Navigator is an assistant tool within the 50skills Journeys system, designed to help users build and maintain workflows (journeys). The tool simplifies the process of designing and configuring steps, emails, forms, and integrations, enabling users to speed up their work and see their ideas turn into actual workflows in a short amount of time.
What artificial intelligence is used in the Navigator assistant tool, and how does it work technically?
The artificial intelligence powering the assistance in 50skills Navigator uses the Claude language model. To ensure security and data segregation, the Claude model has no direct connection to 50skills' databases.
Instead, the system utilizes a tool-calling (or function-calling) approach. This means the AI assistant can only perform actions through highly restricted, pre-defined "tools" that we have designed and fully control.
What data does the AI have access to? Can it be turned off?
The Navigator functionality in Journeys is strictly intended as an aid for designing and configuring workflows (e.g., setting up emails, forms, or integrations).
The "tools" the AI has access to are strictly limited to one area: the journey builder.
Important disclaimer regarding data entry: Because the AI assistant works solely with configuration data, it never accesses personal or sensitive data unless the user specifically inputs such data themselves within the journey builder. For example, if a user manually types a real person's name or social security number into a prompt or hardcodes it into an action, that data will be sent to the AI. We therefore strongly advise users to always use variables instead of real data when designing workflows.
If a customer prefers, the Navigator functionality can be completely disabled company-wide within the system.
Where is the data processed and stored?
The AI functionality is integrated into our solution, and data related to 50skills is stored on our secure infrastructure. When the AI is used during setup, only the exact information strictly necessary to perform the specific requested action is transmitted.
Data hosting and routing: Currently, requests to the AI are routed through servers in the United States (US routing). However, we are actively working on implementing a European routing path (EU routing) to ensure all communications with the AI occur within the European Economic Area in the very near future.
Is customer data used to train AI models?
No, without exception. Data from our customers is never used to train our models, nor is it sent for training to Claude/Anthropic or any other third party. Prompts are used exclusively in real-time to generate a response or an action within the journey builder.
Do 50skills employees have access to AI data or logs?
50skills employees' access to any operational data or system logs is strictly limited. It is based on strict access controls and is only granted temporarily if absolutely necessary for troubleshooting or customer support, in accordance with our security protocols.
Detailed Technical Specifications: Architecture, Access Controls, and Privacy
To ensure the highest level of security when using AI within the solution, the design is based on strict data segregation (Privacy by Design). Here is a detailed breakdown of these factors:
Scope of AI functionality and architecture (Builder vs. Runtime): There is a complete technical separation between the setup phase (Builder) and execution (Runtime). The AI tools can only run within the Builder environment to construct the framework of the workflows. When the workflow is later executed (Runtime) and handles real cases containing personal or operational data, the AI is completely uninvolved.
What data can the AI tools read and write? The AI tools can only read and write configuration data (metadata) for the specific workflow being worked on (e.g., names of steps, form structures, or email template drafts). The tools are intentionally designed so they have no built-in permissions or API calls to fetch real data, user data, or past records from the system's databases.
How access controls are managed: The AI follows the same strict access controls (RBAC - Role-Based Access Control) as the system itself. It never has more privileges than the currently logged-in user. If a user does not have permission to edit a specific workflow, they cannot use the AI to do so either.
Privacy (GDPR) and data security: Because the AI operates exclusively in the setup environment (Builder) and has no access to execution data (Runtime), no Personally Identifiable Information (PII) or sensitive operational data comes into contact with the language model during normal operations. All communications with the language model's API interface are encrypted in transit (TLS). As previously stated, the only exception is if a user deliberately inputs personal information into the workflow setup themselves (e.g., hardcodes names). The system is designed to work with variables to prevent this.