The new benchmark, ‘CRMArena-Pro’, creates a simulated enterprise environment with synthetic data, to evaluate API calls to systems and safeguard personally identifiable information data.

Businesses can use the framework, which Salesforce said acts ‘much like a digital twin of business’, to test the accuracy and efficiency of AI agents.

The company has also launched a benchmarking tool to evaluate AI agents operating in contexts such as customer service and marketing, using metrics including accuracy, cost, speed, trust and sustainability.

Salesforce AI Research said these developments would help give businesses “the trust and tools they need to evolve into agentic enterprises”, i.e. organisations that use digital labour and AI to work alongside people.