Blueprints Financial Services & Insurance Carriers · MGAs
Fewer than half of broker submissions ever get read. Submission Flow puts five agents at the front of the funnel. Routine submissions clear autonomously. Underwriters drive the cases that need judgment, and curate the appetite layer every cycle.
Start a SprintThe submission that waits is the one that places elsewhere. The one that binds for the wrong reason is the one nobody had time to read.
It is a reading problem before it is a pricing problem.
Combined ratio is set here, months before any rate decision is made. Intake and Triage clear submissions against appetite in minutes. Risk Focus and Pricing Support assemble the evidence the underwriter needs to make the call. Bind & Renewal closes the loop and watches drift before the next cycle.
The context
The funnel fact
Underwriters review under half of broker submissions, quote on 20%, bind only 3%.Genpact · SortSpoke commercial-lines benchmarks The bottleneck is not pricing. It is intake. Brokers place the business elsewhere while submissions wait in a FIFO queue. Roughly 30% of submissions drive 80% of premium growth, and most never reach the right desk in time.
The expense fact
Lloyd's expense ratio rose to 35.6% in FY 2025, and the market entered a softening cycle.Lloyd's 2025 Full-Year Results · LMA / ICMR Insights Commercial underwriters spend roughly 40% of their time on administrative work: reading PDFs, rekeying ACORDs, hunting loss runs. Acquisition and operating expense compounds when pricing has nowhere left to run.
The window, already open
AI leaders in insurance have outperformed laggards by 6.1× total shareholder return over five years.McKinsey · Earnix industry meta-analysis McKinsey estimates gen-AI could unlock $50–70bn in industry revenue. Carriers running advanced underwriting analytics are already reporting 3–6 percentage-point combined-ratio improvements and 2.1–4.2 point loss-ratio gains. The gap is being set now.
Submission Conversion
3% bound
Of all incoming broker submissions, fewer than half are even reviewed; the bind ratio collapses to roughly three percent. The good business places elsewhere.
Genpact · Specialty carrier benchmarks
Underwriter Capacity Drag
40% admin
Time commercial-lines underwriters spend on document handling, rekeying, and intake hygiene rather than risk selection or pricing judgment.
SortSpoke · Industry surveys, 2024–2025
Combined Ratio Lever
3–6 pts
Reported combined-ratio improvement from carriers running AI-enabled underwriting analytics. Loss-ratio improvements run 2.1–4.2 points on top of expense gains.
Earnix · McKinsey AI-in-Insurance
AI is not coming for your underwriters. It is coming for your combined ratio. The question is whether it arrives as your weapon, or your competitor's.
The math
What it costs to wait
A submission that waits four days for a quote loses roughly two-thirds of its bind probability. On a £200m specialty book at a 3% bind ratio, every point recovered at the top of the funnel returns ~£60m of additional in-appetite premium reaching an underwriter desk. Most of it never gets there today.
What prevention is worth
Carriers running advanced submission triage report up to 2.8× more high-appetite submissions quoted and 3.7× more bound, with quote times compressed by up to 89%. Combined ratio shifts 3–6 points on the back of better selection, not lower price.
The compounding return
Every bound risk and every loss runs back into the appetite layer. The class that surprised the book last year flags automatically this year. Patterns the lead underwriter approves become rules. Selection compounds before pricing does. That is why programmes that bolt AI on top of legacy workbenches do not replicate it.
The CUO view
The agents do the work. This is the surface where the work shows up. Industry-benchmarked baselines, joint targets refined with a partner in Sprint, illustrative numbers drawn from public benchmarks. The design is the deliverable.
