- Information appears incorrect compared to your analyses or our documentation
- You have questions about how Pagos AI generated a specific result
- You’d like to disable access to specific data sets
Technical Architecture
What AI model and Technology Does Pagos AI Use?
Pagos AI is powered by Anthropic’s Claude Sonnet model (claude-sonnet-4-20250514). The system uses the Model Context Protocol (MCP) to securely interface with customer data through Pagos’s data analytics layer. This is a significant architectural improvement over Legacy Pagos AI.What Data is Pagos AI Trained On?
Pagos AI uses:- Claude’s base training (conducted by Anthropic)
- System prompts that define behavior, constraints, and capabilities
- Page-specific context about the Pagos platform features and metrics
- Real-time access to the customer’s anonymized payments data via MCP queries
The AI does not retain or learn from individual customer queries. Each interaction is stateless from a training perspective.
How Does Pagos AI Access My Payments Data?
Pagos AI uses a structured query interface that provides sandboxed, read-only access to your aggregated payments data. The system:- Can only access metrics and dimensions exposed through the Pagos platform UI
- Cannot access individual transaction records or cardholder data
- Uses page-level context to understand what data is relevant to your current view
- Queries data using the same filters, date ranges, and segments visible in your UI
Data Privacy and Security
Does Pagos AI Have Access to Personally Identifiable Information (PII)?
No. Pagos AI only accesses aggregated, anonymized payments data. Individual cardholder information is already removed from the data pipeline before any AI interaction occurs. The system works exclusively with:- Aggregated metrics (counts, rates, volumes)
- Dimensional breakdowns (by processor, card brand, country, etc.)
- Time-series data
What Data is Shared with the AI Vendor (Anthropic)?
Anthropic receives:
- Your natural language queries to Pagos AI- System prompts and page context- Aggregated query results (no PII)- Conversation history within a session
Anthropic does NOT receive:
- Individual transaction records- Cardholder data- Raw database access- Your organization’s full dataset
Anthropic does not use PagosAI data to train their models.
Under GDPR definitions, Anthropic is not a sub-processor because no cardholder personal data is shared with them.
Can the AI Identify My Organization?
By default, no. The data provided to Pagos AI is anonymized and does not directly identify your organization, except in two scenarios:- You explicitly mention your organization name in your query
- You query merchant account data where account names contain human-readable organization identifiers
What is Your Sub-Processor List for AI Aervices?
AWS remains the primary processor for all Pagos data. Anthropic provides the AI model but does not process cardholder data or function as a sub-processor under GDPR definitions. No cardholder PII is shared with any AI vendor.Accuracy and Reliability
How Does Pagos AI Handle Hallucinations or Errors?
Pagos AI is designed with the following safeguards:- Constrained scope - System prompts strictly limit responses to payments topics and available data sources
- Structured queries - Data requests use templated MCP tools with defined parameters, reducing ambiguity
- Transparency - The AI explains which metrics, dimensions, and filters it used for analysis
- Honest limits - When data is unavailable, the AI explicitly states it cannot answer
Can I See How Pagos AI Generates Results?
Yes. Pagos AI provides explanations with each response, describing which metrics, dimensions, filters, and time periods it queried, and whether the information came from your data, Pagos’ documentation, or general payments knowledge. When it uses synonyms or ambiguous terms, Pagos AI states its interpretation. Any significant assumptions are explicitly stated You can also ask follow-up questions like “How did you calculate this?” or “What filters did you use?” to understand the methodology.How is Pagos AI Tested for Accuracy?
Pagos performs regular testing including:- Validation against known data outcomes
- Testing with ambiguous queries to ensure proper clarification requests
- Regular automated testing to ensure the AI does not drift over time
- Monitoring for off-topic responses
- Verification that safety guardrails prevent unauthorized data access
Control and Oversight
Can Humans Stay in Control of AI Outputs?
Yes, Pagos AI serves as an analytical assistant, not an autonomous agent. It’s designed with the following human oversight principles:- Read-only access - The AI cannot modify your data, configurations, or application state
- Recommendation-only - The AI provides insights and suggestions but takes no automated actions
- User-initiated - All queries are explicitly triggered by users
- Transparent - Users can validate AI analysis through direct platform exploration
- Stoppable - Users can end any conversation and ignore recommendations
What Can Pagos AI Not Do?
Pagos AI has explicit limitations enforced through system prompts and technical architecture. It cannot perform any of the following actions:- Access data outside your Pagos account, external APIs, or the Pagos documentation
- Make competitive comparisons with other companies
- Recommend specific processors, PSPs, or vendors by name
- Export or provide raw transaction-level data (users should use Transaction Level Data Downloads add-on or the Events page)
- Send emails, create alerts, or take automated actions
Conversation Context and Memory
Does Pagos AI Remember Previous Conversations?
Pagos AI maintains conversation context within a single session (multi-turn conversation support), including:- Previously applied filters, dimensions, and time periods
- Referenced metrics and data points
- User-specified preferences within the conversation