Insurance AI Business Case

Underwriting throughput — the revenue case

Underwriting throughput improvement is the strongest first proof point because it connects directly to premium revenue.

Current state (typical commercial lines operation):

• Underwriters review 3-5 new submissions per day
• Each submission takes 45-90 minutes to triage (read ACORD forms, loss runs, supplemental applications, key data into the system)
• Team of 10 underwriters: ~35-50 submissions reviewed per day
• Carrier receives 80-100 submissions per day
• Result: 40-60% of submissions are declined or ignored without review due to capacity constraints
• Assume average bound premium of $15,000 per policy, and 20% of reviewed submissions are bound
• Revenue from reviewed submissions: 35-50 reviewed x 20% bind rate x $15,000 = $105,000-$150,000/day

AI-assisted state:

• AI triages each submission in 5-10 minutes
• Underwriter reviews AI triage summary and makes judgment call: 15-20 minutes per submission
• Underwriters now review 8-12 submissions per day
• Team of 10 underwriters: ~80-120 submissions reviewed per day
• Result: nearly all submissions receive at least initial review
• Revenue potential: 80-120 reviewed x 20% bind rate x $15,000 = $240,000-$360,000/day

Additional premium opportunity: $135,000-$210,000 per day from submissions that were previously unreviewed.

AI cost: 10 underwriting users x $50/month = $500/month = $6,000/year

Annual incremental premium potential: Even capturing 10% of the previously unreviewed opportunity = $3.4-$5.3 million in annual written premium

The ROI is not about saving underwriter salary — it is about capturing premium revenue that is currently walking out the door because your team lacks capacity to review every submission.

Claims cost reduction — the leakage case

Claims leakage reduction is the loss ratio argument, and the numbers are enormous.

Claims leakage categories and AI impact:

Example calculation for a mid-sized P&C carrier:

• Annual incurred losses: $500 million
• Total estimated leakage (conservative 8%): $40 million
• AI-assisted leakage reduction (conservative 15% of identified leakage): $6 million annually

AI cost: 50 claims users x $50/month = $2,500/month = $30,000/year

Net annual saving: $5.97 million on claims leakage alone.

Claims cycle time improvement delivers additional value:

• Faster claim resolution improves policyholder satisfaction and retention
• Shorter cycle times reduce allocated loss adjustment expense (ALAE)
• Faster FNOL triage reduces severity on time-sensitive claims (water damage, business interruption)
• Reduced litigation rate when claims are handled promptly and thoroughly

The combined ratio business case — bringing it all together

The combined ratio business case combines expense ratio improvement (efficiency) with loss ratio improvement (better outcomes).

Expense ratio impact (operational efficiency):

Loss ratio impact (outcome improvement):

• Claims leakage reduction: $6,000,000 (from previous calculation)
• Improved subrogation recovery: $2,000,000 (incremental)
• Better underwriting selection (from reviewing more submissions against appetite): $3,000,000 (estimated loss ratio improvement on incremental written premium)
• Total loss ratio impact: $11,000,000

Combined annual impact: approximately $14 million

On a $1 billion premium book, that is a 1.4 point combined ratio improvement.

AI platform cost: 100 users x $50/month = $60,000/year

This is the business case that gets board approval. A 1.4 point combined ratio improvement for a $60,000 annual investment. Even if you discount the estimates by 50% to be conservative, it is still a 0.7 point improvement — worth $7 million on a $1 billion book — for $60,000 in tool cost.

Pilot design for an insurance carrier

Do not propose company-wide AI deployment. Propose a 6-8 week pilot on one workflow in one business unit. The goal is to produce measurable results that justify broader rollout.

The ideal insurance AI pilot:

Option A: Underwriting submission triage pilot

• Team: 3-5 commercial lines underwriters in a single branch or business unit
• Workflow: Submission intake and triage for one line of business (commercial property is ideal — high volume, standardised forms)
• Duration: 6-8 weeks
• Metrics: Submissions reviewed per day (before vs. after), time per submission, bind rate on AI-triaged submissions, underwriter satisfaction

Option B: Claims FNOL triage pilot

• Team: 5-8 claims adjusters handling one claim type (auto physical damage is ideal — high volume, measurable cycle time)
• Workflow: FNOL severity assessment and claims file extraction
• Duration: 6-8 weeks
• Metrics: FNOL-to-adjuster-contact time, claims cycle time, initial reserve accuracy (measured at 90 days), adjuster caseload capacity

Before the pilot starts:

1. Measure the baseline. Track current time per submission (or per claim), throughput, and outcome metrics for 2 weeks before introducing AI. Without a baseline, you cannot demonstrate improvement.
2. Set up the enterprise AI tool with proper data handling (enterprise tier, no policyholder PII in free tools).
3. Create 5-10 prompt templates for the specific workflow (use the templates from this course as starting points and customise for your forms, appetite guidelines, and processes).
4. Train the pilot team (half-day workshop covering AI basics, prompt templates, and review expectations).
5. Define success criteria in advance. What results would justify expanding the pilot?

During the pilot:

1. Track metrics weekly (time per unit, throughput, quality)
2. Hold weekly 30-minute check-ins with the pilot team
3. Collect specific examples of AI adding value (and examples where it fell short)
4. Iterate on prompt templates based on what works

After the pilot:

1. Calculate before/after metrics with specific numbers
2. Extrapolate to the full operation
3. Document quality outcomes (did AI-triaged submissions have better or worse loss experience?)
4. Present to leadership with a rollout plan for the next line of business

Change management for insurance professionals

Insurance professionals are methodical, risk-aware, and sceptical of anything that promises to change how they work. This is not a flaw — it is exactly the mindset that makes them good at their jobs. Your change management approach must respect this.

What works in insurance:

• Show, don't tell — run a live demo on a real submission or claims file from the team's own book. When an underwriter sees AI triage their actual submission in 5 minutes instead of 45, scepticism dissolves.
• Start with the biggest pain point — ask the team what they hate most about their workflow. It will be data entry, document review, or chasing missing information. Start there.
• Position AI as a tool, not a threat — experienced underwriters and adjusters will resist anything that sounds like it might replace their judgment. Be explicit: AI handles the data extraction and document processing so they can focus on the risk assessment and claims judgment that requires their expertise.
• One champion per team — identify the most enthusiastic early adopter and give them extra time with the tool. They become the peer support for the rest of the team.
• Shared prompt library — create a team repository of tested prompt templates for common tasks. People will use AI when they have a template to start from; they will abandon it if they have to write prompts from scratch every time.

What kills adoption in insurance:

• Launching without proper data governance (compliance and legal block it)
• Starting with a low-value workflow (no visible improvement, team loses interest)
• Overpromising accuracy (first time AI makes an error that an underwriter catches, trust collapses if expectations were set too high)
• No quality review process (leadership gets nervous about AI-generated output going to brokers or policyholders without review)

Key takeaways

1. The underwriting ROI is a revenue story — AI enables carriers to review more submissions, capture more premium, and improve the submission-to-bind ratio without adding headcount.
2. The claims ROI targets leakage — at an estimated 8% of paid losses, even modest leakage reduction produces savings that dwarf AI platform costs by orders of magnitude.
3. The combined ratio business case brings expense and loss ratio improvements together — a 1+ point improvement on a $1 billion book is $10+ million in annual impact.
4. Pilot design is critical — start with one workflow (submission triage or FNOL triage), measure the baseline, run for 6-8 weeks, and produce specific before/after metrics.
5. Change management for insurance requires live demos on real files, starting with pain points, positioning AI as a tool for judgment work, and building a shared prompt library.

Next up: 30/60/90 Day Implementation Plan.