Know which claims need attention
before the loss run shows it
Parsewise Casualty Predict reads adjuster notes, medical reports and legal letters together with the loss run. For each open claim it predicts whether incurred will jump, by how much, and when the claim will close.
We test every model on past claims with known outcomes, using only what was in the file at the time.
Problems show up in the claim file months before the loss run
The adjuster notes it the week it happens. Incurred catches up later.
Illustrative claim
1. You send loss runs and claim files
Hosted in Europe or in your cloud
2. We read each file as of each valuation date
Every fact linked to its source
3. You get a prediction for every open claim
Deterioration, incurred and closure
Four long-tail lines
One core set of signals, plus a few per line.
Workers' compensation
- Work status
- Medical escalation
- Return to work
- Attorney
General liability
- Litigation stage
- Demand near limits
- Venue
- Multiple defendants
Commercial auto
- Injury severity
- Early representation
- Treatment
- Multiple claimants
Medical malpractice
- Injury severity
- Expert evidence
- Settlement stage
- Local caps
- United States
- United Kingdom
- European Union
A glossary per state and country maps local terms to the same signals.
Questions it answers
Claims teams
- Which open claims are likely to jump in the next 12 months?
- Which files should a senior handler see first?
- What is the next step on this claim?
Actuaries and reserving
- Which case reserves look light?
- How much will incurred grow over 6, 12 and 24 months?
- Which claims are likely to close this year?
Reinsurers, run-off and diligence
- Where is development building in a book we are taking on?
- Which large losses are still moving?
- Which claims hold most of the expected development?
Four rules for every model
No hindsight
We read a claim using only what was written by its valuation date, so a test uses only what a handler could have seen.
Every fact cited
Each value is linked to the sentence it came from, so a handler or an auditor can check it.
Reasons with every prediction
Every prediction comes with the drivers behind it, in plain language.
Tested on years it never saw
We score each model on held-out years and check that no information leaks in from the future.
Security & sovereignty
SOC 2 Type II and GDPR
Compliance status and documents in our trust centre.Hosted in Europe
Documents, data and model inference stay in Europe.Private cloud (VPC) deployment
Run Parsewise in your own cloud.
Encryption and access controls
Encrypted at rest and in transit. MFA included, SSO on Business and Enterprise.Documented data management
Retention and deletion policies, ready for your security team.
How pricing works
Backtest on your book
A fixed fee for a scoped test on your own closed and open claims, credited against your first year if you continue.
Portfolio monitoring
An annual fee based on open claims, lines, refresh frequency and deployment (Europe-hosted or your own cloud).
Deal diligence
Priced per transaction, on the deal timetable.
Common questions
What is Parsewise Casualty Predict?
Parsewise Casualty Predict forecasts how open casualty claims will develop. For each claim it gives the probability of deterioration, expected total incurred and the probability of closure at 6, 12 and 24 months. It reads the claim file as well as the loss run, and lists the reasons for every prediction.
Which lines of business do you cover?
Workers' compensation, general liability, commercial auto and medical malpractice. All four use the same core signals plus a few per line, and we calibrate and test each model on your own data before you use it.
Which geographies?
US states and European countries. A glossary per jurisdiction maps local terms to the same signals, for example a Section 32 agreement in New York or a court-appointed expert in Italy. We take local rules such as damage caps and settlement law from a reference table rather than from the AI.
How is this different from actuarial reserving?
Reserving works on triangles and aggregates. Parsewise Casualty Predict works claim by claim and reads the file, so it can flag a deteriorating claim before the change reaches the loss run. It complements your actuarial view rather than replacing it.
What data do you need?
Loss runs (paid, incurred and reserves by valuation date) and the claim file: adjuster notes, medical and legal reports and correspondence. For a backtest, a few thousand closed and open claims with several years of history is enough.
Is our data secure, and is it used to train models?
Parsewise is SOC 2 Type II and GDPR compliant. Documents, data and model inference stay in Europe, or Parsewise Casualty Predict runs in your own cloud. We encrypt data at rest and in transit and redact direct identifiers before analysing any text. We use your claims only to calibrate and test the models for your book.
Is a prediction legal, actuarial or reserving advice?
No. Predictions are statistical estimates to support your team's judgement. Reserving and settlement decisions stay with you.
How does billing work?
You start with a fixed-fee backtest on your own claims, credited against your first year if you continue. After that, portfolio monitoring is an annual fee and deal diligence is priced per transaction. See pricing above.
Try it on your own claims
Send us a sample of closed and open claims. We'll show you which of them had warning signs in the notes before the loss run moved.
Hosted in Europe · SOC 2 Type II · GDPR compliant