Submission Flow · Live
CUO Scorecard · Trailing 30 days · Updated daily
Combined Ratio
88.2%
↓ 4.1 pts vs. baseline
Baseline 92.3% · book mid-cycle
Loss Ratio (Attritional)
46.8%
↓ 2.6 pts vs. baseline
Lloyd's market 47.9% · FY25
Quote-to-Bind
2.4d
↓ 7.6 days vs. baseline
Was 10 days at intake review
Hit Ratio (in-appetite)
38%
↑ 14 pts vs. baseline
Federato adopter cohort 2.8×
Combined-Ratio Trend · 12 weeks
Latest 88.2%
Baseline 92.3% (Sprint) · Joint target 85% (Enable)
Agent Activity · Last 24 hours
Tier 1 autonomous · 84% clear without underwriter review
Quality drivers · What changed the curve
In-appetite hits surfaced
Out-of-appetite declined <1h
Loss-run gaps flagged pre-bind
L3 appetite patterns active
Selection is the lever. Combined ratio follows it.
In-appetite premium reached desk
£14.2M
Modeled against a £200m book at the funnel uplift observed in trailing 90 days.
Numbers illustrative · drawn from public benchmarks. The first 90 days of a Provectus engagement establish the actual baseline before any target is committed.
The Blueprint
Each agent shares state with the next, so a Risk-Focus finding strengthens Triage on the next submission. Systems of record do not change. Pick a workflow, then any agent, to see how it routes.
Select a workflow
Layer 01 · Accountable
Accountable · decides · liable
Layer 02 · Provectus Specialist Agents
Reviewed · routed · audited
Layer 03
Submission Flow adapts. No replatform.
Agent
Extracts named insured, exposures, schedules, and loss runs from broker emails, ACORDs, SOV spreadsheets, and PDF appendices. Normalises into your data model and dedupes against in-flight inventory. Addresses the most common failure at the top of the funnel: the 50%+ of submissions that today never get reviewed because no one had time to open the attachments.
What it reads and writes
Where the underwriter drives
Routine ACORD-clean submissions clear at Tier 1 confidence and never hit the queue. The underwriter sees what arrived ambiguous: free-form loss narratives, schedules that conflict with the cover note, broker submissions that span lines. Pattern candidates surfaced here require underwriter approval before they become appetite rules.
In this workflow
Full pipeline. Five agents, top-of-funnel through bind, with renewal watch and accumulation feedback. The underwriter owns selection and pricing exceptions; agents handle structured extraction, appetite matching, and pre-bind compliance gates.
Knowledge Base · The Keystone
Rule-pack vendors ship static logic. Provectus ships the refresh loop and a three-tier rule structure. Underwriters curate the rules. The carrier owns the resulting KB.
Tier 1 · Universal
Lloyd's bylaws, PRA / FCA guidance, OFAC and sanctions lists, OECD jurisdictions, NAIC model bulletins. Common across every carrier. Updated as regulations change.
The structure
Three tiers, layered. Tier 1: market regulation, sanctions, OFAC, Lloyd's bylaws. Tier 2: class-of-business and peril-specific rules (wording libraries, exposure schemas, peer comparables). Tier 3: institution-tuned, learned from your appetite and your loss history. The system recommends. Underwriters curate.
Tier 2 · Class & peril
Class-of-business rules, wording libraries, peril-specific schemas (cat, cyber, marine, casualty, D&O), broker league tables, market peer comparables, exposure rating context.
The lever
Every submission, every quote, every bound risk, and every loss runs back through the KB per class × peril. The loop runs on five steps: Submit, Observe, Diagnose, Update, Deploy. Underwriters review what the system learned before it becomes a rule. The system compounds.
Tier 3 · Institution-tuned
Patterns from your bound book and your loss history. Appetite carve-outs, broker performance scores, treaty cession rules, the wording your team negotiates that the next carrier does not have. The carrier owns this tier at the end of the engagement.
The asymmetry
Rule packs are copied inside a quarter. An institution-tuned refresh loop is not. The KB carries the wording your team negotiates, the broker performance signal nobody else has, and the appetite carve-outs your underwriters built over years. The carrier owns it at the end.
fig. 12 · KB refresh loop
the loop compounds
What incumbents ship
Vendor-defined logic. Updated when the vendor updates it. Common across every carrier who buys it. Copyable inside a quarter. The next vendor ships the same pack with a different skin.
