Roots scored 99% for claims accuracy Mistral AI reached 62%
Roots scored 93% for underwriting accuracy GPT 5.0 only 80%
Roots scored 93% for underwriting accuracy Gemini 3.0 Pro got 84%
Roots Outperforms General Knowledge LLMs
98%+ Accuracy Guaranteed
Claims Extraction Benchmark Results
Roots AI models top OpenAI and Mistral for accuracy across all key insurance fields, showing a considerable advantage with a confidence threshold.
Underwriting Extraction Benchmark Results
Roots AI models outperform state-of-the-art GPT and Gemini models across key underwriting extraction fields.
GPT-4.1
% Accuracy
GPT-5.0
% Accuracy
Gemini-3.0-Pro
% Accuracy
Roots
% Accuracy
Underwriting Field Extracts
Overall Accuracy
81
80
84
Loss Run Extraction Benchmark Results
Roots AI models beat commonly used frontier, state-of-the-art models across key loss run insurance data extraction fields.
December 2025 results.
Data Extraction Category
GPT-4.1
% Accuracy
GPT-5.0
% Accuracy
Gemini-3.0-Pro
% Accuracy
Roots
% Accuracy
Data Field
Accident Description
Claim
88
53
81
Accident State
Claim
90
53
83
Allocated Expense Reserves
Claim
85
60
91
Allocated Expenses Paid
Claim
88
57
86
Carrier
Claim
75
56
74
Claim Number
Claim
83
54
76
Claim Reported Date
Claim
89
59
86
Claimant Closed Date
Claim
93
61
90
Claimant Name
Claim
88
58
84
Date of Incident
Claim
88
58
85
Indemnity Paid
Claim
66
53
81
Indemnity Reserves
Claim
80
53
84
Line of Business
Claim
91
60
86
Medical Reserves
Claim
86
60
91
Nature of Injury
Claim
53
40
72
Paid Medical
Claim
90
59
89
Policy Number
Claim
84
53
81
Policy Year
Claim
81
53
74
Recoveries
Claim
97
61
91
Status
Claim
70
46
69
Total Incurred
Claim
73
48
72
Total Paid
Claim
74
52
73
Total Reserve
Claim
69
54
83
Type
Claim
36
26
32
Data Extraction Category
GPT-4.1
% Accuracy
GPT-5.0
% Accuracy
Gemini-3.0-Pro
% Accuracy
Roots
% Accuracy
Data Field
Allocated Expense Reserves
Policy
81
89
Allocated Expenses Paid
Policy
80
87
Carrier
Policy
73
72
75
Experience Mod
Policy
88
90
90
Expiration Date
Policy
72
65
74
Inception Date
Policy
81
74
Indemnity Paid
Policy
77
88
84
Indemnity Reserves
Policy
81
88
87
Line of Business
Policy
76
81
80
Medical Reserves
Policy
86
90
88
Named Insured
Policy
90
92
91
Paid Medical
Policy
84
87
84
Policy Number
Policy
72
77
81
Policy Year
Policy
86
86
88
Recoveries
Policy
86
91
89
Total Claims
Policy
84
83
87
Total Closed Claims
Policy
80
85
68
Total Incurred
Policy
75
78
79
Total Open Claims
Policy
89
85
91
Total Paid
Policy
69
72
73
Total Reserve
Policy
64
81
76
Validation Date
Policy
78
63
Loss Run Extraction Benchmark Results
Roots AI models beat commonly used frontier, state-of-the-art models across key loss run insurance data extraction fields.
July 2025 results.
Data Extraction Category
GPT4.1
% Accuracy
Deepseek-r1
% Accuracy
Gemini-2.5-flash
% Accuracy
Roots
% Accuracy
Data Field
Indemnity Paid
Financials
65
83
74
Total Incurred
Financials
66
52
71
Recoveries
Financials
52
64
71
Paid Medical
Financials
44
74
74
Medical Reserves
Financials
44
59
64
Indemnity Reserves
Financials
63
70
74
Total Paid
Financials
67
85
78
Total Reserve
Financials
59
69
81
Allocated Expenses Paid
Financials
86
84
Allocated Expense Reserves
Financals
76
55
Claimant Name
General
80
90
Date of Incident
General
80
89
Claim Reported Date
General
83
89
82
Claim Number
General
45
43
61
Line of Business
General
76
95
95
Carrier
General
75
88
93
Policy Number
General
65
81
91
Policy Year
General
78
94
Status
General
57
88
68
Accident State
General
88
90
96
Claimant Closed Date
General
72
89
85
Accident Description
General
54
60
77
Key Features of InsurGPT™
Trained on Millions of Insurance-Specific Data Points
InsurGPT™ is fine-tuned using millions of proprietary insurance documents, parameters, and business processes, giving it a deep understanding of the language and complexities of the insurance industry.
Patented Human-in-the-Loop (HITL) Capabilities
The system incorporates real-time human oversight, allowing subject matter experts to intervene when necessary, ensuring data accuracy and process control.
Reduced False Positives & Hallucinations
Unlike generic AI models, InsurGPT™ significantly reduces errors and hallucinations, ensuring more reliable outputs.
Enterprise-Ready & Compliant
InsurGPT™ is built for enterprise-scale operations with robust security, including SOC2-Type 2, ISO27001, CCPA, GDPR, HIPAA, and NYCRR 500 compliance.
.png?width=300&name=accuracy%20(1).png){kind=icon}
Improved Accuracy
Achieve over 98% accuracy in interpreting complex insurance documents and data.
AI & Human Collaboration
With built-in Human-in-the-Loop capabilities, InsurGPT™ combines the power of AI with the oversight of industry experts, ensuring the highest levels of control and precision.
Compliance & Security Built In
Trust in InsurGPT™’s enterprise-level security with industry-standard certifications, protecting sensitive data and ensuring compliance.
Our Commitment to Responsible AI
Insurers must prioritize fairness and transparency to make unbiased, fact-based AI decisions. Roots simplifies this with an ethical AI framework based on eight core principles, ensuring explainable decisions, strong data protection, and ongoing improvement to maintain trust and compliance.
Roots’ AI Research and Development
Roots is dedicated to AI research that yields practical solutions. Our work emphasizes advanced post-training methods and reasoning capabilities, enabling AI models to simplify complex operations, make business processes more effective, automate routine tasks, and improve decision-making.