This guide outlines when and how the Salesforce APIs are used to help estimate performance impact.

API usage scenarios

1. Model retraining

Purpose

Periodically retraining customer models using historical case data.

API Used

Salesforce Bulk Query v2 API

• Asynchronous process with results available after approximately 1 minute
• One bulk query job per week of historical data
• Typically processes 6-12 months of data (26-52 bulk query jobs total)
• Up to 4 concurrent jobs

Frequency

• Higher during onboarding (multiple times weekly)
• Lower during full deployment (every few weeks)
• Can be adjusted based on customer needs

2. Case processing

Purpose

Real-time processing of individual cases.

API used

Standard REST API

• 1 GET request to fetch a case
• 1 PATCH request to update multiple fields (e.g., topic, language, sentiment)

Frequency:

• Example case lifecycle (illustrative):
  • New case event → GET request + PATCH request
  • Case closed event → GET request
• The actual number of API calls will vary based on your specific implementation and workflow
• Additional GET requests may occur if you configure update events between case creation and closure

3. Event monitoring

Purpose

Real-time event notification.

API used

Salesforce PubSub API

• Consumes platform events or Change Data Capture (CDC) events
• Each event consumed exactly once
• Implemented with guardrails to minimize consumption

Performance considerations

• Bulk operations run asynchronously with minimal impact on Salesforce performance
• Case processing calls are made only on specific events (new, update, closed)
• PubSub API is designed for efficient real-time communication

This information should help estimate the performance impact of implementing Tekst's intelligent capabilities within Salesforce.