Three in four policy records migrate untouched.
BAF Financial & Insurance, a century-old Bahamian financial services group, is consolidating about 220,000 policy records from three legacy systems onto InsureEdge, across the Bahamas, Cayman Islands and Turks & Caicos. A book of business that old carries its evidence in roughly 950,000 documents. Damco Solutions built an intelligent document processing layer that finds, reads, reconciles and scores every record — so BAF's reviewers spend their judgment only on the fields that need it.
- of routed policies reach InsureEdge with no manual touch, up from a 45–50% ceiling
- 75–80%of routed policies reach InsureEdge with no manual touch, up from a 45–50% ceiling
- peak straight-through week, validated independently by Damco QA
- 84%peak straight-through week, validated independently by Damco QA
- policy records in scope, across three territories and five product lines
- ~220Kpolicy records in scope, across three territories and five product lines
Straight-through conversion comes off the production dashboard, not a model. Live in BAF production since 1 May 2026; Damco operates the platform today.
How it works
A record earns its own migration
A century of policies means a century of paperwork: roughly 950,000 documents holding the evidence for 220,000 records. The platform finds the documents for each policy, reads them, reconciles them and decides which records need a reviewer's judgment.
A record whose supporting document is missing is not a silent failure. The platform raises a document request automatically, and the record re-enters the pipeline when the document arrives — nothing quietly disappears.
How we did it
A live migration, changed carefully
Four decisions made a migration this size move at machine speed.
A vision pass broke the extraction ceiling.
Base document extraction settled at 45–50% straight-through, with nothing left to tune. The extraction team proposed a vision-model pass over the same output, aimed at handwritten forms and older scans — exactly where base extraction loses confidence. Straight-through conversion rose to 75–80% sustained, with an 84% peak.
The live migration was never bet on the change.
The vision pass touched the extraction core of a live migration, days before mass processing was due to start. It was built and validated outside delivery hours, both versions ran side by side for two weeks, and Damco QA validated the comparison independently — confirming the new path beat the old at a lower cost per document. No delivery day was lost to it.
Every field can explain itself.
Each extracted field carries a confidence score and the source document it came from, logged. High-confidence records approve automatically into the migration pipeline, a reviewer sees only the fields that need judgment, and where a document is missing the platform raises a request rather than passing the gap along.
One new step, not a change programme.
The data sources, the review roles and the migration pipeline all stayed as they were. The platform slots in as one step in a flow BAF already trusted, which is why it went live without a change programme wrapped around it.
What it changed
A migration moving at machine speed
In one live week, the platform routed 4,802 policies across all three territories and cleared more than two-thirds of them automatically; reviewers saw only flagged fields, never whole records. Indexing runs alongside: by mid-May, about 500,000 of the 950,000 documents had been matched to their policies, and that work continues while throughput carries on. Damco has operated the platform in BAF production every working day since 1 May 2026.
- of 4,802 policies in one live week cleared automatically, across three territories
- 3,230of 4,802 policies in one live week cleared automatically, across three territories
- documents matched to their policies by the indexing agent, of roughly 950,000
- ~500Kdocuments matched to their policies by the indexing agent, of roughly 950,000
- product lines on one platform: life, health, annuity, pension and mortgage
- 5product lines on one platform: life, health, annuity, pension and mortgage
If a modernization is waiting on thousands of records nobody has time to read, the first useful conversation is about which records deserve a human at all.
We will tell you which numbers to trust.