How a rating engine turns an application into a premium
You sign off the rate pages with your pricing team, file them with the regulator, and hand them to someone else to build into the quoting system. Months later, a review of small home policies finds quotes below your own minimum premium. Every rate in the system matches the filing; the quoting system took the steps in a different order from the one set by the rate pages.
The order of the steps is part of the rate, so you write it down and test it like the rates.
The rating algorithm, step by step
A rating algorithm modifies a base rate with multipliers, addends or another defined mathematical expression. It also states how the components combine, the order in which rating variables are considered, the effect of each variable, any maximum or minimum premium, and the rounding procedure.
The demo case starts with a base rate of 420 for a home policy.
The factors that describe the risk come first and multiply: sum insured band, territory, construction and deductible. The new home and claims-free discounts are added inside one bracket and give × 0.75; taken one after the other, they would give × 0.765, so whether discounts add or compound is a pricing decision. The fee goes on after the discounts, and the minimum is checked last, so this policy's minimum of 250 never comes into play.
This follows the shape of the homeowners example in Werner and Modlin’s Basic Ratemaking.
Where the order bites
The places to specify and test are:
- Discounts can add inside one bracket, or compound as separate multipliers.
- A maximum discount or surcharge can limit the total effect after individual discounts or debits have been applied.
- The fee can be added before or after the discounts.
- The minimum premium can be applied last, or before the discounts, which can quote a policy below it.
- Rounding can happen after each operation or only at the end.
In a demo case, one small policy takes the same discounts, fee and minimum in two orders: one quote at 250 and the other at 214, below the minimum of 250.
| Order | Steps | Quote |
|---|---|---|
| Discounts, fee, then minimum | 260 → 195.00 → 185.25 → 225.25 → minimum 250 | 250 |
| Fee and minimum first, discounts last | 260 → 300 → minimum holds at 300 → 225.00 → 213.75 | 214 |
In a decision table, a missing numeric value is treated as zero. A missing sum insured or vehicle value can therefore match the lowest band, such as a band between 0 and 1000, unless a row catches the missing value deliberately. The table can test for null or for a value that is not a number. The decision table guide explains row order and matching behaviour in more detail.
Where pricing stops and an underwriter decides
Underwriting guidelines can specify that a risk is accepted, declined or referred. Werner and Modlin give the example of a household with two or more losses in the last 12 months that may not be eligible for insurance or may need review by a senior underwriter.
A risk can receive a tier, and the rating algorithm can use that tier to determine the premium. A decision engine can also write an outcome such as “refer” with a reason. The underwriter reviews the application in the insurer’s own system or in a case.
For the work that follows a referred life application, see how life underwriting turns an application into terms.
Rates change on a date
Rates are filed and take effect on a date. Policies in force usually keep the rates that applied when they started, so a policy effective on 1 March can be based on rates effective on 1 January.
Renewals can have a separate effective date. Washington’s rule permits a different renewal date when the filing requests it, subject to a maximum interval of 60 days. The rate release therefore needs a clear distinction between new business, renewals and policies already in force.
A transition rule can limit the maximum or minimum change in premium at a single renewal. In the United Kingdom, the FCA’s ICOBS 6B.2 rule says that a renewal price for home and motor insurance must not be higher than the equivalent new business price. A renewal quote has to be compared with the price for the same customer as new business.
A release takes effect when it is deployed, so the day the new rates start is the day your release plan deploys them.
Test a rate change on the book before it goes live
A proposed rate change should be run against the existing book before deployment. The extension of exposures method rerates every policy to restate its historical premium under the current rates. Dislocation analysis then quantifies how many customers receive each amount of change. If the result falls outside tolerance, the proposed change can be revised.
A decision engine can perform a similar check against past decisions. The impact analyzer runs two revisions in the same decision flow over the same rows and shows which decisions differ. The rows can come from an uploaded dataset or from the decision archive, selected by a date window and a row limit. A whole release can also be backtested against decisions already made by production endpoints, up to 50,000 rows, with 10,000 rows as the default.
Testing one changed decision table can show the effect of a specific rate table revision. Backtesting the whole release shows the effect of the complete set of deployed logic. The result can report every decision that would change, either by comparing with recorded decisions or by replaying both releases on the same rows.
The rate table itself can be pasted from Excel or Google Sheets into a decision table. A formula can express the calculation with variables, arithmetic operators and functions such as min, max and round. The minimum premium can be written as the higher of the calculated premium and the minimum, and the rounding rule can be made explicit.
The release starts producing decisions when it is deployed to an endpoint. An endpoint can require a sign-off sequence, and deployment happens only after sign-off. Each response carries the name of the release that produced it, which connects the production result to the approved version.
And claims
A claim is run as a case, a set of tasks. A task runs again when a report replaces a value it read. See how claims work as a case.
Frequently asked questions
Should discounts add or multiply?
The homeowners example adds the new home and claims-free discounts inside one bracket, then applies the multi-policy discount as a separate multiplier. The rule must state which discounts belong together.
Where does the minimum premium go?
In the worked policy, the fee is added after the discounts, the higher of the calculated premium and 250 is selected, and the result is rounded to the whole unit. Moving the minimum earlier can produce a different result.
How do I test a new rate table before release?
Run the changed table against past decisions with the impact analyzer, then test the whole release against production decisions. Review which decisions would change and the amount of dislocation before the release goes through sign-off.
Does a renewal get the new rates on the effective date?
A renewal can have its own effective date, and Washington permits a different renewal date subject to a maximum interval of 60 days. You deploy the release on the day the new rates start.
