Customer Agent Launch

A scalable knowledge architecture for the HubSpot Customer Agent using structured JSON data and a published URL instead of manually maintained static files. The solution was designed to improve answer reliability, reduce maintenance effort, and create a stronger technical foundation for AI-powered customer service.

The project focused on optimizing the knowledge source behind a HubSpot Customer Agent.

For AI-based customer service, the quality of the answers depends heavily on how accurate, structured, and up to date the underlying information is. Different knowledge-source options were therefore evaluated, including manually uploaded JSON files, website content, predefined Q&As, and publicly accessible JSON URLs.

The goal was to create a technically robust setup that could handle frequently changing structured data while minimizing manual maintenance and keeping the Customer Agent's knowledge as close as possible to the source system.

The Customer Agent needed to provide reliable answers based on information that changes regularly, such as products, services, locations, or availability.

Using manually uploaded files created a major limitation: every data change required a new export and upload into HubSpot. This increased maintenance effort and created the risk that the AI agent would respond based on outdated information.

The challenge was therefore to find a knowledge-source architecture that was both structured enough for reliable AI interpretation and scalable enough for continuously changing data.

The initial setup relied on a manually uploaded JSON file.

While JSON provided a clean and machine-readable data structure and worked well for validating the concept, the file itself only represented a static snapshot of the source data.

There was no automated update path between changes in the underlying data and the HubSpot Customer Agent. As the amount of information and update frequency increased, this approach would have created unnecessary operational effort and a higher risk of inconsistent customer responses.

The knowledge architecture was redesigned around a publicly accessible JSON URL.

This maintained the advantages of structured JSON data while creating a more scalable update process. Instead of repeatedly exporting and uploading new files, changes can be reflected directly in the published data source, which can then be recrawled by HubSpot.

The solution included:

• evaluation of HubSpot Customer Agent knowledge-source options
• comparison of static JSON files and dynamically published JSON sources
• definition of a consistent and AI-readable JSON structure
• use of a publicly accessible URL as the central knowledge source
• reduction of manual data-maintenance processes
• consideration of additional sources such as website content and predefined Q&As
• ongoing review of Customer Agent conversations to identify missing information, incorrect interpretations, and opportunities to improve answer quality

The result is not only a technical integration, but a scalable knowledge-management approach designed around AI reliability, maintainability, and the actual customer experience.