Walk into any underwriting or claims shop and you’ll find the same thing: a deeply organized folder tree. The year, the insured, the submission or claim number, the documents inside. That structure isn’t incidental — it’s how the business tracks work, and it already encodes what most tools ask you to re-enter: which account this is, which submission, what kind of run. One-way ingestion throws that away, asks a user to re-type it into an upload form, and then leaves the extracted results sitting in a separate system nobody’s process points at.
InsightXtract treats the folder tree as a first-class input — and treats delivery as a round-trip, not a drop-off.
The folder is the metadata
A folder template maps path segments to run metadata, so the structure your teams already maintain becomes the run’s context automatically. A leaf folder is one run; its path tells the platform the account, the submission, and often the line of business.
2026/
Acme-Logistics/
SUB-88431/ ← one run
broker-email.pdf
acord-125.pdf
exposure.xlsx
loss-run.pdf
# derived automatically:
year = 2026 insured = "Acme Logistics" submission_id = "SUB-88431"
No upload form, no re-keying — drop the documents where you already file them, and the run is fully contextualized.
Round-trip: results go back where they belong
When extraction finishes, the structured output doesn’t sit in a silo. It’s written back to the source — a results file into an /Output subfolder in SharePoint, an object into cloud storage, a record into the policy system — so the next person in the process finds it exactly where they’d look. Watched locations exclude their own output, so writing back never re-triggers a run.
SharePoint · S3 · GCS)] -->|new leaf folder| ING[intake
path → run metadata] ING --> EX[classify · extract ·
consolidate] EX --> WB[write-back] WB -->|/Output results| SRC WB -->|record| PAS[(policy system)]
Connectors on both ends
The same connector framework reads and writes, across the systems carriers actually use:
Why folder-native, round-trip intake matters
- Zero re-keying. The account, submission, and line of business come from the path your teams already maintain — not a form someone fills in again.
- It fits the existing process. People keep filing work where they always have; automation happens around them, not in a separate tool they have to adopt.
- The loop is closed. Results land back in the system of record, so the extraction actually reaches the next step instead of stranding in a dashboard.
- Read and write. Two-way connectors mean InsightXtract participates in your document flow, rather than being a one-way import.
Automation that meets people where they are
The fastest adoption comes from not asking anyone to change their habits. When intake is “keep saving files where you save them” and delivery is “the results appear next to the inputs,” the tool disappears into the workflow — which is exactly where extraction should be.
Read next →
See how systems call it directly instead of via folders: the agentic API. And what gets written back cleanly: resolved entity identities.