Ask a pricing team where to find the most detailed rating information in US property and casualty insurance and many will point to the state filing systems. Before an insurer can use a new rate or rating rule, most states expect it to be filed with the department of insurance. Once filed, much of that material becomes public.
So the base rates, the factor tables, the eligibility rules and often a set of worked premiums are already written down. The hard part has always been getting them out of the documents and into something that can price a quote.
Anatomy of a filing
No two filings look quite the same. The state, the line and the carrier all shape the package. Most rate and rule filings carry some mix of these parts:
- Rate pages. The base rates plus every relativity table: territory, age, construction, protection class, deductible, limit and so on.
- The rule manual. How the factors combine, who is eligible, how rounding works, and where minimum premiums, fees and surcharges come in.
- The actuarial memorandum. The carrier's case for the change: the overall rate level effect and the analysis that produced it.
- Exhibits. Loss history, trend and development, expense loads and the working behind each factor.
- Sample premium calculations. Step-by-step examples for named risks. These are gold, because they show the order of operations and give you exact answers to check against.
- Regulator correspondence. Objection letters and the carrier's replies. Rates sometimes change during review, and this is where you see it.
Plenty of filings also lean on a rating bureau such as ISO, NCCI or AAIS, adopting its loss costs or rules by reference instead of printing them in full.
Where to look
Most states take filings through SERFF, the System for Electronic Rate and Form Filing run by the NAIC. Many publish them through SERFF Filing Access or their own department website. What you can search, and what you can download, differs by state, and some exhibits may be kept confidential.
Rebuilding the rater, step by step
- Trace the lineage. One filing rarely holds the whole rating plan. Usually it amends an earlier one. Walk back to the filing that launched the programme, then forward through every change to the version in force.
- Lift the tables. Get every rate table into structured rows, keeping band edges and values exactly as filed.
- Work out the algorithm. The rule manual and the sample calculations together tell you which factors apply, in which order, and where rounding and minimums sit.
- Write the rules. Eligibility becomes declines and referrals.
- Prove it. Turn every sample calculation into a test. When the model matches all of them to the cent, it is done.
- Keep the paper trail. Note which filing each table and rule came from, so any number in the model can be traced to a page.
In Swallow, steps 2 to 5 map straight onto the model format. Tables become collections, the algorithm becomes a run of calculation steps, rules become exclusions and refers, and the filing's examples become the tests saved on the version.
Where AI earns its keep
A single programme can span hundreds of pages across several documents. Tables are split over page breaks, some PDFs are scans and the rules are scattered. That is exactly the kind of reading AI is good at:
- Finding and structuring tables, including ones that run across pages, from PDFs and spreadsheets.
- Mapping the chain, by spotting which earlier filings a document refers to and which pages it replaces.
- Drafting the order of calculation from the rules and examples, for a person to check.
- Reconciling the model to the filing by running every sample premium as a test and explaining any gap, such as rounding in the wrong place or a factor that never got applied.
This is the work behind Swallow's filings platform. It has ingested every California filing since 2005 and can turn a filing into a structured rating model in Swallow's JSON format, then check that model against the worked examples in the source filing. The result runs on the same Swallow engine as any other model.
AI drafts. People still check. Every table and rule should be reviewed against the filing, and any decision the filing leaves open should be made by someone who understands the product.
Know the gaps before you start
Filings are rich, but they are not always the full story:
- Bureau content. If a carrier adopts bureau loss costs or rules by reference, the bureau manuals usually aren't in the filing. They are copyrighted and normally need a licence.
- Confidential sections. Some states let carriers withhold parts of a filing, such as proprietary scoring models.
- Third-party scores. Insurance scores, credit-based scores and other external data feed factors whose models are often not in the filing.
- Filed is not always approved. Check the correspondence and the final disposition so you build the rates that were actually approved.
- State by state. Each state has its own filing and effective date, and the same programme can differ across states.
None of these stops a build. They just need to be spotted early, then filled from the right source or covered by a written assumption.
What teams use it for
- Moving your own filed rates onto a new engine. Your approved filing is the right reference to build and test against.
- Entering a new state, starting from filings you already have and adjusting for local rules.
- MGA programmes written on a carrier's filed rates.
- Understanding the market: which factors competitors rate on, how they structure discounts and how their rate levels compare.
When you study other carriers' filings, use them to inform your own work. Your rating plan should rest on your own analysis and be filed in your own name, and bureau material should only be used under licence.
Next steps
See what a finished model looks like on our model building page, or how testing proves it prices like the source. If you have a filing you want to build from, talk to our team.
