A valuation block the adjuster can't dismiss.
Every Predict prediction exports as a ready-to-paste demand-letter valuation block — PDF or Word — with the number, the 90% confidence band, the cited comparable cohort, and the four facts that moved the prediction. The artifact that turns "we value the case at $148K" into "we value the case at $148K and here's why."
New to Predict? Predict is a settlement-value model for plaintiff MVA and premises cases, jurisdiction-tuned and shown with confidence bands. You can run a free prediction before you sign up; the demand-letter export ships with the trial.
Comparable cohort
This prediction is grounded in 312 same-jurisdiction, same-severity rear-end MVA settlements from Harris County, Texas, between 2018 and 2025. The cohort excludes catastrophic-injury outliers and out-of-state cases. The 90% confidence interval reflects the spread of the comparable cohort, not a generic margin.
The four facts that moved the prediction
- Injury severity tier: moderate — soft-tissue, six months of treatment, no surgical intervention. Anchors the prediction to the moderate-severity Harris County cohort (n=312), not the minor cohort (n=894, median $42K).
- Medical specials: $18,400 across orthopedic and chiropractic providers. Inside the 25th–75th percentile band for the comparable cohort.
- Property damage: $4,200 — low-speed impact. Note: PD is weakly correlated with settlement value in the moderate-severity cohort; the model deweights it accordingly.
- Defendant insurance posture: commercial policy with $1M/$2M limits. Removes the lowball-floor constraint typical of state-minimum personal lines.
Methodology & reliance
The valuation is produced by Predict (predict.law), version v3.2, refreshed 2026-04-01. The model is trained on 312,000 verdicts and reported settlements from 50 US jurisdictions; held-out 2024–2025 MdAPE accuracy (median absolute percentage error against actual settlements) on Texas MVA cases: 92%. Full methodology and recalibration policy are public at predict.law/methodology.
Four blocks. No fluff. Defensible end-to-end.
The number + the band
Predicted settlement value with the explicit 90% confidence interval. Never a single number — never a range without a median. The pair is the load-bearing claim.
The comparable cohort
5–10 verdicts and reported settlements that informed the prediction — same case type, same jurisdiction, same severity tier. With n, year range, and the filtering criteria used.
The four facts that moved it
Plain-language reasons for the number, computed per prediction (the model attributions, surfaced as SHAP values): which inputs pushed the value up, which pushed it down, by how much. The "why this number" question, answered.
Methodology & reliance
Model version, training-set size, MdAPE accuracy on the case-type fold, and a public link to the methodology page. The adjuster can't claim "the AI made it up."
Read how the band is computed and recalibrated at predict.law/methodology.
The same demand letter, on either side of Predict.
The legal arguments are the attorney's. The valuation paragraph — where every demand letter either lands or gets dismissed — is what Predict changes.
"Based on the severity of our client's injuries, the duration of treatment, and comparable cases handled by this office, we value the case at $120,000. We are prepared to litigate."
One number. No band. No cohort. No citation. The adjuster reads this as a negotiating opener and counters at $32K.
"We value the case at $148,000, with a 90% confidence interval of $126,000–$170,000, derived from a cohort of 312 Harris County moderate-severity rear-end MVA settlements. The full methodology, model version, and held-out accuracy figures are public at predict.law/methodology."
Number, band, cohort, methodology link. A cited, banded number gives the adjuster something to respond to on the merits rather than a round opener to discount. We don't claim a specific dollar lift; we show the methodology and let the file speak.
Two clicks from a prediction to a paste-ready block.
- 1
Start the trial, then run the prediction
Start the free trial, then answer six questions (30–60 seconds). The prediction comes back with the band, the cohort, and the four facts already attached.
- 2
Click Export → PDF or Word
The valuation block downloads as a single-page PDF or a Word document with editable headings — so the firm can drop it under its own letterhead and adjust tone without rebuilding the substance.
- 3
Paste into the demand letter
Drops in as the valuation paragraph. The legal arguments, recovery theory, and tone are the attorney's. The math is Predict's — and the methodology is in the open.
Quick answers.
Is the export available on the free trial?
Yes — the trial unlocks the full export including the cohort and the feature attributions. The public predict-your-case calculator returns the number and the band, but not the export.
Does it work on letterhead?
The Word export ships with editable headings and no fixed letterhead. Drop your firm's letterhead in the document header; the valuation block sits underneath as the substance of the demand.
How do I cite Predict in the demand letter?
The export auto-includes a methodology-reliance paragraph with the model version, MdAPE accuracy on the case-type fold, and a link to predict.law/methodology. Counsel on the other side can read the methodology; the page is public.
What does it cost after the trial?
$499/month, single tier, with unlimited exports on every prediction. The 14-day trial is free, no credit card at signup, and you can cancel in one click. At day 15 the card on file is charged $499; before then, nothing. Full pricing and limits are at predict.law/pricing.
Can I edit the number before exporting?
No — and that's a brand commitment. The valuation block exports the model's number, not an attorney-adjusted number. Editing the model output and presenting it as model output is exactly the failure mode the methodology page is built to prevent.
Try one export on a real case from your docket.
The 14-day trial unlocks the export, the cited cohort, and the four-facts attribution on every prediction. No credit card at signup. The first valuation block ships the moment the first prediction completes.