Our team turns years of submission and claims files into verified data, tests what separated the losses, and builds a model from what holds up, for your own systems or for Lexcel.
Illustrative workbench: claims from a study, each with its verified type and source file.
Insured
Claim Type
Claim Notice
Pellingrove Robotics
Software defectVerified
pdfClaim 24-0312.pdf
Lumitrove Software
Data lossVerified
pdfClaim 22-0950.pdf
Kestrova Systems
Project delayVerified
pdfClaim 23-0127.pdf
Brennick Controls
Installation errorVerified
pdfClaim 21-0536.pdf
Tarnwell Automation
Installation errorVerified
pdfClaim 24-0088.pdf
Wexholt Design
Design errorVerified
pdfClaim 23-0611.pdf
Halvoran Systems
Software defectVerified
pdfClaim 22-0274.pdf
Vesperline Robotics
Software defectVerified
pdfClaim 23-0442.pdf
Tessamere Systems
Data lossVerified
pdfClaim 22-0718.pdf
Tessamere Systems
Project delayVerified
pdfClaim 24-0196.pdf
Uncapped risks
JRJordan Reyes
Pull our claims on risks with no liability cap.
Lexcel agent
Listed Submissions(1 filter)
Listed Claims
Read file →Claim 24-0312.pdf
Read file →Claim 23-0127.pdf
Read file →Claim 24-0088.pdf
Here they are, marked in the table.
Lexcel agentClaims
Ask Lexcel…
02
Find what held up
We compare what you had at underwriting on risks that led to a claim and risks that did not. The findings show what beat chance, what made no difference, and what to request at submission.
Your firm’s work
Bound policies
1,284
Quotes
3,912
Claims
184
Endorsements
2,050
pdfKestrova Systems, Application.pdf
pdfTarnwell, Client Agreement.pdf
pdfBrennick Controls, Loss Runs.pdf
pdfPellingrove, Claim 24-0312.pdf
pdfBrayford Systems, Policy 2023.pdf
Study findings
What separated risks that led to a claim
Held up
No liability cap in client agreements
Held up
No written sign-off on delivered work
Held up
Claims in the last five years
Needs more data
Safety-critical systems
No difference
Years active
Recommendation
Build a claims model
Request at submission
Liability caps
How work is signed off
03
Build the model
If the data supports it, we build a model to guide selection and pricing, tested on held-back files. Use it in your own systems or in Lexcel, where each new submission gets a rating, drivers, and source pages.
Illustrative workbench: submissions on the desk with their claims rating, stage, and annual revenue.
Name
Claims Rating
Stage
Annual Revenue
Tollwick Robotics
Yellow
Quoting
$52.6M
Pellingrove Robotics
Red
Underwriting
$58.4M
Kestrova Systems
Yellow
New submission
$41.9M
Tarnwell Automation
Red
Quoting
$33.0M
Lumitrove Software
Green
Bound
$47.1M
Wexholt Design
Yellow
Quoting
$12.4M
Vesperline Robotics
Yellow
Underwriting
$72.6M
Brennick Controls
Green
New submission
$29.4M
Halvoran Systems
Green
Bound
$38.8M
Brayford Systems
Yellow
New submission
$36.2M
Claims modelTollwick Robotics
RatingCloser reviewYellow
Raises
No liability cap in client agreements
Client Agreement §11
Raises
No written sign-off on delivered work
Application.pdf p. 6
Lowers
No claims in the last five years
Loss Runs.pdf p. 1
A fraction of the time and cost,
reviewed by our team and open to your actuaries.
Actuarial analysis at this depth has taken a large team a year or more.
Agents do the reading, so a lean team of experts gets you to an answer
far sooner.
We review the method and results, and your actuaries can review them
too. The model and findings are used only for you, and your data never
trains a general model or reaches another customer.
Can our claims history predict which submissions turn into claims?
Sometimes, and finding out is the first job of a claims study. Lexcel tests which details you had at underwriting separated the risks that led to a claim.
The findings cover what held up, what showed no difference, and what to request at submission, with a recommendation on what to do next. Lexcel does not guarantee predictive performance, and your underwriters make every decision.
How many claims does a claims study need?
It depends on the line and how strong the effect is. Claims are thin across much of specialty, so a small count can point to trends worth testing without supporting a reliable score. The first phase shows how far your data goes and what more it would take to go further.
What if our data cannot support a claims model?
You keep the findings and the dataset, on workbenches your team can search. Lexcel then configures a workflow that compares each live submission with past claims. That workflow does not produce a claim probability, but it puts the closest history in front of the underwriter.
Do we have to use Lexcel to run the model?
No. Lexcel does the analysis and builds and tests the model, and you can implement it in your own systems. Running it in Lexcel is the simplest path, because the rating, its drivers, and its source pages sit on the submission record.
Does a claims study replace our actuaries?
No. Much of what separates one specialty risk from another sits in documents that rarely reach a model. A claims study turns those files into structured, source-linked data for your actuaries to work with. Lexcel reviews the method and results, and your actuaries can review them too.
Is a claims study an actuarial opinion?
No. It is analysis to support underwriting decisions, not a statement of actuarial opinion, a rate filing, or a reserve study. The model can guide selection and pricing, but your team and your actuaries set the rate. If you or a capacity partner need an independent model rebuild or a challenger model, that is scoped separately.
Who owns the structured underwriting history?
The organization that produced the work product owns it. Agentic AI runs on that organization's materials, under its control. Claims history and internal analysis are among the categories of data that require the highest care.
Connects everywhere you work, and works with your existing systems.
Security and compliance
Built for enterprise security.
SOC 2 Type II
Start with one process. We’ll run it on Lexcel with you.
Bring a line of business or a review your team wants to move faster, and we’ll walk through it on the workbench, from the first email to your systems.