Case studySnapsheet

Reading the claim, so adjusters stop re-keying it.

I designed the Snapsheet AI layer that captures a claim's documents with OCR and vision, extracts the fields that matter, and turns them into suggestions an adjuster reviews in a tap. No black box.

Staff Product Designer, Claims + AI OCR + vision capture AI suggestions Human in the loop
AI suggestionverify before action
snapsheet · ai · claim summarizer flow
Snapsheet AI flow builder run, all steps complete, with execution results
01 / Context

A complete claims system, learning to think.

Snapsheet is claims management software used across 170+ insurance customers, including 16 of the top 20 US property and casualty carriers. It already unifies a claim's documents, communications, notes and vendors into one view.

Snapsheet's stated goal for AI is to take it out of the black box. I lead design on the Claims and AI teams for that layer: the suggestions, summaries and document capture that sit on top of the claim, and the builder that lets the business shape them.

Role
Staff Product Designer, Claims + AI
Company
Snapsheet
Timeline
2024–present
What I owned
The AI suggestion experience, OCR capture to claim data, the adjuster AI tray, agent and flow builder direction
Surfaces
Snapsheet AI tray, AI Actions, Agent and Flow Builder
Partners
Claims and AI product, engineering, data science
02 / The problem

Adjusters spend their day re-typing, not deciding.

Every claim arrives as a pile of documents: the first notice of loss, police reports, repair estimates, photos. The platform gathers them in one place, but a person still had to read each one and key the facts into fields by hand.

Re-keying is slow, easy to get wrong, and it buries the work that actually needs a human: judging coverage, setting a fair reserve, deciding the next step. The documents were the bottleneck, and the adjuster was the OCR.

From one document alone

A single police report can hold the parties, citations, narrative and a dozen more facts an adjuster would otherwise transcribe by hand, one field at a time.

A representative claim, before this work:

9
Documents attached to the claim
14
Fields buried in one police report
5
Exposures to reconcile across sources
100%
Keyed in by hand, by the adjuster

All the data was in the file. None of it was structured.

03 / Approach

Capture, extract, suggest. Never silently fill.

The model was simple to say and hard to earn trust for. OCR and vision read the documents. Extraction turns them into structured fields. The system proposes those fields and the next action as a suggestion. A person reviews and confirms. Nothing reaches the claim on its own.

Attachments tray with agents that read documents, each scoped to its parent
Agents that read documents

Capture, scoped to the source

Extract a police report into 14 fields. Read damage photos for severity and a repair estimate. Compare a shop estimate line by line. Each agent is scoped to its parent document, so the adjuster always knows exactly what was read and where a value came from.

AI Action configuration for Summarize Claim
One building block

Every capability is a defined action

A summary, a fraud check, a liability estimate: each is a named, versioned, configurable AI Action. Designing the action as a first class object meant the same block could power the tray, a workflow step, or a button, without rebuilding it three times.

Claim documents
FNOL, police report, estimate, photos
Capture
OCR and vision read the page
Extract fields
Turned into structured data
AI suggestion
Proposed, with its sources
Adjuster reviews
Confirm, edit, or dismiss
04 / The adjuster experience

One tray. Pick an agent, or just ask.

From inside a claim, the adjuster opens Snapsheet AI. They can run an agent, summarize the claim, draft an insured update, or ask a question in plain language. The summary and the suggestions arrive together, each carrying its sources and a single clear action.

Snapsheet AI tray, pick an agent
01 Pick an agent
Chronological summary of the claim
02 Read the summary
Suggestions that need review with sources and actions
03 Act on suggestions

The suggestions are where capture turns into closure. A coverage gap on the rental, a missing police report holding up reserve approval, a reserve sitting below the comparable median. Each one shows the policy section or claim data behind it, and offers one move: draft the heads-up, request the document, adjust the reserve. The adjuster stays in command of every decision.

05 / The system behind it

The workbench that makes a suggestion trustworthy.

A simple tray for the adjuster sits on a serious builder for the business. The suggestion is only as good as the system that produced it, so I designed that system to be inspectable, composable, and tunable.

Agent library

Specialized agents, versioned and governed

A managed library of agents, Summarize Claim, Fraud Detector, Liability Estimator, Damage Vision, each with a status, a last modified date, and a model. Treating agents as governed objects is what lets a carrier trust them in production.

AI Agents library with status and model per agent
AI AgentsEach agent is a managed, versioned object, not a one off prompt.
Visual flow builder

Agents compose into a flow on a canvas

A claim is validated and triaged, then claim data, exposures, history and documents are extracted in parallel, analyzed, and merged into one reconciled summary. Building blocks for AI processing, flow control and outputs let a non engineer assemble this without writing code.

Flow builder canvas with parallel extraction and an add step palette
Flow BuilderParallel extraction, a merge step, and a palette of composable building blocks.
Per step control

Each step picks its own model and prompt

Every step chooses its foundational model (Claude, GPT-4o, Gemini), its own token, temperature and sampling settings, and its own system prompt. Designing this control surface let the team tune accuracy task by task, and stay model agnostic as the field keeps moving.

Step configuration panel with contextual inputs, model selection, settings and a system prompt
Step configThe panel runs long; this shows the top. Click to view all of it: contextual inputs, model selection, settings, and the system prompt.
06 / Designing for trust

Take AI out of the black box.

In claims, a wrong answer delivered confidently is worse than no answer. Three principles kept the AI accountable to the person responsible for the outcome.

01

Show the work

Every suggestion cites its sources, a policy section, the FNOL, comparable claims. Every summary links back to the dates and documents behind it, so a surprising result invites a look instead of blind trust.

02

Confirm before action

Suggestions stay pending until a person acts. Every response says it plainly at the bottom: AI generated, audit logged, verify before action. The human, not the model, moves the claim.

03

Fail loudly, not silently

When a step times out or a dependency is missing, the run stops, names the failed steps, and returns partial results, rather than papering over a gap with a confident guess.

Flow run failed with two errors, showing per step status and partial results
Designed failureA failed run is a designed state: it names what broke, why, and what it could and could not complete. Screens use representative demo data.
07 / Impact

What the work made possible.

14
Fields captured from a police report in one pass, ready to review
2.1s
To read, extract and summarize a full claim file end to end
0→1
Manual re-keying replaced by reviewed AI suggestions
100%
Of AI output cited, audit logged, and confirmed by a human

This was less about adding an AI feature and more about operationalizing AI responsibly inside work adjusters already do. The design made the model legible: every read is traceable to a document, and every action is a human's to confirm.

08 / Reflection

The goal was never to replace the adjuster's judgment. It was to hand them the facts already sorted, so judgment is all that's left to do.

Capture and extraction are only worth anything if the adjuster trusts them, and trust comes from showing sources and never acting alone. The most important pixels on the screen were not the answer. They were the citation and the confirm button.

The simple tray and the complex flow builder turned out to be the same idea at two altitudes: make the AI legible to the person who is accountable for the outcome. That is the through-line I carry into every AI surface I design now.