We believe in complete transparency about how our AI document analysis platform works. This report details our AI model, training data, accuracy metrics, limitations, privacy protections, and commitment to responsible AI usage. We want every user to understand exactly what our AI can and cannot do.
DetectHiddenFees operates an AI-powered document analysis platform that helps consumers identify hidden fees, pricing risks, and negotiation opportunities in contracts, invoices, estimates, bills, receipts, and service agreements. Our AI is designed as a screening tool — it highlights potential issues for human review, not as a replacement for professional legal or financial advice. This transparency report documents our AI practices in accordance with emerging AI ethics standards and our commitment to consumer protection.
Related:AI Analysis Methodology | Security Overview | Our Evaluation Process
Our AI analyzes uploaded documents to: (1) Identify pricing structures and fee line items, (2) Flag vague or suspicious fee descriptions, (3) Compare charges against industry benchmarks, (4) Highlight contract clauses that may impose additional costs, (5) Generate risk scores across multiple dimensions, (6) Provide specific questions to ask and items to challenge during negotiation.
Our AI does NOT: (1) Provide legal advice or legal opinions, (2) Make binding determinations about contract enforceability, (3) Guarantee that all hidden fees are identified, (4) Store or retain uploaded documents beyond the analysis window, (5) Share document data with third parties, (6) Use uploaded documents for AI training purposes.
Our document analysis platform uses a proprietary ensemble of natural language processing models specifically fine-tuned for financial document analysis. The system combines transformer-based language models for understanding document context with specialized classification models trained specifically on fee patterns, pricing language, and deceptive billing practices. The models are hosted on secure infrastructure and are not accessible to external parties.
We do not use general-purpose AI models like ChatGPT or GPT-4 for document analysis. Our models are purpose-built and specifically trained for the narrow domain of hidden fee detection, which provides more accurate and consistent results for this specific use case.
The AI models are trained on a curated dataset of over 100,000 documents including: (1) Anonymized contracts and agreements across industries, (2) Invoices and billing statements, (3) Estimates and quotes from home service providers, (4) Medical billing documents, (5) Financial services agreements and disclosures, (6) Automotive purchase and financing documents, and (7) Subscription service terms and conditions. Documents are labeled by fee type, risk level, and industry category by trained analysts. Training data is continuously updated to reflect new fee patterns and pricing practices.
Our AI achieves approximately 87% precision and 92% recall for identifying known hidden fee patterns in structured documents. Accuracy varies by document type and complexity. Simple invoices and bills achieve the highest accuracy (90%+), while complex multi-party contracts and medical billing documents have lower accuracy (78-85%). The system is tested quarterly against a held-out validation set of 5,000 documents. Results are published internally and used to refine model performance.
Every AI system has limitations. Our platform's key limitations include: (1) Inability to understand highly specialized industry jargon, (2) Reduced accuracy for handwritten documents or poor-quality scans, (3) Lower reliability for novel or unprecedented fee structures not represented in training data, (4) Potential false positives for legitimate charges that resemble hidden fee patterns, (5) Inability to verify whether fees were verbally disclosed outside the document. Users should treat AI analysis as a starting point for investigation, not a definitive determination.
We are committed to ensuring our AI does not produce biased outcomes. Our training data intentionally includes documents from diverse geographic regions, income levels, and service providers. We regularly audit model outputs for demographic bias and have found no evidence of systematic bias in our analysis. However, because hidden fee patterns vary by region and industry, the AI may be more accurate for fee types common in the training data. We continue to expand our training data to improve coverage.
AI analysis in our platform is supplemented by human oversight. Trained analysts review a random sample of AI analyses weekly to verify quality and identify areas for improvement. Additionally, user feedback on analysis accuracy is collected and used to refine model performance. For users who require additional human review, we recommend consulting with a qualified professional.
Privacy is foundational to our AI operations. All documents are encrypted in transit (TLS 1.3) and at rest (AES-256). Documents are automatically deleted after analysis completion. No document content is used for AI training. No document data is shared with third parties. Our systems are designed to prevent unauthorized access and comply with applicable data protection regulations. See our Privacy & AI Security page for detailed information.
We encourage responsible use of our AI platform. The tool is designed to empower consumers, not to replace professional advice. We recommend: (1) Using AI analysis as a screening tool before major financial decisions, (2) Consulting with qualified professionals for complex contracts or significant financial commitments, (3) Not relying solely on AI analysis for legal interpretations, (4) Providing feedback on analysis accuracy to help us improve, (5) Understanding that no AI system can guarantee complete fee detection.
Our AI models are continuously improved through: (1) Quarterly retraining on updated datasets, (2) User feedback integration, (3) New fee pattern detection and model updates, (4) Regular accuracy benchmarking, (5) External security and privacy audits. We publish significant updates to this transparency report annually or when major model changes occur.
Q: Is the AI analysis considered legal advice?
A: No. AI analysis is not legal advice and should not replace professional legal review for important decisions. It is a screening tool that identifies potential issues.
Q: How often is the AI model updated?
A: The model is updated quarterly with new training data. Significant updates are published in this transparency report.
Q: Can I trust the AI analysis?
A: The AI is highly accurate for identifying known fee patterns (87% precision), but it is not perfect. Always review AI findings critically and consult professionals for important decisions.
Q: What happens to my document after analysis?
A: Documents are automatically deleted after analysis is complete. No documents are used for AI training. See our Data Handling Policy for details.
Q: How is this different from using ChatGPT for document analysis?
A: Our AI is purpose-built for hidden fee detection, trained on 100,000+ financial documents. General-purpose AI like ChatGPT is not optimized for this specific task. Learn more about the differences.
Upload any document for free AI-powered hidden fee detection.
Analyze My Documents →