Singapore – Salesforce has launched Koa, its first CRM reasoning model for Agentforce, built on NVIDIA Nemotron. According to Salesforce, the model already matches or exceeds leading model performance on CRM actions while producing three times fewer errors on the company’s CRM benchmark, which tests real-world tasks such as updating an opportunity, routing a case and scheduling a follow-up.
The model was developed through what Salesforce describes as deep technical collaboration with NVIDIA. It was built by post-training NVIDIA Nemotron 3 Super with a proprietary synthetic dataset modelled on enterprise knowledge drawn from nearly three decades of CRM deployments.
Salesforce says the resulting reasoning model is grounded in how businesses actually operate, reflecting their processes, workflows and operational policies. Rather than functioning as a general-purpose assistant, Koa is purpose-built to help Agentforce agents reason through complex, multistep workflows and select the right tools to complete them.
Crucially, no customer data was used to train the model. Salesforce says the training corpus was built entirely from synthetic scenarios designed to reflect the reasoning, tool use and decision-making skills that Agentforce agents perform across CRM workflows, including generating leads, qualifying opportunities and resolving service cases.
These scenarios were built to simulate enterprise workflows across more than 14 industries, among them manufacturing, financial services, healthcare and travel. Each scenario paired a persona with specific tasks, then mapped the sequence of actions and tool calls an agent would need to take to complete them.
To post-train the model, Salesforce applied Supervised Fine-Tuning alongside reinforcement learning using Group Relative Policy Optimization, drawing on NVIDIA NeMo RL, NeMo Gym and NeMo AutoModel. By focusing training on a prioritised set of enterprise tasks, the company says the model developed deeper task-specific expertise, learning not just to produce correct answers but to take the right sequence of actions to reach a goal.
Salesforce retains control of the model weights and says it performs both post-training and inference entirely within its own trust boundary. This, the company says, ensures no customer data crosses that boundary during inference, giving customers a specialised, Salesforce-hosted option for their Agentforce deployments.
Marc Benioff, Chair and CEO of Salesforce, framed the launch as a shift in where the company’s value lies. “The most valuable thing Salesforce has built isn’t our platform — it’s the accumulated knowledge of how enterprise business actually works.”
“With Koa, the knowledge is put inside the model itself. We trained a reasoning engine that understands the structure of a deal, the lifecycle of a service case, and the workflows that vary across industries. That’s a different kind of intelligence, and it runs entirely inside your trust boundary,” Benioff said.
NVIDIA Founder and CEO Jensen Huang linked the release to a broader shift in enterprise software. “AI is creating a much larger opportunity for software. Every company needs useful AI, tailored to its knowledge, expertise, and work. NVIDIA Nemotron open models give Salesforce the foundation to turn decades of enterprise expertise into specialised AI with Koa, creating a CRM model that can reason and securely take action.”
Beyond Agentforce, Salesforce and NVIDIA are also extending Nemotron-based models and accelerated computing into Missionforce, the companies’ platform for government and regulated organisations. That extension carries the same emphasis on control over model, data and deployment environment, with mission-specific AI being brought into private clouds, air-gapped networks and other secure environments.
Koa is already operating inside Salesforce itself, including an agent within Slack that helps employees locate information and complete everyday tasks. The model is now moving into customer pilots with 1-800Accountant, Baxter Credit Union, Engine, Formula 1, UChicago Medicine and Xero.
Ryan Teeples, Chief Strategy Officer at 1-800Accountant, pointed to the complexity the model is designed to manage. “Accounting requires navigating tax rules, financial data, documents, and the unique circumstances of every customer.”
“Koa gives our agents the reasoning to work through that complexity step by step and use the right tools along the way. That means we can extend more of our accountants’ expertise across every customer interaction and help people get to the right outcome faster,” Teeples added.
At Baxter Credit Union, John Sahagian, SVP and Chief Data Officer, described a similar emphasis on context. “Our members come to us with goals, whether that’s buying a home, managing their money, or planning for what’s next.”
“Koa can help our Digital agents understand the full complexity and context behind those goals and reason across the information, tools, and policies needed to move them forward. It gives us a powerful way to make every interaction more intelligent, personal, and useful,” Sahagian continues.
Elia Wallen, Founder and CEO of Engine, said the model’s reasoning capability was central to the company’s interest in the pilot. “At Engine, we’re building the future of business travel, and that starts with equipping our teams to deliver exceptional service.”
“Business travel has countless moving parts, and what we need from AI isn’t a model that sounds confident — it’s one that can reason precisely through complex, multi-step problems. Collaborating with Salesforce and NVIDIA on a reasoning model purpose-built for Agentforce lets us push that vision further, faster,” Wallen said.
Meanwhile, Andrew Chang, Chief Marketing Officer at UChicago Medicine, highlighted the operational, behind-the-scenes nature of the work Koa is intended to support. “Some of the most important work in healthcare happens behind the scenes, coordinating information, navigating complex processes, and making sure the right next step happens at the right time.”
“Koa can work across those longer, multi-step workflows and help our teams manage that complexity more effectively. That creates more time and capacity for what matters most, caring for patients,” Chang concluded.
