PASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTED
PASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTED
PASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTED
PASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTED
PASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTED
PASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTED
PASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTED
PASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTED
PASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTED
PASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTED
PASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTED
PASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTEDPASSWORD PROTECTED

Protected Content

This project is password protected. Please enter the password to view.

Case Studies

A Unified Claims Platform

Equipping insurers with an AI-native platform to process claims, deployed at scale to bring clarity and speed to complex multimodal workflows.

Role
Founding Designer
Company
QuantCo
Timeline
2025 —
Focus
AI-native, Product Strategy
A Unified Claims Platform - project cover

The Big Questions

What does AI look like when embedded across every stage of insurance claims processing?

What does it mean to design an AI-native claims platform from scratch?

How do you create clarity across tools, teams, and workflows when none exist?

Framing the challenge

I led the design of the QuantCo Platform from 0 → 1, working closely with engineering leads, AI researchers, and product leadership to transform a fragmented toolset into a coherent product surface. The design challenge was less interface design, more strategic reframing of how our technology was positioned and delivered.

At the time, the tooling landscape was fractured: internal prototypes stitched together by scripts, powerful tools siloed into specific teams or clients, and every project rebuilding the same scaffolding from scratch instead of assembling it from existing work. The product didn’t exist yet. But the ambition was clear, to create a modular, extensible platform that made the entire AI-native claims processing workflow observable, governable and composable.

Unified claims platform

A platform for managing claims from first notice of loss to payout, built with agentic claims processing in mind.

Claimant experience

Digital, multi-channel intake to collect structured data about claims.

Automated follow-ups

Inbound voice agents

Upfront quality checks

Self-service updates

Claim handler experience

A human-in-the-loop workspace to process claims in collaboration with our agents.

Call assistant

Document workspace

Suggested actions

Observability & audit

Claims agents

Reusable decision units. Each one stands alone, and a claim draws on whichever it needs.

Triage

Coverage

Recourse

Question engine

Invoice checks

Data unification

One centralised place for everything needed to power the agents.

Claim data

Calls, photos, documents, policy

Agent knowledge

Rules, procedures, exceptions

Platform integrations

Connects to existing systems and data sources.

Core claims system

Document stores

Company wikis

The platform in layers: the systems we integrate with feed one unified data layer, that data powers a composable set of claims agents, and the agents power the two experiences on top

The platform had to bring together the work of many teams, each building tools for a different area of the claims lifecycle, while the agentic technology underneath them was still being developed. What an agent could reliably do shifted month to month, and nobody had a settled answer for how that capability should appear to a claim handler. An agent that reads, drafts or decides needs a defined place for a person to step in, and in insurance that place isn’t a preference: decisions have to be explainable, auditable and attributable to someone. Every agent touchpoint was a product question and a regulatory one at once.

Instead of designing screens for an ever-shifting capability set, I developed a framework for categorising different levels of agentic automation and defined custom interaction patterns for all levels of human-in-the-loop that needed to be supported. Next, I designed the frame our tools would arrive into: a modular structure where each tool could feel native to its own users while sitting inside one navigation system, one design language and familiar interaction patterns. Teams could build into the platform without inventing their own patterns, and a claim handler moving between tools didn’t have to learn a new product each time. It let the platform scale both within existing customers and to new clients adopting our capabilities for the first time.

High-stakes, low-complexity claims

Supervised automation

AI capable, but not allowed

AI-led, human-supported

AI suggests action, human approves

High-stakes, high-complexity claims

Supported manual

AI not fully capable

Human-led, AI-supported

AI does what it can, hands off to human

Low-stakes, low-complexity claims

Full automation

AI capable and allowed

AI-run

AI executes

Not a target state

Compliance, legal, and regulatory requirements demand human oversight.

Low-stakes, high-complexity claims

Audited automation

AI imperfect, but allowed

AI-run, human-audited

AI executes, human spot checks

Complexity

The automation framework assigns a level of automation based on what a claim costs to get wrong and how hard it is to handle. Each level then has its own human-in-the-loop interaction pattern

From client delivery to a product mindset

The landscape I arrived into at QuantCo was heavily driven by client-specific engineering. Forward-deployed teams of data scientists and ML engineers embedded inside insurers, building what each client needed, then moving on. In practice however, these engagements often became long-running, with no clear handover point. Ultimately, the shift into becoming more product driven came from the need to deliver outcomes at scale in a more sustainable, less labour-intensive format. Early on, I embedded with forward-deployed teams at the client and put product framing into engineering decisions early enough to shape what was built. For every capability we scoped, I asked the same question: which parts of this belong to this client, and which belong to the product?

Then came the advocacy piece: these teams had never worked with a designer, nor on a centralised product effort. To break out of the tunnel vision of client delivery I ran vision workshops to define what a QuantCo platform could look like, hosted outreach sessions on product and design thinking, and introduced a roadmap and prioritisation framework to hold client commitments and product bets in the same view. Within 12 months of joining, the role of design had shifted from helping teams deliver better tools, to shaping how these tools came together to create QuantCo’s claims product experience.

Product roadmap and prioritisation process

A roadmap and prioritisation framework built to hold client commitments and product bets in the same view

Impact and leverage

The platform is now in use at several major insurance companies and has become central to how QuantCo communicates its value to customers, investors, and partners.

By creating clarity across tools, teams, and workflows, the platform helped QuantCo shift from fragmented prototypes to a product mindset and from powerful models to composable, visible outcomes that had a clear path to integration with complex client systems.

  1. Joined QuantCo

    Joined as first designer to shape the platform experience

  2. Design embedded in delivery

    Worked directly with customer teams to understand real workflows and constraints

  3. Initial framing

    Synthesised delivery insights into core journeys and experience architecture

  4. A shared vision

    Ran workshops across teams to define what a QuantCo platform could be

  5. Agentic exploration

    Explored agent-based workflows and generative platform capabilities

  6. First MVP in production

    The first release live with a client, not a prototype

  7. Roadmap and prioritisation

    A framework holding client commitments and product bets in the same view

  8. Patterns for AI features

    Extended the design system with patterns for integrating AI features

  9. Levels of automation

    One human-in-the-loop framework, standardised across every agentic tool

  10. The 2026 vision

    A working prototype of the platform the whole company could build into