All AI product case studies

Case study 01Customer-facing insurance intake

AutoClaims Intake AI

A safety-first conversational prototype that helps drivers organize accident information one clear question at a time and prepare a review-ready summary.

AutoClaims Intake AI product avatar
Prototype statusCompleted independent prototype
My roleProduct Owner · Workflow Designer · Responsible-AI Strategist
Primary audienceU.S. personal-auto drivers and policyholders preparing accident information for an authorized insurance or claims professional.
Product boundaryIndependent prototype · Human review required

Problem framing

Start with the workflow—not the technology.

01 · Problem

The business and user challenge

After an accident, customers may be stressed, uncertain about what matters, and likely to omit information during the first reporting conversation. That creates an inconsistent experience and avoidable follow-up for both the customer and the claims professional.

02 · Hypothesis

What the prototype is designed to test

If accident intake begins with safety, routes by incident type, and asks one relevant question at a time, customers can prepare more complete information with less confusion while preserving human claims authority.

03 · Discovery status

What is known—and what is not

Workflow-based discovery and scenario design are complete for the portfolio prototype. Representative customer interviews and production-environment research would be required before implementation.

04 · Example scenario

A realistic workflow moment

A driver has been involved in a collision, has photos and partial information, and needs a calm way to document what happened before speaking with an insurer.

Product flow

From user need to a structured human handoff.

The workflow is intentionally visible so recruiters and product teams can see how the experience is designed—not only the final output.

  1. 01

    Start with safety

    Confirm immediate safety, injuries or pain, and whether emergency assistance may be needed before gathering claim details.

  2. 02

    Establish the reporting path

    Clarify the insurer and whether the driver is reporting through their own insurer or another party’s insurer.

  3. 03

    Route by incident

    Collect state, date, location, incident type, vehicles, people, property, police information, damage, and relevant documents.

  4. 04

    Preserve uncertainty

    Separate what the customer provided from information that is missing, unclear, estimated, or still needs confirmation.

  5. 05

    Prepare the handoff

    Generate a structured summary the customer can review before sharing it with an authorized claims professional.

Product artifact

A structured summary that keeps uncertainty visible.

This illustrative artifact shows how the prototype organizes customer-supplied information for review without making claims decisions or filling in missing facts.

Sample accident-intake handoff

Illustrative · Privacy-safe
Confirmed first

Immediate safety

Safety, injuries or pain, and the need for emergency assistance are addressed before intake continues.

Information supplied

Incident details

Incident type, date, location, involved vehicles, people, property, police information, and available photos are organized for review.

Needs clarification

Incomplete details

Any unclear, estimated, or partially supplied information remains visibly marked for customer review.

Professional handoff

Claims review

The completed summary is reviewed by the customer before an authorized professional handles coverage and claims decisions.

Illustrative output structure only. It contains no real customer, policy, or claim information.

AI and technical approach

Technical fluency that turns business needs into buildable product decisions.

These decisions show how I translate business needs into a buildable AI product concept, including the workflow, information structure, human review, and integration considerations.

Prototype environment

Custom GPT for workflow validation

The prototype tests conversation architecture, output structure, and guardrails before any secure claims-system integration is considered.

Conversation architecture

Incident-aware branching

Question order changes according to collision, theft, weather, vandalism, or other supported personal-auto scenarios.

Information design

Known, missing, and unclear

The output preserves uncertainty instead of filling gaps, making the final summary easier to review and correct.

Human review

Customer review before professional handoff

The assistant organizes information; the customer verifies it and authorized professionals retain all coverage and claims decisions.

Technical and product constraints

What the current prototype does not claim.

  • Limited to U.S. personal-auto accident intake; it is not designed for commercial, property, or international claims.
  • No direct connection to a carrier, policy, claims, document-management, or identity system.
  • Avoids unnecessary sensitive information and never invents facts that were not supplied.
  • Cannot provide legal, medical, coverage, liability, repair, fraud, settlement, or payment advice.

Product leadership

The decisions and tradeoffs I owned.

01

Safety before data collection

I prioritized the customer’s immediate situation over intake speed because the first product responsibility is safe routing.

02

One question at a time

I chose a guided sequence over a long form to reduce cognitive load and make missing information easier to identify.

03

Summary instead of a decision

The product prepares information for review rather than interpreting fault, coverage, or settlement.

Validation

Evidence today. Questions for the pilot.

Prototype flow, output structure, decision boundaries, and test scenarios have been designed. Formal customer testing and production pilot results are not claimed.

Evidence in the current prototype
  • Safety-first opening and required question order documented
  • Incident-routing and missing-information logic defined
  • Structured customer and professional handoff outputs designed
  • Common and edge-case test scenarios prepared
Still to validate
  • Customer comprehension and completion rates
  • Missing-information reduction
  • Accessibility and mobile task completion
  • Secure integration, auditability, and operational fit

Pilot success measures

How I would measure value responsibly.

These are proposed measures for a future controlled pilot—not results the independent prototype has already achieved.

Measure

Intake completion rate

Understand whether users can finish the guided flow without unnecessary abandonment.

Measure

Missing-information rate

Measure whether the structured workflow improves the completeness of the initial handoff.

Measure

Follow-up questions required

Evaluate whether claims professionals receive more review-ready information.

Measure

Customer confidence score

Assess whether users feel clearer about what they documented and what happens next.

Product output

A review-ready accident-intake summary that separates supplied facts, missing details, unclear information, and appropriate next questions.

Potential product value

Creates a calmer customer experience and a more consistent starting point for professional review without automating protected claims decisions.

Reflection

What I learned.

In a stressful workflow, clarity and restraint are product features. The assistant becomes more useful when it asks less at once, preserves uncertainty, and makes the human handoff explicit.

Next responsible iteration

Conduct representative user testing, measure completion and information quality, refine accessibility, and explore secure integration requirements with approved claims systems.

Explore the product

See the workflow in action.

The live agent is an independent portfolio demonstration using fictional or user-supplied sample information.

Try AutoClaims Intake AI