Clearing ledger rebuild

A mid-market insurer was reconciling premium and claims across three overnight batch jobs and a workbook that only two people understood. Month-end took eleven days. Audit asked for provenance the system of record could not produce. Every close was a reconstruction, not a report.

The incumbents were not leaving. File-based ingest from the policy admin system had to stay for at least a year. The finance team would not accept a stack they could not explain to the examiners.

We replaced the batch chain with an append-only ledger and a small set of posting rules. Ingestion stayed file-based; the change was the accounting, not a streaming rewrite. Every posted amount carries a source document and a rule version. Exceptions land in a queue the two spreadsheet keepers already knew how to work.

The same team owns the jobs, the schema, and the close checklist. Promotion to production uses the same path as every other system we ship. We sat the close with them for two cycles before calling the engagement done.

Close dropped from eleven days to three. Every posted amount has provenance an examiner can follow. The two people who used to rebuild the month now review exceptions. Finance can answer audit from the ledger instead of from memory.