Introducing Inbox Agent: The First Step of Underwriting, Handled Before Anyone Opens the Email

Today we're launching Inbox Agent, the first step of underwriting handled before anyone opens the email. As submissions arrive, Inbox Agent reads every message and every attachment, classifies the documents, extracts the data, and checks the submission against your appetite. Underwriters open their workbench to a cleared desk, not an inbox.

Every underwriting decision begins in a shared inbox. New emails, replies on existing chains, forwarded threads, and attachments in every format imaginable all land in the same place. Before anyone can assess a risk, someone has to work out what came in, what it relates to, and whether it's worth pursuing. That first pass is the work Inbox Agent takes on.

The submission inbox is where the underwriting day disappears before underwriting starts

Shared inboxes are messy by nature. Whether an SOV arrives as its own file or buried as inline content depends on a broker's email client. Scanned attachments carry filenames like "scan0042.pdf." One reply chain can be the fifth update on a submission already in progress; another, under a familiar subject line, can be an entirely new risk.

Sorting all of that eats hours at the top of the day before any underwriting happens, and those hours are hours a book doesn't move forward. Instead of evaluating submissions, underwriters end up triaging documents.

What Inbox Agent does inside the email

Across every message, Inbox Agent reads the email and every attachment it contains, scanned or native, regardless of file name. Each document gets classified by type, the fields that matter get extracted, and the submission gets checked against your appetite rules. By the time an underwriter picks up the file, the structured view they would otherwise build by hand is already there.

Appetite comes in as a Skill written in plain language, so the rules your team actually underwrites to are the rules Inbox Agent applies. In-appetite submissions move forward with their data already keyed. For out-of-appetite submissions, the flag arrives with a reason tied back to the specific guideline, so a human review takes seconds instead of a full read.

Threading intelligence keeps the inbox from double-counting itself

Shared inboxes also need memory. Four messages deep on an existing thread, a reply is an update on a submission already in flight. With no history behind it, a new email is a new risk. Treating both the same way is how an operation ends up with duplicated submissions, missed updates, and clearance work that quietly becomes a second job.

Inbox Agent sees the difference. Replies get routed to the submission already in progress, carrying their new information into the existing file. New emails open as new submissions, with their own clearance checks and their own place on the workbench. Nothing gets duplicated and nothing gets missed.

When an underwriter opens the workbench, what they see is submissions already read, classified, and cleared. The inbox stops being a to-do list and starts being a starting point.

If you want to see Inbox Agent running against your own submission inbox, get in touch.