On this page
Rate filing work is a chain of calculations: build the triangles, develop losses to ultimate, select trends, bring premium to current rate level, add expense and profit provisions, weigh credibility, and turn it all into an indication and the exhibits a reviewer reads. Each step is standard; the time goes into moving data between spreadsheets, checking the arithmetic and documenting what was done.
Evermont is an insurance regulatory filing desk inside the AI assistant your team already uses (Claude, ChatGPT or Codex). For the actuarial side it gives the assistant a library of deterministic calculation modules. You describe the analysis in plain language; the assistant picks the modules, runs them on your data and lays out the results, and your actuary reviews every number before it goes anywhere. The rest of the filing job, from state requirements to objection responses, is described here.
What the calculations cover
The modules follow the steps of a property and casualty rate indication, plus the classification and reserving work around it:
- Data: profiling and cleaning a dataset, building paid, incurred and count triangles from claim transactions, summarizing experience and reconciling totals.
- Losses: loss development with triangle diagnostics and Berquist-Sherman adjustments, loss trend fits, catastrophe and large loss loads, severity distributions, benefit level adjustments.
- Premium: on-leveling to current rates and premium trend.
- Provisions: expenses and the permissible loss ratio, underwriting profit and investment income, the net cost of reinsurance, and loss cost multipliers for advisory organization adoptions.
- The indication: overall and loss cost indications, credibility and its complement, the rate indication exhibit and sensitivity to key selections.
- Rating plan: territory and classification relativities (including minimum bias and a GLM), increased limits and deductible factors, off-balance, the rating algorithm, rate page calculations and policy rerating to show the impact on customers.
- Reserving: reserve estimates and their variability.
The assistant can list every module and the inputs each one takes, then chain them: trend dates feed the trend fits, which feed the indication exhibit.
Working from your own data
Upload the data to a filing as an Excel or CSV file: a loss triangle, a policy listing, experience by year. A calculation reads the sheet and range you name straight from the uploaded file, so a table is never retyped from page text. The files stay in your organization's filing, and the assistant can turn the results into a Word or Excel exhibit saved alongside them for review.
From data to a reviewed indication
Upload
Add triangles, premium and exposure data to the filing as Excel or CSV.
Run
Ask for the analysis; the assistant runs the modules in order on your data.
Check
Each result comes back with its quality checks and an audit record.
Review
Your actuary reviews the selections and results before anything is filed.
Why deterministic calculations matter
A general AI chatbot can write a formula or describe a method, but its arithmetic and its handling of a large table are hard to verify. Actuarial standards expect the work to be documented well enough that another qualified actuary could assess it (ASOP No. 41), and reviewers ask how each selection was made. Evermont keeps the math in versioned code with recorded inputs, so the same inputs give the same answer, and the audit record says exactly what ran.
The modules return checks rather than silence. A pass is a pass; a warning or a review item is a point for the actuary to weigh, such as the tail factor used or a default assumed; a failure, such as a fit that did not converge, blocks the result.
What stays with your actuary
- Selections and judgment. Which trend, which tail, which credibility standard, and whether a result is reasonable are actuarial decisions. The modules disclose the assumptions they use so they can be challenged.
- Sign-off. Evermont's output is not actuarial advice. A qualified actuary reviews and signs the memorandum and exhibits, as the CAS ratemaking principles and ASOP No. 53 assume.
- Data quality. The modules check the data they are given; they cannot know what is missing from it.
For what a filing reviewer expects to see, read the actuarial memorandum and rate filing exhibits. The docs list the calculation tools, and the demo shows Evermont at work.
Questions and answers
Does a language model calculate the numbers?
No. Calculations run as deterministic Python modules on Evermont's servers. The assistant chooses the module and passes the inputs; the module returns the numbers, its quality checks and an audit record with hashes of the code and the inputs.
Which calculations are available?
Modules cover the rate indication (overall and loss cost), loss development and triangle diagnostics, loss and premium trend, premium on-leveling, credibility, expense and profit provisions, catastrophe and large loss loads, increased limits and deductibles, territory and classification relativities including a GLM, rating algorithms, policy rerating, off-balance, sensitivity and reserve estimates. The assistant can list them with their inputs.
Can it read our data from a spreadsheet?
Yes. Upload an Excel or CSV file to the filing, and a calculation can read a named sheet and range from it directly, so the assistant does not retype the table.
Who signs off on the result?
Your actuary. Every result is marked as computed and unreviewed and requires qualified review, and a result with a failed quality check must not be carried forward. Evermont's output is not actuarial advice.
See it work: watch the 75-second demo or read the docs for setup, the tool list and coverage.
Sources
General information, not legal or actuarial advice. Requirements differ by state, line and filing; check the current rules with the department of insurance before you file.