Koa unveiled but not available

Salesforce introduced Koa, its first homegrown AI model for CRM, at Investor Day on September 16. Only select customers can pilot it now, with broader U.S. availability coming later.

Post-training gives control without cost

Salesforce post-trained Koa on NVIDIA Nemotron's foundation, controlling the weights without building from scratch. This buys control over behavior and deployment while specializing for CRM workflows.

Benchmark claims need real proof

Salesforce says Koa produces three times fewer errors on CRM actions than competitors on its own benchmark. Independent validation would require accuracy in live deployments and measurable impact on paid adoption.

Winter release timing delays revenue

Enterprise customers typically deploy around the Dreamforce Winter release, so September's unveiling converts to billable consumption months later. Agentforce ARR already exceeded $1.5 billion, up over 240% year-over-year.

“AI is delivering value across every layer of our platform. We're seeing incredible demand for our AI and data products, with ARR about to cross $4 billion.”

Microsoft and ServiceNow have edges

Microsoft already ships Copilot inside its productivity suite, a distribution advantage Koa's benchmark doesn't address. ServiceNow is monetizing its own agentic tier on similar consumption logic.

Pricing power is the deciding factor

The bull case depends on whether Koa lifts Agentforce win rates through the Winter release. The bear case is that model capability alone doesn't guarantee commercial outcomes or pricing power.

“The deciding variable is whether Koa carries pricing power and whether pilot accounts convert to production before fiscal year-end.”