Dearborn Labs
Case Study · Claims AI
TerranceBot — Agentic Claims Copilot

TerranceBot gives every adjuster half a day back, every week.

Clearcover Insurance Company deployed TerranceBot, affectionately known internally as Terry, an agentic AI Claims copilot, across their claims workflows. Each active adjuster gets roughly half a workday back, every week. Time that used to go into pulling documents, cross-referencing systems, and composing research summaries Terry now handles in 34 seconds. Across the 25-person front line staff, that adds up to 617 hours returned every month: a nearly 20% productivity lift, without a single new hire.

// Headline Numbers

Productivity Lift
~20%
Equivalent to 4.7 FTE of research and documentation work returned to the team, without adding headcount.
Hours / Mo Returned
617
Monthly hours returned to adjusters. Roughly half a workday back per active user, every week.
Adoption
92%+
Not a pilot. Not an experiment. This is how the claims team works now. Clearcover internal analytics, 2025.
Daily Invocations
~500
AI research requests handled per day. The tool is in the workflow, not sitting on a shelf. Clearcover internal analytics, 2025.
Response Time
34s
Median time to research a claim and deliver a structured answer. Faster than the manual cross-system research that adjuster had previously done.
Built With Clearcover Insurance Company

The people at Dearborn Labs built Terry with Clearcover Insurance Company, our affiliate within the Clearcover, Inc. family. Every number in this document reflects production operations at Clearcover Insurance Company. DL carrier partners don't receive a copy of Terry. They get their own agentic claims copilot, built on the same platform and primitives, tailored to their workflows, their systems, and their organizational context.

// The Challenge

Processing an auto insurance claim is research-intensive. For each new claim, an adjuster must verify policy coverage, review the reported incident, read claim documents (police reports, medical records, repair estimates), assess liability, set reserves, and compose detailed internal documentation, typically spread across a dozen separate systems.

A single first notice of loss review can take over an hour. Supervisors spend significant time on follow-up reviews and quality checks rather than coaching.

// The Build

Rather than replacing adjusters, Clearcover built an AI Claims CoPilot that lives inside Slack, the tool their team already works in all day. An adjuster can speak directly with Terry, or invoke him on shared Slack channels for team collaborations on deep dives and investigations. Anything from a routine coverage check to a complex special investigations unit referral, it researches the claim across every connected system and returns a structured answer for the adjuster's review. From there, a rep can then ask questions of Terry who will use the context learned about the claim to surface answers immediately, instead of having to sift through thousands of pages of documents, and hundreds of notes and customer correspondence.

Terry routinely assembles research contexts exceeding 55,000 tokens as it reads documents, photos, searches claim history, and cross-references financial records. Any time claim complexity exceeds an LLM's context window, Terry will intelligently fall back on a recursive summarization routine, so that he reads all relevant context before building his response. Median response time: 34 seconds.

// Measured Impact — Top Claims Workflows

Median handling time compared between AI-assisted and manual claims, controlling for claim complexity. Excludes simple claim types (glass, roadside, hail-only).

Claims WorkflowVolumeWithout AIWith AIReductionHrs / Mo
First Notice of Loss — Auto Damage>14,00073 min57 min−21%192
Supervisor Follow-Up Review>1,000286 min239 min−16%65
Supervisor New Claim Review>5,000115 min100 min−12%64
Claim Closure Review>3,00043 min24 min−43%50
Customer Contact Follow-Up>1,000186 min155 min−16%30
The Signal

First Notice of Loss review — the first and most time-consuming step in every claim — was completed over 14,000 times in 2025. AI assistance cut the median review time from 73 to 57 minutes. That's 192 hours returned to adjusters in a single month, on a single workflow. Supervisors saw meaningful savings too, freeing time for coaching and quality work instead of follow-up research.

Kyle Nakatsuji · CEO, Dearborn Labs

“The goal was to stop asking trained adjusters to spend their days on document retrieval. Reserve-setting, coverage interpretation, customer communication: those decisions require judgment that only comes from experience. Terry handles the research. The adjusters own the decisions.”

// What the Agent Can Do

Terry is organized into six capability modules, supported by 17+ underlying tools, that together give Terry deep access to every aspect of a claim — from the underlying insurance policy to active litigation.