What Submission Flow ships
Tier 3 is yours, learned from your bound book and loss history. The system recommends rules; your team approves. Weekly refresh per class × peril. The KB carries patterns no rule pack can ship. You own it at the end.
North Star · The Reimagination
Underwriting today is a queue of submissions across email, broker portals, the workbench, the rating engine, exposure tooling, and treaty ledgers, plus a shared drive of Excel SOVs, scanned loss runs, and PDF wordings. The reimagination is one chat that reaches every system and every loose file underneath.
One conversational control plane
Claude Cowork
direct the work · audit every agent decision
Structured systems
System
UW Workbench
Quotech / Risqué · policy lifecycle
System
Broker Portals
submissions · clearance · RFIs
System
Reinsurance Ledger
treaty · fac · cessions · recoveries
Loose files
File
SOV spreadsheets
schedule of values · TIVs
File
Loss runs
scanned PDFs · broker exports
File
Wordings & endorsements
negotiated language · per deal
Human oversight
Every agent action is assigned a confidence tier. The tier determines whether it proceeds autonomously, flags for review, holds for approval, or escalates to an SME. The UW manager sets the thresholds.
Tier 1 · Confidence >95%
Agent proceeds and logs for audit. The 99% of routine work that clears at high confidence never reaches the queue.
Standard ACORD submission, in-appetite class, clean loss runs, no sanctions hit
Tier 2 · 75–95%
Agent acts and flags for review. Underwriter can approve, modify, or override. Pipeline does not block.
Minor SOV inconsistency auto-corrected, flagged for verification at next touch
Tier 3 · <75%
Agent pauses for explicit underwriter approval. The case surfaces with full reasoning, citations, and recommended action.
Edge-of-appetite class, novel cat exposure, broker performance below cohort threshold
Tier 4 · Novel / compliance
Escalates to an SME: chief actuary, compliance, treaty buyer, or legal. Full audit trail attached.
Multi-jurisdiction sanctions edge case, novel cyber wording, treaty conflict
Key use cases
UC-01
Broker submissions arrive in any format: email body, ACORD PDFs, SOV spreadsheets, loss-run scans. Intake extracts and normalises; Triage scores against appetite and broker performance; Clearance runs sanctions and license checks. Standard packets clear at Tier 1 and route autonomously to the right desk. The underwriter sees a rank-ordered queue with a recommendation and a citation. Target: out-of-appetite declined under one hour, in-appetite at desk same day.
UC-02
Risk Focus surfaces class-specific risk drivers from the submission, your historical book, and external data, using GraphRAG to keep every claim auditable to source. Pricing Support assembles peer comparables, technical-price context, and exposure rating inputs. The actuary or underwriter prices; the agent gathers the evidence. Target: combined ratio improvement 3–6 pts; loss ratio improvement 2.1–4.2 pts.
UC-03
Pre-bind compliance gate validates wording, endorsements, treaty fit, and accumulation. Bind agent issues policy documents, MRC slips, and bordereaux entries. Reinsurance cessions auto-route to fac or treaty per your rules. The underwriter signs off; the agent prepares the package. Target: bind-day documentation complete on first pass above 95%.
UC-04
Three knowledge layers injected into every agent prompt at runtime. L1: market and regulation. L2: class, peril, and broker. L3: your appetite, your loss history, your negotiated wordings. Renewal watch flags exposures drifting against treaty capacity and accumulation tolerance. The UW manager curates which Layer 3 candidates promote to active rules. Target: zero accumulation surprises at the next renewal cycle.
Side by side
Underwriting timeline · where losses surface today
Phase · click any node
What goes wrong
Cost
The information to prevent this loss existed at submission. No one had time to read it.
Outcome
Roughly half of submissions never reviewed in time. The bound risk that did not match appetite produces the loss months later.
The information to prevent this loss existed at submission. No one had time to read it.
