Why We Built InsightUW: An AI-Native Workstation for P&C Commercial Underwriting

Commercial underwriting in 2026 still runs on email threads, ACORD PDFs, screenshots, and tribal knowledge. We didn’t build InsightUW to bolt AI onto that — we built it to replace the workstation, so the AI is the foundation, not a sidecar.

The submission gauntlet hasn’t changed in 20 years

Walk into any commercial P&C carrier today and watch a new submission land. A broker emails an ACORD application, loss runs, a SOV, and a few back-and-forth clarifications. From the moment the email hits the shared mailbox to the moment the underwriter has a quoteable risk in their workstation, you’re looking at:

• Intake & routing — a person manually triages mailboxes, opens attachments, files them into folders, and assigns the submission to the right line of business team.
• Clearance — another person runs OFAC checks, looks for duplicates and broker conflicts, verifies capacity, territory, and licensing — usually in PDFs and screenshots, sometimes from memory.
• Risk data assembly — an analyst copies values from the ACORD into the rater, retypes the schedule of values, geocodes addresses, pulls a loss summary from a separate system, and hands a packet to the underwriter.
• Underwriting decision — the UW reviews, asks for missing info, negotiates, and ultimately writes a quote in a tool that has no memory of how the answer was assembled.

Across these four phases, the average submission touches 6 to 12 systems and 3 to 5 people. The actual underwriting judgement — the part that needs human expertise — takes a small fraction of the elapsed time. The rest is logistics.

The wrong answer: bolt a chatbot onto the PAS

Most carriers we talked to had already tried the obvious thing — license an AI assistant or a document extractor, integrate it into an existing PAS, and hope productivity goes up. The pattern we saw repeatedly:

• The AI sees a slice, not the workflow. A document extractor that pulls fields from ACORDs doesn’t know which mailbox the submission came from, which broker has a pending BOR conflict, or that capacity in this territory is already over for the quarter.
• The PAS sees rows, not work. Legacy policy admin systems were built to hold the policy of record. They don’t model intake, clearance, triage, or the day-to-day work that produces a submission. So the AI gets bolted onto a layer that doesn’t have the right primitives.
• The audit trail is a memo. When an AI does propose a recommendation, it ends up as a free-text note in a comment box. Six months later, when compliance asks "why did we bind this risk at these terms?", the trace is gone.

You can’t fix that with a smarter model. It needs a different host.

What we built instead

InsightUW is an AI-native commercial underwriting workstation. By "AI-native" we mean three concrete things:

• Every workflow stage is owned by an agent. Submission intake, clearance, broker conflict detection, data assembly, peer benchmarking, quote follow-up — each is a bounded agent with a specific job, a specific output schema, and a specific authority level. They don’t replace the underwriter; they remove the friction.
• The workstation owns the contract. When the workstation needs broker history, claims data, or market intelligence, it calls the agent it owns — the agent talks to whatever upstream system has the data. Swap the upstream, no UW page changes.
• Every decision is auditable. Every agent step, every rule applied, every confidence score, every override is logged with timestamp, user, and reason. When compliance asks six months from now, the trace is a query, not an archaeological dig.

The six pillars

InsightUW is organised around six product pillars, each owning a piece of the underwriting lifecycle:

Where the agents change the work

The 30+ modules across these pillars are built on a layer of AI agents that handle the parts of underwriting that have always been logistics. A few examples of what changes:

None of these replace the underwriter. The underwriter still owns the decision, the negotiation, and the relationship. What changes is what they spend their day on. Less retyping. Less screenshot collection. Less "what was the right SIR last time we saw this risk?" Less "who’s broker of record now?"

Who we built this for

InsightUW is built for commercial P&C carriers writing — or wanting to write — in the lines where submission volume, document density, and renewal complexity create operational drag. Lines where a clean intake matters more than a clean PAS UI. Specifically:

• Excess & Surplus carriers — high-volume intake, low-margin work, brokers who change relationships often.
• Specialty commercial — document-heavy submissions, lines where the SOV is half the risk picture and the loss runs are the other half.
• Standard commercial — carriers modernising off a legacy PAS without a forklift replacement, wanting to add an AI-native intake layer that talks to what they have.
• Programs & MGAs — teams running multi-program books with bespoke clearance rules per program.

What’s next on this blog

This is the first post in a series. In the coming weeks we’ll publish deep dives on specific agents and modules — how clearance actually runs, why the integration hub catalogues data instead of moving it, how the workflow engine treats LOBs as configurations rather than code, how the audit trail makes compliance a query. Each one will be a working note, not a marketing page.

If you’d like to see the workstation against your own data — a real broker email, your line of business, your clearance rules — we run a 45-minute working session. Get in touch.