Salesforce's Koa model reaches pilots with two accounts of what it was trained on
Salesforce and Nvidia post-trained Koa on simulated CRM workflows and claim three times fewer errors than leading general-purpose models. The published record describes the training data two different ways.
The Product Desk
What happened
- Salesforce said at Dreamforce that it worked with Nvidia to build Koa, its first reasoning model, post-trained on Nvidia's open-weight Nemotron for sales, marketing and customer-support tasks.
- Kumar said the training set was synthetic, simulating enterprise workflows across 14 industries including healthcare, finance and manufacturing, so no customer data was used at all.
- Salesforce plans to roll out Nvidia's Nemotron models inside Missionforce Operations in October for customers running air-gapped networks and private clouds.
Why it matters
- What doesn't add up contradiction The two published accounts of the training data do not line up, so a data protection review cannot sign off on "no customer data" until Salesforce says in writing what the synthetic generator was conditioned on.
- Who has to decide decision Agentforce owners now decide per task whether reasoning stays inside Salesforce or leaves for Claude, and the ClaudeForce deal keeps the frontier path funded whichever way they route it.
- What it limits constraint Anyone outside the pilot has to take the error claim on the vendor's terms until winter, because the published figure is a ratio against an unnamed comparison set.
- What's now possible capability Government and other regulated buyers can put Nemotron reasoning next to their most sensitive data inside air-gapped networks from October, before Koa itself is generally available.
Next story · Invest · Sep 28
Nvidia's Huang defends AI distillation as competition despite Treasury's July sanctions threatConfirmed2 publishers
A privacy reviewer who reads both write-ups of Koa's training comes away with two answers. SiliconANGLE reports that Salesforce "used its customer data to build a massive synthetic dataset crafted out of almost 30 years' worth of internal CRM deployment experiences," in the same paragraph where it quotes Chief Platform and Engineering Officer Rohan Kumar saying the company avoided using customer data at all [20][1]. TechCrunch reports that Salesforce and Nvidia did not use any actual customer data and instead crafted synthetic data that mimicked customers' patterns [5]. Those are different statements, and they lead a data protection officer to a narrow written question: what was the generator conditioned on?
Kumar framed it as institutional knowledge. "There is so much knowledge that we've sort of gathered in building out the CRM over the last 27 years," he said [9]. Jayesh Govindarajan, EVP of Salesforce AI, described the environment they built instead of using records: "We actually simulated a customer service environment with a persona customer service professional, including irate customers that call into the customer service center, all the way to a sales professional who's trying to close a deal" [6].
Provenance is also why Salesforce picked the base model. Govindarajan said Nemotron was the first available pre-trained American model that was state of the art and had clear data provenance, adding: "We have no idea what Qwen trains on" [18].
The reliability number comes from Salesforce itself. In its benchmark tests, the company says Koa matched or exceeded multiple leading general-purpose models on CRM-based tasks with three times fewer errors [7]. Three times fewer puts the error rate at a third of the baseline, a reduction of about 67 percent [16]. Salesforce described the comparison set only as leading general-purpose models and published the ratio without absolute error rates [17]. A pilot team therefore cannot tell whether the baseline missed three steps in a hundred or thirty.
For a team already on Agentforce, this changes a routing rule. Multistep reasoning used to leave the platform for a frontier model such as Claude or ChatGPT through the Agentforce gateway [14]. "But reasoning has always been something that we've relied on the frontier model providers for. Until now," Govindarajan told TechCrunch [13]. Salesforce also announced ClaudeForce with Anthropic, which lets companies use Claude as their interface while data stays in Salesforce's system of records [19].
Then there is cost. TechCrunch reports Koa uses fewer tokens to do the same work [21], and Nvidia's Kari Ann Briski said, "we have a unique architecture for inference to be token efficient" [22]. The Friday review asks a rollout owner about the cases that closed without a human going back in, not tokens per prompt.
Only one pilot voice is on the record, and it is qualitative. Ryan Teeples, chief strategy officer at 1-800Accountant, said "Koa gives our agents the reasoning to work through that complexity step by step and use the right tools along the way" [15]. Koa stays in an expanded pilot with select customers until general availability in winter [8], so that window is where the two claims can be tested. The comparison that settles it is the same hundred cases sent down both paths, scored on the share that closed without a reopen and the tokens each close consumed.
What to watch
- Whether Salesforce names the general-purpose models and publishes absolute error rates behind the three-times-fewer-errors claim before winter general availability.
- Whether Koa is metered or priced separately from other Agentforce models when it reaches general availability.
- Whether the Nemotron rollout for Missionforce Operations actually ships in October for air-gapped and private cloud deployments.
Clarity's read
What the record supports and how the coverage leans. The claims behind it follow.
