →
$148,000
± $22,000 · 90% CI
TX · HARRIS · MVA
Soft-tissue rear-end, low-speed impact, ~$18K med-specials, 6 months of treatment. The case the carrier is valuing at $48K.

Where Predict fits in an MVA workflow.

With Predict
Without
Intake call
30-second predicted band, before the retainer
"Sounds like a good one" — anchored to PD photo + caller affect
Retainer decision
Band ≥ firm's case-value threshold → sign
Sign first, value at the demand-letter draft 60 days later
Demand letter
Valuation block: number + band + cohort, exportable PDF/Word
One number with no methodology — easy for the adjuster to ignore
First offer counter
Counter to the lower bound of the 90% CI, defensible on the record
Counter to a round number 20% above the offer — anchored to nothing
Second-opinion
Same model, same band, same cohort across the firm
Senior partner re-runs the gut analysis from scratch

If you already license a verdict-search database, the difference is what happens after retrieval. A verdict database gives you the comparable cases. Predict scores their similarity, weights them to your jurisdiction and injury severity, and returns a calibrated value with a confidence band, so you cite a defensible number instead of a stack of comps you still have to value by hand. Predict vs. a verdict-search database →