One pipeline, inbox to record
Six layers between a broker's email arriving and a reviewed, structured record existing in your policy administration system. Every one of them is auditable, and a person approves before anything posts.
What comes out
A typed record, every field scored, and anything uncertain routed to a person rather than guessed.
- Named insured
- Ridgeline Freight LLCHigh
- DOT number
- 3184722High
- Effective date
- 2026-09-01High
- Power units
- 42High
- Radius of operation
- 500+ milesHigh
- Prior carrier
- illegible on p.7Review
One field could not be read confidently, so it is flagged rather than guessed. A person resolves it before anything posts to your policy admin system.
Illustrative example built from representative field values. Not a customer record.
The six layers
Distinct stages, one product — because a gap between any two of them becomes manual work again.
Ingestion
Broker emails and every attachment are collected from the submissions address you already publish. No forwarding rules, no upload portal, and nothing for brokers to adopt — which matters, because any requirement placed on a broker is one they can decline by sending the risk elsewhere.
Classification
Each page is identified before anything is read from it: ACORD 125, 127 and 140, loss runs, schedules of vehicles, MVRs and program supplementals. Getting this right first is what allows the extractor to apply the correct field logic instead of pattern-matching text.
Extraction
Insureds, drivers, vehicles and claims are pulled out and typed against your field schema. VINs are decoded and reconciled against the count on the application, loss runs are itemised row by row and merged across carrier formats, and every field carries a confidence score with a pointer to its source page.
Review workspace
Extracted records land in a workspace where a person can see each field beside the document it came from, with low-confidence values surfaced first. Corrections are attributable, which is what makes the record defensible later.
Policy admin sync
Approved records are written into your system of record. Dayd ALIS is live in production, Vertafore IMS is in development, and anything else is connected through a custom API integration built against your instance.
Audit trail
Every value keeps its provenance — source document, page, model confidence and any human correction. When a rating question or an E&O query arrives two years later, the trail is intact rather than reconstructed from an old email thread.
The platform, answered
What does the RM Assistant platform actually do?
It automates the step between a broker emailing a submission and a structured record existing in your policy administration system. That means ingesting the email and its attachments, classifying each document, extracting and typing the underwriting data with a confidence score per field, presenting it for human review, and writing the approved record downstream.
Is it one product or several modules?
One pipeline. The stages are distinct but they are not separately licensed products you assemble — a tool that extracts data but leaves someone to get it into the system of record has moved the manual work rather than removed it.
What happens to documents it cannot read confidently?
They are flagged rather than guessed. Low-confidence fields are surfaced at the top of the review queue with the source page alongside them, so a person decides. The failure mode is a human look, never a silent wrong value in the system of record.
Where does our data live?
Approved records are pushed into your own system of record. RM Assistant retains the submission documents and extraction history that produce the audit trail, isolated per customer. Data is not pooled across customers and is not used to train shared AI models.
Which document types are supported?
ACORD 125, 127 and 140, carrier loss runs in varied formats, schedules of vehicles as spreadsheets or PDFs, MVRs and program supplementals. Filled fillable-PDF forms are handled including the case where values live in the form-field layer with nothing rendered on the visible page.
See the whole pipeline on your own submission
30 minutes. A real submission from your book, run end to end while you watch — including what gets flagged and why.
Book a demoNo sales pressure. We confirm within one business day.