Article / Insurance

Underwriting Still Starts in the Inbox. The Operating Model Shouldn’t.

The underwriting opportunity is not simply to extract data from submissions faster. It is to redesign how intake, evidence, authority, pricing, and judgment move from broker submission to bind.

Ravi MisraCOO, mantra.ai

Advises insurance and healthcare leaders on AI-enabled transformation, digital platforms and operating-model modernization

August 12, 20269 min read

The inbox can remain a broker channel. It should stop being the carrier’s operating system.

A broker submission arrives. An underwriter opens the email, reviews attachments, extracts facts, checks what is missing, searches internal systems, asks follow-up questions, evaluates the risk, accesses pricing tools, records a decision, and moves the case forward.

A great deal of sophisticated insurance activity still begins with a person interpreting an unstructured package. The broker is working in the channel that works for the broker. The operating problem is what happens after that submission enters the carrier.

Point automation can extract, classify, enrich, or price parts of the case while the underwriter still coordinates the work. The opportunity is to turn the submission into a decision-ready case before expert attention is consumed.

The goal is not to eliminate the inbox. It is to stop unstructured intake from dictating the structure of underwriting work.

Move from extracted fields to a decision-ready case.

The unit of work should not be a document. It should be a case that carries structured submission data, source evidence, completeness, outstanding questions, appetite signals, enrichment, pricing inputs, authority requirements, and the next action.

That shared case state makes it possible to route different risks through different levels of automation and human judgment.

Document automation

Classifies attachments and extracts fields. Valuable, but still leaves the underwriter to reconstruct the case.

Decision preparation

Assembles evidence, missing information, appetite, enrichment, and pricing context around the decision.

Underwriting operating model

Routes cases by complexity and authority while maintaining governance, portfolio context, and feedback across the full journey.

Not every risk should travel through the same operating lane.

01Straight-through

Clean, well-understood risks inside appetite and authority thresholds can move automatically with evidence retained.

02Assisted decision

The system prepares the case, applies rules, retrieves context, and recommends a next action. A human reviews and owns the decision.

03Expert judgment

Complex, ambiguous, strategic, or high-value risks remain human-led while administrative preparation is removed.

04Authority and escalation

Confidence, materiality, exceptions, and delegated authority determine when a case moves from one lane to another.

05Portfolio feedback

Case outcomes feed back into appetite, pricing, concentration, broker performance, and future recommendations.

A modern underwriting system protects expert attention by making human judgment the scarce resource it is.

Automating tasks without redesigning the case flow leaves the operating model intact.

OCR can work. A pricing model can work. An appetite engine can work. A document classifier can work. Underwriters can still spend their day reconciling outputs, chasing missing data, moving cases, and reconstructing context.

The failure is not the model. It is the absence of one controlled state for the submission.

Extracted but incomplete

Fields are available, but missing information and contradictory evidence are not resolved as part of the workflow.

Intelligent but unauthorised

A recommendation exists without a clear rule for who can act on it or when human review is mandatory.

Fast but fragmented

Automation accelerates individual steps while queues, handoffs, and exception paths remain manual.

Case-by-case but portfolio-blind

Individual decisions do not continuously reflect or improve portfolio-level appetite, concentration, and performance context.

The interface should protect judgment, not reproduce administration.

What arrived

A normalized view of the submission, documents, source evidence, and data quality.

What is missing

Explicit gaps and automatically prepared requests for clarification.

What matters

Appetite, exposure, prior decisions, external enrichment, and material risk signals.

What the system recommends

Pricing or routing recommendation with evidence, confidence, and relevant policy.

What the underwriter owns

The decision, override rights, authority limits, and broker-facing judgment that cannot be delegated.

Design around underwriting attention, not around AI feature coverage.

The strongest operating model removes repetitive extraction, re-keying, validation, status chasing, document assembly, and standard follow-up while increasing time spent on risk judgment, broker negotiation, portfolio construction, and complex pricing.

Human ownership should become more explicit as automation increases. The system should know which decisions it can make, which require review, what evidence must accompany a recommendation, and who owns an adverse outcome.

01Prepare the case before routing it

Normalize intake, evidence, gaps, enrichment, and appetite before consuming expert capacity.

02Route by complexity

Use risk, confidence, materiality, and authority to choose straight-through, assisted, or expert-led lanes.

03Make human control explicit

Encode approval rights, overrides, escalation, evidence, and audit history rather than relying on informal review.

04Connect case and portfolio

Feed outcomes back into appetite, pricing, broker, concentration, and performance decisions.

We build the connected operating layer behind the outcome.

Intelligent intake

Convert emails and attachments into structured, evidence-linked case state.

Decision orchestration

Coordinate appetite, enrichment, pricing, authority, referrals, and next actions.

Core integration

Keep policy, pricing, document, CRM, and data platforms in place while connecting the workflow between them.

Governance

Make human review, delegated authority, exception paths, and evidence part of the operating model.

Evidence standard

Point of view

A practitioner point of view informed by current commercial-lines research on AI-enabled underwriting operating models. The piece distinguishes the source research from Mantra’s operating-model interpretation.

  1. The future underwriting operating system: From inbox to AI nerve centerMcKinsey & CompanyJune 2026 research on commercial and specialty underwriting moving from fragmented inbox workflows toward machine-first, human-governed operating models.
The Insight Loop

A sharper read on AI, workflows, and the systems reshaping enterprise performance.

A concise field note on AI, systems, and the workflows shaping enterprise performance.

By subscribing, you agree to receive The Insight Loop. Unsubscribe at any time. Privacy.

Underwriting modernization

Start with the part of underwriting where expert attention is being spent on coordination.

Map one submission path from intake to decision and identify what should be automated, assisted, or explicitly human-owned.