No. Unlike large language models (LLMs) that learn from raw data, such as information appearing on the web or provided by users, AppZen’s AI models do not learn from your raw data. Instead, they are fine-tuned using aggregated data, ensuring a more secure and compliant learning process.

Aggregated data refers to data that has been de-identified and anonymized, then converted into mathematical models that can’t be traced back to their original form. This process is different from financial data aggregation, where financial reports are consolidated across multiple entities.

In our AI context, aggregated data is used solely for fine-tuning, to enhance performance in areas like expense report auditing and invoice processing. This data includes receipts, expense reports, vendor invoices, vendor emails, and purchase orders—but always in an anonymized, secure form. Users cannot access raw data, and our AI models do not enable any conversational prompts.

In Expense Audit, AppZen’s AI models learn from audit policies set by your organization. The AI flags high-risk expense reports based on potential compliance violations. Over time, the AI becomes more adept at identifying high-risk transactions, ensuring the most critical reports are brought to your attention. The AI also learns receipt formats, enabling it to quickly audit transactions from new merchants.

In Autonomous AP, the AI learns to match invoice lines to PO lines based on your company’s unique business practices. This enables touchless processing of invoices, eliminating the need for manual review. AI models also learn to apply accurate accounting codes for invoices not tied to POs, ensuring accurate accounting automation. The AI continuously learns new invoice formats to handle invoices from new vendors without disruption.

In AppZen Inbox, our AI models automatically categorize emails based on their content to automate response workflows. The AI also learns how to automatically generate vendor responses based on the content and attachments found in incoming emails. You always have the option to review and edit these replies before they’re sent.

Our AI models are trained on aggregated data across millions of anonymized expense reports, receipts, credit card transactions, invoices, vendor emails, and user actions. Before training, all data is stripped of personally identifiable information (PII) and any details that could identify individuals, businesses, or the nature of any business practices, ensuring the highest standards of privacy.

Yes. Any data processed by AppZen is converted into aggregated data. This aggregated data is then used to fine-tune our models, continually improving accuracy and efficiency for customers using our solutions.

In Expense Audit, aggregated data is used to ensure high-risk expenses are prioritized for auditor attention and only compliant expenses are streamlined. In Autonomous AP, it’s used to improve invoice entry, PO matching, and GL coding accuracy for faster rates of touchless processing. In AppZen Inbox, it’s used to enhance email categorization for faster, more accurate, touchless responses.