Salesforce CRM: Testing AI agents with hundreds of requests

Since the end of October, AI agents can be used within Salesforce CRM to communicate with customers and exchange goods or place orders without human assistance. Salesforce introduced the Agentforce platform for this purpose in September.

According to Salesforce, the AI agents can be created in the Agent Builder without programming knowledge, with little effort and in a very short time. They can be provided with instructions in natural language and fed with PDF documents on the products sold, for example. Furthermore, the same rules and limits apply to AI agents as to Salesforce CRM.

Independently acting AI agents

However, it may not be quite so easy to outsource customer service to an AI agent. This is because such agents need to be tested and configured without interfering with day-to-day business.

Artificial intelligence that makes budget-relevant decisions is entering uncharted territory and harbors risks. Its manufacturer Salesforce promises a low hallucination rate: in case of doubt, the AI should not respond at all before it produces errors. However, many customer companies are still likely to be cautious.

Sandbox for interaction with the AI

The Agent Builder contains a chat window in which individual agents can be tested, but only manually and to a limited extent. Salesforce is now offering the "Agentforce Testing Center", in which usage scenarios can be run through in a sandbox environment.

Companies will be able to run through how end customers interact with an agent in natural language. The Testing Center should be able to automatically generate hundreds of synthetic interactions and test them in parallel.

The "Salesforce Sandbox" is intended to provide an isolated environment in which the production data of a customer company is mirrored. It should then be possible to migrate changes to the process to production. (akr)