Claim Intelligence
Retrieve involved parties and their roles, view all coverage lines on the claim, browse and read file attachments — including reading hundreds of PDF pages in seconds.
Policy Verification
Pull the full policy as of the date of loss — coverage limits, listed vehicles and drivers, garaging address. Review the complete history of endorsements, cancellations, and coverage changes.
Compile Financials
Surface reserves, payments, pending disbursements, and recovery amounts across every coverage on the claim — for adjuster review.
Intelligent Note Search
Search the full claim file history using semantic similarity, keyword matching, or date-range filters — powered by a real-time vector database that indexes notes as they're written.
Settlement & Demand Tracking
Surface settlement offers, demands, and key communications across every coverage on the claim — for adjuster review.
Company Knowledge Base
Search internal procedures, guidelines, and best practices via a RAG-powered knowledge base. The agent cites the relevant company procedures and guidelines, with sources.

// Real-Time Data Architecture

The agent's understanding of every claim evolves in real time. It keeps pace with the adjusters themselves. A change-data-capture pipeline streams events from the claims management system through Kafka, where downstream processes index data into a vector database, transcribe audio recordings, and extract structured information from incoming claim notes.

// Capture
CDC streams every claim event (new notes, payments, status changes, incoming documents) into Kafka in real time.
// Enrich
Downstream processors transcribe audio, extract structured data from notes, and generate vector embeddings for semantic search.
// Index
Enriched data is indexed into a vector database, enabling semantic, keyword, and date-filtered search across every claim file.
// Reason
The agentic AI selects from 17+ tools, builds context averaging 55K tokens, and responds in a median of 34 seconds.

// How Adjusters Utilize Terry

Review this claim and complete the FNOL template
Summarize all prior claims for this vehicle
Pull the policy in force on the date of loss and summarize coverage
Calculate rental day eligibility under the policy
Find prior notes referencing the same accident location
Summarize facts relevant to a liability evaluation for adjuster review
Was the driver listed on the policy at date of loss?
Prepare the subrogation referral with carrier details

// Methodology

We identified the 20 most frequent claims workflows completed in 2025, excluding simple claim types (glass replacement, roadside assistance, hail-only). For each workflow, we compared the median handling time between AI-assisted and manual claims of the same claim complexity, controlling for inherent difficulty differences between, say, a multi-vehicle highway collision and a simple parking lot incident.

We computed a complexity-weighted average across all claim types and scaled the per-workflow time savings to a monthly estimate using observed AI-assisted claim volume. Workflows where AI showed no improvement were included in the analysis but excluded from the headline figure. The headline 617 hours/month figure aggregates time savings across all measured workflows; the table above displays the top five workflows by hours returned.

// Transparency matters. We measured all 20 workflows honestly.

// What's Next

Terry currently handles 16 of Clearcover's highest-volume claims workflows. The team is extending coverage to the remaining workflows identified in the 2025 analysis, with particular focus on litigation docket tracking and subrogation management, two areas where research burden remains high and AI assistance has the most room to run.

The architecture behind Terry (the CDC pipeline, vector search layer, and claims knowledge structure) is also the foundation of DL's carrier partner deployments. Carrier partners don't get a prototype. They get the same system Clearcover Insurance Company's adjusters use every day, tested at operational scale, with the results in this document, adapted to their guidelines and embedded alongside their team.

// Interested in deploying this?
Dearborn Labs · Applied AI for Insurance
Terrance · Live in Production

Clearcover Insurance Company is the insurance carrier referenced in this material. Dearborn Labs is an affiliate within the Clearcover Insurance Holdings family. Dearborn Labs is not an insurance company.
Results reflect Clearcover Insurance Company's deployment. Outcomes for other carriers will vary based on claims volume, mix, system integrations, and deployment scope.
Dearborn Labs does not issue, underwrite, or sell insurance, and nothing in this material is an offer of insurance or a representation of coverage.
Copyright 2026 Clearcover, Inc. All Rights Reserved.
Tracking reference: Web_CaseStudy_National_0426_TER  |  Clearcover Marketing log entry: [TO BE ASSIGNED]
Last revised: [PUBLICATION DATE PENDING]