Skip to content
PYD ENTERPRISES

Insurance · Regional US property and casualty carrier

A P&C carrier lifted 30 years of state rating endorsements out of COBOL and into decision tables actuaries could check.

The carrier's commercial auto and homeowners rating engine ran on COBOL, CICS and VSAM. PYD ENTERPRISES extracted 2,380 rating rules into DMN 1.5 tables organized by state and line of business, reviewed by actuaries in four weeks.

ENGAGEMENT PROFILEInsurance
Customer
Regional US property and casualty carrier
Code
1,140,000 LOC across 3 systems
Stack
IBM Enterprise COBOL 6.2CICS TSVSAM KSDS with alternate indexesJCLIMS DB (policy history)Endevor
rating and underwriting rules extracted
2,380
rating and underwriting rules extracted
of endorsement logic separated into per-state DMN tables
41 states
of endorsement logic separated into per-state DMN tables
to actuarial sign-off, against a 6-month estimate
4 weeks
to actuarial sign-off, against a 6-month estimate
equivalence fixtures generated for the new rating service
1,200
equivalence fixtures generated for the new rating service

The challenge

What they were up against.

The rating engine had been amended every time a state filing changed for three decades. State-specific endorsements lived as nested `EVALUATE` blocks keyed on a two-character state code, territory factors sat in VSAM tables loaded at CICS region start, and a handful of rate overrides were driven by 88-level condition names on a policy flag byte that four different programs set. The carrier planned to replace the engine with a commercial rating platform and needed every factor, table lookup and rounding step documented so the actuarial team could prove the new platform produced the same premium. The internal estimate for doing that by hand was six months of seven people.

The approach

How the work ran.

Separating the rate algorithm from the state exceptions

PYD ENTERPRISES parsed the rating, policy issuance and billing interface systems together, so a premium could be traced from the CICS rating transaction through the batch re-rate jobs to the billing extract. Program slicing on the final premium field isolated the base algorithm: base rate, territory factor, class factor, limit factor and the experience modifier, applied in a fixed order with ROUNDED at two specific steps.

State endorsements were then extracted as separate decision tables, one per state and line of business, with a DMN hit policy of FIRST to match the original EVALUATE ordering. Where a state rule depended on the policy flag byte, the 88-level condition names became named FEEL inputs, so POL-NY-SURCHARGE-EXEMPT reads as a boolean input rather than a bit test.

Actuarial review and equivalence fixtures

Actuaries reviewed the tables in the sign-off view and compared them against the carrier's filed rate manuals. They found three territory factor tables in VSAM that no longer matched the filed manual; the code had been right and the manual out of date. The behavioral test generator then produced 1,200 boundary fixtures (state, class, limit and modifier combinations at every branch edge) with expected premiums captured from the existing engine.

Outcome

The carrier finished rating discovery in four weeks instead of six months and handed its rating platform vendor a complete set of DMN tables plus fixtures. The new rating service passed the full fixture suite before parallel run began, and the parallel run itself was shortened because the edge cases had already been exercised.

“It turned a 4,000-line procedural rating routine into decision tables our actuaries could verify immediately.”
Lead Business Analyst, regional US property and casualty carrier

More stories

Want to check the arithmetic against your own estate? Run the ROI calculator.

Start with one system.

Request a Discovery Scan for one system up to 500,000 LOC, or try PYD ENTERPRISES on 100,000 lines for 14 days.