AI & Automated Systems Disclosure

Effective Date: January 1, 2026 Version: 1.2 Last Updated: March 1, 2026

Estimatics is committed to transparency regarding the artificial intelligence and automated systems embedded in the Platform. This disclosure describes which AI technologies are used, for what purposes, what their limitations are, and how outputs should be interpreted by professionals.

---

1. Purpose of This Disclosure

Estimatics serves professionals in the U.S. property insurance and restoration industry — an environment where documentation accuracy, defensibility, and professional accountability are paramount. AI-assisted features accelerate workflows, but they do not replace professional judgment. This document ensures users have a clear and accurate understanding of how AI operates within the Platform.

---

2. AI Systems and Providers

2.1 OpenAI

Estimatics uses models provided by OpenAI for:

• Photo damage detection — identifying and classifying damage types in field photographs
• Photo folder classification — organizing media by area and damage category
• AI Walkthrough findings — extracting structured observations from guided video inspections
• Document extraction — parsing policy and permit documents for relevant data
• Compliance checking — evaluating scope items against structured rule sets

Models used: GPT-5.4 (vision and analysis tasks), GPT-5.4-mini (classification and volume tasks).

2.2 Anthropic

Estimatics uses models provided by Anthropic for:

• Scope generation — converting damage findings into structured line-item scopes
• Report narratives — generating professional damage report text
• Inspection narratives — summarizing inspection findings in human-readable form
• Causation analysis — reasoning about damage causation from evidence
• Code Trigger Engine — evaluating scope items against building code requirements (CERTIFY plan)
• In-app AI assistant — answering user questions about the Platform

Models used: Claude Opus 4.6 (narrative and reasoning tasks), Claude Sonnet 4.6 (scope and code tasks), Claude Haiku 4.5 (fast conversational tasks).

2.3 Provider Selection Architecture

Estimatics operates a proprietary AI Provider Router that selects the optimal model for each task based on accuracy requirements, cost efficiency, and capability. Model assignments are updated as new frontier models are released. This document reflects the current state as of the last updated date above.

---

3. AI Features by Platform Tier

---

4. How AI Processes Your Data

4.1 Image Analysis. When you upload photographs to a job, the Platform may transmit those images to OpenAI's API for damage detection and classification. Images are processed transiently — they are not retained by OpenAI for model training under our enterprise data processing agreements.

4.2 Text and Scope Generation. Structured data from your inspection (damage findings, property dimensions, area tags, material types) is transmitted to Anthropic's API to generate scope line items and narrative text.

4.3 No Training on Your Data. Under our agreements with OpenAI and Anthropic, your Content (photographs, property data, scope items) is not used to train or fine-tune any third-party AI models.

4.4 Data Residency. API calls to AI providers are processed on infrastructure in the United States.

---

5. Accuracy and Reliability

5.1 Known Limitations

AI damage detection may produce false positives (finding damage where none exists) or false negatives (missing damage that is present). Accuracy is affected by:

• Photo quality, lighting, and angle
• Damage type and severity
• Similarity to training data distributions
• Obscured or partially visible damage

5.2 Geometry Accuracy

These figures are general guidance based on internal testing. Actual accuracy varies by property complexity, coverage quality, and regional satellite imagery availability.

5.3 Scope Generation Limitations

AI-generated scopes are based on visual evidence submitted. They may not capture:

• Concealed damage not visible in photographs
• Code-required items not triggered by visible damage (CERTIFY plan mitigates this with the Code Trigger Engine)
• Regional material or labor variations not reflected in current pricing data

---

6. Human Review Requirement

All AI-generated outputs require professional review before use in any official claim submission, report, appraisal, or legal proceeding.

The Platform is designed to assist licensed professionals — not to replace their judgment. AI outputs are starting points. The professional using the Platform bears responsibility for the accuracy and completeness of any work product derived from AI features.

---

7. Automated Decision-Making

Estimatics does not use AI to make fully automated decisions that significantly affect user rights without opportunity for human review. All AI outputs in the Platform are presented as recommendations or drafts that require explicit user review and action before being finalized or submitted.

---

8. CERTIFY Plan: Certified Evidence and AI

On the CERTIFY plan, the Certified Evidence feature applies a cryptographic hash chain to your evidence package — photographs, videos, metadata, GPS coordinates, timestamps, and AI findings. This certifies:

• The state of the evidence at the time of capture
• That the evidence has not been modified since certification

The Certified Evidence feature certifies the evidence, not the AI's interpretation of that evidence. AI findings are included in the certified package as a record of what the AI observed, but the underlying photographic evidence is the primary certified artifact.

---

9. Transparency Commitment

Estimatics will update this disclosure when:

• We add new AI capabilities or use cases
• We change AI providers or model tiers
• We change our data processing agreements with AI providers
• Material limitations or accuracy changes are identified

---

10. Contact

Questions about our AI systems:

Estimatics, Inc. legal@aiestimatics.com aiestimatics.com/legal/ai-disclosure