The bet
The best underwriter on your team carries appetite patterns, broker intuition, and class-specific tells that took a decade to learn. Submission Flow extracts that knowledge, encodes it in three layers, and lets every agent act on it. Then it learns from what it gets wrong.
The reading
The funnel is the lever, not the rate. Submission Flow's inline citations link every agent decision to source: ACORD field, SOV row, loss-run line, treaty clause, market peer comparable. Auditable. Defensible. Consistent across the team.
The posture
Submission Flow sits above your UW workbench, PAS, exposure tooling, and reinsurance ledger via APIs and MCP. You do not replatform. Quotech stays Quotech, Risqué stays Risqué. The agents extend what your underwriters already do inside them.
The asymmetry
Every bound risk and every loss becomes a pattern candidate. Every underwriter-approved pattern becomes a rule. The next submission is screened against a smarter book. The learning loop requires the agentic architecture to exist first. That is why programmes that bolted AI on top of legacy workbenches cannot replicate it.
The window
Pricing softened in 2025. Lloyd's HY 2025 combined ratio rose to 92.5%, expense ratio to 35.6%.Lloyd's HY 2025 Results The carriers building agentic prevention before the next hard market arrives carry the operational advantage when the cycle turns.
The risk
What we don't
Your specific appetite carve-outs, your broker league table, your historical loss taxonomy, the Layer 3 patterns your team has accumulated over years. We find these out together, on one specialty line, before we claim anything further. Sprint produces a confidence range, not a promise.
What we own
Submission Flow sits above your UW workbench, PAS, exposure tooling, and reinsurance ledger, adapting to what you run today. Routine submissions clear autonomously with full audit trail; Tier 3 and Tier 4 escalations require underwriter sign-off. Regulatory controls (PRA, FCA, Solvency II, NAIC) stay yours. The accountability line stays with your team.
The engagement model
Three phases. One specialty line in Enable. Provectus operators join your underwriting team and run old and new in parallel through one renewal cycle. We scale only once the comparison favours Submission Flow on every metric.
01
Weeks 1–2
Assess and pick one specialty line.
02
One renewal cycle
Run old and new in parallel. Same submissions.
03
Cycle over cycle
Scale across lines. Own the outcomes.
Fig. 07 · Enable phase · one specialty line, two workflows, one renewal cycle
Source · Sprint output
Same broker mix and class taxonomy split into two parallel runs.
Track 01 · Today
Customer underwriting team. Same specialty line. Same renewal cycle. Run as it runs today, no changes to existing process or systems.
Today's result
Scorecard A
Track 02 · Submission Flow
Provectus operators on the customer crew. Five agents with the three-tier KB. Same line, same cycle, parallel run.
Submission Flow result
Scorecard B
Compare · head-to-head
No progress without proof
Provectus operators sit on the crew. We run one specialty line with your team through one renewal cycle. The same submissions produce two scorecards. The head-to-head decides whether we scale, measured on combined ratio, loss ratio, hit ratio, and quote-to-bind.
Commitments
Bounded confidence
One specialty line in Enable. One renewal cycle of parallel runs. Head-to-head on combined ratio, loss ratio, hit ratio, and quote-to-bind. We scale only when the comparison favours Submission Flow.
Honest unknown
We do not know the ceiling of agent accuracy on your book until we run them against yours. The fit of your Layer 3 pattern library, the broker performance signal, the wording your team negotiates. Sprint returns a confidence range. Enable turns it into evidence, or it doesn't.
Named trade-off
Submission Flow sits above your workbench and ledger, not inside them. We accept that constraint. It keeps the regulatory audit story clean and the adoption path short. It also means we will not chase problems that belong inside Quotech, Risqué, or your PAS.
Next Step
If Submission Flow looks like the right shape for your team, the next step is a Baseline Assessment. One to two weeks. Read-only. No workflow change for the underwriter. We measure the funnel by class × broker, identify the beachhead slice, and decide together whether to proceed. If the assessment does not support a strong business case, we say so openly and recommend the right next step.