Koa, Salesforce’s first CRM reasoning model for Agentforce based on NVIDIA Nemotron, was unveiled today by Salesforce and NVIDIA. Koa was created in close technical cooperation with NVIDIA with the goal of assisting agents in using the appropriate tools to complete tasks and reasoning through intricate, multi-step workflows.
Using a proprietary synthetic dataset based on enterprise experience from almost thirty years of CRM implementations, NVIDIA Nemotron 3 Super was post-trained to create Koa. The end product is a reasoning model based on corporate operations, including workflows, operational policies, and processes.
Koa already equals or surpasses leading model performance on CRM actions with three times fewer errors in Salesforce’s CRM benchmark, a model benchmark that comprises a variety of real-world operations, including upgrading an opportunity, routing a case, or scheduling a follow-up. Salesforce provides clients with a specialized, Salesforce-hosted solution to power their Agentforce use cases by managing the model weights and carrying out post-training and inference completely within its own trust boundary.
In order to give government and regulated organizations the same control over models, data, and deployment environments, Salesforce and NVIDIA are also integrating Nemotron-based models and accelerated computing into Missionforce. The businesses are working together to deliver mission-specific AI to secure environments such as air-gapped networks and private clouds.
Why Koa Matters for the Future of CRM
The Koa reasoning model was trained without using any customer data. The entirety of its training corpus was constructed from artificial scenarios that demonstrate the ability to reason, use tools, and make decisions. Agentforce agents generate leads, qualify opportunities, and resolve service incidents throughout the customer lifecycle in a variety of CRM workflows.
These scenarios were designed to mimic actual company workflows in over 14 industries, such as manufacturing, financial services, healthcare, and travel, rather than depending on generic material. In each scenario, a persona was assigned particular tasks, and the order of actions and tool calls an agent needed to do those tasks was sketched out.
Salesforce used Supervised Fine-Tuning (SFT) and reinforcement learning with Group Relative Policy Optimization (GRPO) with NVIDIA NeMo RL, NeMo Gym, and NeMo AutoModel to post-train the model. Through training on a specific set of priority corporate tasks, the model gained deeper expertise and learned how to take the correct action, step by step, to achieve a goal in addition to producing the correct answer.
Salesforce was able to include CRM intelligence into the model itself because to NVIDIA Nemotron open models. Salesforce ensures that no customer data crosses the trust boundary during inference by managing the weights and operating Koa within its own infrastructure.
Koa is now being used in client pilots with 1-800Accountant, Baxter Credit Union (BCU), Engine, Formula 1, UChicago Medicine, and Xero. It is already in use within Salesforce, including an agent in Slack that assists staff members in finding information and doing daily chores in Slack.
Salesforce Perspective
“Our platform isn’t the most valuable thing Salesforce has created; rather, it’s the collected understanding of how an enterprise company truly functions. Koa incorporates the knowledge into the model itself. We developed a reasoning engine that can comprehend transaction structures, service case lifecycles, and industry-specific procedures. It operates completely within the boundaries of your trust and is a distinct type of intellect. — Marc Benioff, CEO and Chair of Salesforce
NVIDIA Point of View
AI is giving software a lot more opportunities. Every business requires practical AI that is suited to its work, knowledge, and skills. Salesforce can use NVIDIA Nemotron open models to transform decades of company knowledge into customized AI with Koa, resulting in a CRM model that is capable of reasoning and acting securely. ― Jensen Huang, NVIDIA’s founder and CEO
Conclusion
And that’s a wrap on this Salesforce update. We hope this guide gave you a clear understanding of what’s new, how it works, and what it could mean for Salesforce users and businesses.
As Salesforce continues to evolve with new innovations, staying informed can help you make better decisions and get more value from your CRM. Keep following CloudMetic for the latest Salesforce updates, product announcements, insights, and industry news.
Stay tuned for more Salesforce updates and insights from CloudMetic.
Note:We are a top-rated Salesforce consultants and Salesforce service provider with a 5-star rating on Appexchange. Have questions or need assistance? Connect with us: Sales@cloudmetic.com.
