How we built and tested the insurance intake and reporting workflow

We measured the work that competed with underwriting.

To find where preparation was taking up underwriting capacity, we mapped the work from submission intake through carrier reporting. Underwriters, assistants, operations staff, and reporting staff logged their work in 30-minute increments for one working week.

We compared those logs with twelve months of submission-inbox and policy-system records. Underwriters spent ~33% of their week on intake, clearance, and missing-information follow-up, while 40% of submissions didn't fit the program but still received a full manual read.

Monthly carrier reporting took 40 staff-hours per program. That baseline covered all of a program's reporting work for the month.

One set of fields connected intake and reporting.

The participating teams used different fields, so we mapped them to an agreed set and built a shared record for each submission. That record connected the broker's email and attachments to the policy and carrier report.

The intake assistant structured the incoming information, checked program rules, flagged missing details, drafted requests back to brokers, and helped prioritize the underwriter's list. Because the carrier-report builder used that same record, staff could prepare monthly reporting without repeating entry across their systems.

AI prepared the information; their underwriters kept the decisions.

We deployed the workflow in a private environment for sensitive submission information and gave their risk team ownership of AI-use rules.

The assistant structured, checked, and prepared information, but every underwriting decision remained with an underwriter acting within the carrier's delegated authority.

We tested the workflow against their underwriters' past decisions.

Their underwriters selected real past submissions and marked the decisions we'd use as references. We tested the intake assistant and report builder on held-out submissions in one program group, so the pilot had to work on submissions reserved for evaluation.

The pilot had to deliver at least 33% less non-underwriting time per submission with no drop in agreement with their underwriters' marked decisions. It passed both requirements. The published result is a minimum reduction achieved; it doesn't specify an exact after time or agreement percentage.

Agreement with their marked decisions was the quality measure for this test. Underwriters still had responsibility for live decisions, and this test didn't measure a change in loss ratio.

We measured reporting work and profit separately.

The published results follow our founder's account of the engagement and its finance findings.

• Carrier-report work: monthly reporting started at 40 staff-hours per program and fell 75%. The 10-hour result comes from 40 × 25%, leaving 30 fewer staff-hours of reporting work per program each month.
• Non-underwriting effort: the pilot in one program group achieved at least 33% less time per submission with no drop in agreement with their underwriters' marked decisions. This measures time per submission, separately from the baseline share of an underwriter's week.
• Net recurring profit: their finance team independently measured an increase of $1.96M. We report that finance result separately because released staff hours alone don't establish a profit increase.

The pilot's time-and-agreement result applies to the program group tested. The reporting and profit figures describe outcomes of the engagement; they don't establish that every program matched the pilot's reduction.

The group also recorded $16M in additional bound premium from existing incoming submissions and $2.4M in additional commission. Commission is revenue, separate from the $1.96M net recurring profit result.

We helped their teams use and extend the workflow.

Once the pilot passed, we helped the participating teams use the workflow daily and retire repeated retyping across their systems. We monitored usage and accuracy each week as the workflow entered their day-to-day work.

Because the group needed to keep adding programs, we trained their shared-services and data teams to add programs and companies to the workflow. It became the standard for new programs, with their risk team owning AI-use rules and their underwriters retaining decision authority.