COI360
Reengineering document automation with AI for speed, accuracy, and auditability.
INTRODUCTION
About
COI360 (formerly CTRAX) is an end-to-end compliance and document processing platform handling thousands of Certificates of Insurance (COIs) each month for agencies, ports, construction firms, school districts, cities, and more. In addition to verification, it manages policy tracking, expiry alerts, and audit reporting. This case study focuses on the verification workflow how we transformed a manual, error-prone OCR process into an AI-powered, user-centric experience.
Problem
The legacy OCR (Artsyl) delivered <50% accuracy, supported only ACORD 25, and couldn’t learn from corrections. Specialists spent 800+ hours/month on copy-paste workflows, context switching, and error-prone reviews without confidence scores or prioritization leading to fatigue, low morale, and compliance risks.
Solution
We reimagined verification with AI: flexible document classification, adaptive field extraction, and continuous learning. Intuitive UX with visual overlays, confidence heat-maps, inline editing, and robust ML models cut review time, reduced errors, and scaled seamlessly across evolving COI formats.
Success Metrics
RESEARCH
Contextual Inquiry, Shadowing & Interviews
Goal
I aimed to gain a deep, empathetic understanding of specialists’ end-to-end verification workflows to surface both explicit frustrations and hidden challenges, and pinpoint critical UX pain points for targeted AI-driven solutions.
Prior Research
Evaluated five leading intelligent document processing platforms for benchmarking.
Shadowing
Spent two days observing three specialists processing live COIs in their natural environment.
Logged every interaction, click, and hesitation; captured workarounds (e.g., sticky notes for reminders, printed spreadsheets for cross-checks).
Interview
Asked specialists to demonstrate the exact steps they take when verifying a COI in COI360, probing for pain points at each click.
Questions Asked:
“Walk me through how you locate and verify the insured name across different COI layouts.”
“When you see a low-confidence field, how do you decide whether to trust or correct it?”
“Tell me about the most frustrating part of correcting OCR errors, what would make that easier?”
“Describe how you handle multi-page docs or unexpected attachments today.”
“How do you currently prioritize which COIs to review first, and why?”
Focused on uncovering workflow gaps, decision criteria, and emotional responses rather than abstract preferences.
Key Pain Points Discovered - Current System
Poor Scan Quality
Blurry, skewed scans from varied sourced equipments reduce OCR accuracy and create rework.
No Confidence Score
Without confidence indicators, ~5,000 forms are all reviewed equally.
No Triage or Insights
All COIs are processed in order, with no risk or deadline prioritization.
Cognitive Overload
Specialists juggle dozens of COI templates across tools, causing errors and fatigue.
Limited Template Support
OCR only supports ACORD 25, forcing manual extraction for all other forms.
100% Manual Review
Every document requires full review, consuming 800+ hours monthly.
Time-on-Task Study
What we did:
Instrumented the prototype and legacy system with stopwatch timings across 20 COIs each. We measured durations for extraction review, correction, and save steps.
Key findings:
Instrumented the prototype and legacy system with stopwatch timings across 20 COIs each. We measured durations for extraction review, correction, and save steps.
3-5m per COI on average, 65% of total task time spent aligning data to form.
2-4m per COI on average, with specialists spending 25s searching for paste targets.
Problem Statement
Team need a context-aware, interface that pairs visual document mapping with ML classification/extraction to auto-triage and continuously learn, leaving humans to focus on edge cases.
IDEATE
Ideation & Concept Development
Design Workshop
We ran a sticky-note brainstorm with the verification team, 50+ ideas dot-voted and clustered into Visual Mapping, Smart Triage, Batch Efficiency, and Image Enhancement, then converged on the north-star concepts: inline overlays, confidence heat-maps, batch-edit panels, and AI template detection to cover every COI layout.
Sketches & Low-Fidelity Wireframes
We began with paper Crazy 8s across core concepts (dual-pane mapping, inline overlays, confidence cues), converged the best ideas into a unified low-fi wireframe, and built a focused Verification prototype to align stakeholders on the core flow. With that validated, we moved straight to high-fidelity mocks, confident the foundation worked.
High-Level WorkFlow
DESIGN
Dual-Pane Visual Mapping
Dual‑pane overlays link each field to its exact spot on the PDF, so reviewers never lose context while verifying.
Auto‑Detect Classifier
The AI classifier tags each upload by type the moment it arrives, eliminating manual sorting and guesswork.
Confidence‑Driven Triage
Confidence badges and filters route only risky COIs to humans, letting clean files pass without extra effort.
Mobile/Email Upload
Drag‑drop, email, and QR capture meet users where they are, accelerating intake and improving image quality.
Future Scope
Vendor/Agent Portal – Secure uploads, live status tracking, and self-corrections for vendors.
Real-Time Compliance Pre-Check – Instant Pass/Warning/Blocker validation with fix prompts at upload.
Mobile Capture & Review App – Camera-assisted capture, offline support, and push-based re-submission.