Cognitive Underwriter
A pre-quote triage layer that scores submissions against carrier appetite and historic loss data, so producers spend their hours on the accounts most likely to bind.
[ capabilities ]
- —Appetite matching across your carrier panel
- —Loss-run parsing and risk scoring
- —Deploys in under a week with no AMS migration
[ core systems & integration architecture ]
Reads submissions out of the policy system, scores them in an isolated environment, and writes the triage result back as an activity or note — never as an unapproved bind.
Guidewire PolicyCenter
Native REST APICloud Submission and Account APIs pulled per new submission
Duck Creek Policy
Native REST APIAnywhere Managed Integrations for submission and appetite reads
Applied Epic
Native REST APIEpic API reads the opportunity, writes back an activity
Salesforce FSC
Embedded SidecarTriage score rendered inside the producer record page
Sync model
Real-time on submission creation, typically under 20 seconds
Standards spoken
Time to connect
Days — read-only pilot in under a week
Enterprise guardrails
- —No customer data used to train models — zero-retention inference only
- —OAuth 2.0 / mTLS against your existing identity provider
- —Field-level audit trail on every value written back
- —Underwriter approval required before any write to the policy record
[ mcp microservice interface ]
This module ships as an MCP microservice — underwriting-mcp. It consumes governed tools from the core-system adapters and publishes its own, so your assistant, orchestrator, or another module can call it directly. No credentials ever reach the model runtime, and every write lands in the human gate.
Consumes from core systems
- gw_get_submission
- gw_check_appetite_signals
- epic_search_client
- acord_parse_form
Provides to other agents
triage_submissionreadScore a submission against appetite and return the reasoning
explain_declinereadPlain-language rationale for a declined or referred risk
stage_triage_resultwrite · human gateWrite the triage outcome as an underwriter activity
[ implementation playbook ]
How an agency actually stands Cognitive Underwriter up — read-first, human-gated, and measured from day one.
Setup, in order
- 1
Pick one line of business to pilot — small commercial BOP or workers' comp works best — and export 30 days of submissions as the baseline.
- 2
Connect read-only access to your AMS or submission inbox so the module can see incoming submissions as they arrive.
- 3
Write down your appetite rules in plain language (class codes, revenue caps, states, loss-ratio limits) — the module drafts against these, not vibes.
- 4
Run shadow mode for two weeks: the module scores and summarizes every submission, but producers keep working exactly as they do today.
- 5
Compare the module's triage order against what your producers actually quoted; tune the appetite rules where they disagree.
- 6
Turn on the daily triage digest for the pilot team, then expand to other lines once the hit rate holds for a month.
Agency use cases
Monday-morning submission triage
Instead of producers opening every email in the shared inbox, the module delivers a ranked list: quote these five today, these three need more info, decline these two with reasons drafted.
Appetite check before the carrier call
A producer about to pitch a account runs the risk through the module first — it flags which of your appointed carriers fit the class code and which will decline, saving a wasted market call.
New-producer ramp
Junior producers learn appetite by seeing why the module ranked a submission the way it did — every score carries the reasoning, so training happens inside the daily workflow.
The first win to aim for
Cut average time-to-first-touch on new submissions from days to same-day for the pilot line.
Guidance by agency role
Producers
Review the ranked queue before prospecting blocks, challenge appetite scores with field context, and record why you pursued or passed so the model learns from real placement decisions.
Account managers
Confirm submission data and missing items before a risk reaches market, then use the summary to keep producers and clients aligned on next steps.
Agency administrators
Own appetite-rule versions, carrier access, and the weekly accuracy report; expand lines only after the pilot meets its agreed hit-rate and response-time targets.
[ built for ]
[ where this sits in the vendor market ]
We recommend buying first wherever the market already solves a problem well. Here is what you can buy instead of building this — and where this module earns its place beside what you buy.
Products that already cover this ground
Distribution network connecting brokers and MGAs to carriers
Buy it when: Specialty writers and MGAs that need structured access to carrier capacity.
Where it falls short: Mainline agencies whose panel is already handled through direct relationships.
Commercial lines quoting and underwriting workflow
Buy it when: Organizations with written appetite and guidelines they want enforced inside a system.
Where it falls short: Small retail agencies with no guidelines library to encode.