Case study: 41% faster claims at North Meridian Insurance

North Meridian processes 1.1 million claims a year. When they engaged us, all of them ran through a mainframe workflow engine written in the 1990s, understood in detail by exactly two employees, and formally end-of-support in 26 months.

Two previous replacement programmes had been abandoned, at a combined cost of roughly $14m. Both had been structured as a single cutover.

Why the previous attempts failed

We spent the first three weeks reading the post-mortems and interviewing people who had worked on both attempts. The pattern was consistent and it was not technical.

Both programmes had tried to replicate the full behaviour of the existing engine before switching anything over. Because nobody fully understood the existing behaviour, the requirements phase never converged — every month of analysis surfaced new edge cases, and the scope grew faster than the build. The second attempt was cancelled with the requirements document still open.

What we did differently

We did not try to understand the whole system. We segmented claims by type and complexity, and found that 38% of volume was low-value, low-variance auto and property claims with a narrow set of paths. We built the new platform for that segment only, ran it alongside the mainframe, and reconciled every claim across both systems automatically.

Once the reconciliation was clean for four consecutive weeks, that claim type moved over for real. Then the next one. Nineteen migration waves in total, roughly one a fortnight.

Two things made this work that are worth naming. First, the reconciliation harness was treated as a first-class system with its own team — it was about 16% of build effort, and it was the reason anyone trusted the migration. Second, the process redesign for each claim type ran ahead of its migration wave, with the adjusters who would use it. The system was configured to the redesigned process, not the existing one.

The adoption problem we nearly missed

Six weeks in, the adoption dashboard showed workaround rates climbing in one regional team — adjusters were completing claims in the new system and then re-entering key data in a spreadsheet. The reason, when we asked, was that a report their manager relied on had not been migrated and they were reconstructing it by hand.

That is a two-day fix if you find it in week six. It is a cultural fact about the new system if you find it in month nine. The only reason we found it was that workaround rate was an instrumented weekly metric rather than something surfaced in a steering committee.

Results

All claim types migrated in 19 months, four months inside the vendor's end-of-support deadline. Average cycle time fell from 16 days to 9.4. Straight-through processing rose from 4% to 31%. The mainframe was decommissioned on schedule, removing $2.4m of annual run cost — and decommissioning had a named owner from the first week of the programme, which is why it actually happened.

Twelve North Meridian engineers were embedded in the delivery team throughout. They run the platform today. We have not had a support ticket from them in eleven months.