Reality
- Evidence45
- Adoption20
- Hype gap+35
- Incentives75
- Confidence55
Perspective Coverage
3 publishers- Builder
- Builder 35%
- Operator
- Operator 38%
- Investor
- Investor 27%
Claim ledger
Ranked by verification strength, evidence, and original report placement.
- [1]
By using synthetic data that simulates realistic enterprise workflows across 14 industries, including healthcare, finance and manufacturing, Salesforce avoided using any customer data at all, Kumar said.
- [2]
Salesforce said at its Dreamforce user conference that it partnered with Nvidia to train and release Koa, a specialized AI model custom built for reasoning about CRM data.
- [3]
Koa is based on Nvidia's Nemotron model architecture and is designed to help AI agents reason through complex, multistep workflows within Salesforce's Agentforce platform.
- [4]
Koa is Salesforce's first reasoning model, built on Nvidia's open-weight Nemotron model; the two companies worked together to post-train it to excel at sales, marketing and customer-support-related tasks.
- [5]
TechCrunch reported that Salesforce and Nvidia did not use any actual data from Salesforce's customers and instead crafted synthetic data that mimicked customers' patterns.
- [6]
Govindarajan said: "We actually simulated a customer service environment with a persona customer service professional, including irate customers that call into the customer service center, all the way to a sales professional who's trying to close a deal."
- [7]
In Salesforce's benchmark tests, Koa matched or exceeded the performance of multiple leading general-purpose models on CRM-based tasks with three times fewer errors.
- [8]
Koa is not yet generally available; it is being made available to select customers via an expanded pilot program within Agentforce, with general availability slated for winter, Kumar said.
- [9]
Rohan Kumar, Salesforce Chief Platform and Engineering Officer, said: "There is so much knowledge that we've sort of gathered in building out the CRM over the last 27 years."
- [10]
According to Kumar, Salesforce combined supervised fine-tuning with reinforcement learning and a technique known as group relative policy optimization to teach Koa to execute multistep tool calls with accuracy.
- [11]
Because Salesforce hosts and runs Koa within its own trust boundaries within Agentforce via Nvidia's open framework, customer data never leaves its controlled environment, Kumar said.
- [12]
Salesforce said it will roll out Nvidia's Nemotron models for Missionforce Operations in October, letting customers run them within air-gapped networks and private clouds and automate back-office workflows such as procurement and supply chain management.
- [13]
Jayesh Govindarajan, EVP of Salesforce AI, told TechCrunch: "But reasoning has always been something that we've relied on the frontier model providers for. Until now."
- [14]
Before Koa, if an agent needed to reason through a long-running or multi-step task, those prompts would be routed to a frontier model like Claude or ChatGPT through Agentforce's AI gateway.
- [15]
Ryan Teeples, chief strategy officer at 1-800Accountant, said: "Koa gives our agents the reasoning to work through that complexity step by step and use the right tools along the way."
- [16]
Three times fewer errors puts Koa's error rate at one third of the baseline, a reduction of about 67 percent.
- [17]
Salesforce stated the benchmark result as a relative error ratio, describing the comparison set only as multiple leading general-purpose models and without absolute error rates.
- [18]
Govindarajan said that until Nemotron came along there was no sovereign American pre-trained model available that was state of the art and had clear data provenance, adding: "We have no idea what Qwen trains on", referring to the Chinese open-weight model produced by Alibaba.
- [19]
Salesforce announced a partnership with Anthropic called ClaudeForce that allows companies to use Claude as their AI interface while their data remains in Salesforce's system of records, secured by its infrastructure.
- [20]
SiliconANGLE reported that the company "used its customer data to build a massive synthetic dataset crafted out of almost 30 years' worth of internal CRM deployment experiences."
- [21]
TechCrunch reported that Koa helps reduce AI spending because it uses fewer tokens to do the same work.
- [22]
Kari Ann Briski, Nvidia's VP of Generative AI Software for Enterprise, told TechCrunch that with Nemotron, "we have a unique architecture for inference to be token efficient."
Sources
3 independent publishers whose own reporting we read for this story.
- salesforce.comSalesforce Koa, built on Nvidia Nemotron | Salesforce
2 articles · September 15, 2026
- siliconangle.comSalesforce debuts Koa, a specialized model built to reason about CRM data
2 articles · September 15, 2026
- techcrunch.comSalesforce and Nvidia’s new reasoning model is everything the AI labs should fear
2 articles · September 15, 2026
Topics and entities
Topics
- Synthetic Training Data
- Open-Weight Model Publishing
- Domain-specific reasoning models
- Enterprise AI Agents
Entities
- NVIDIA Nemotron
- Jayesh Govindarajan
- Nvidia
- Agentforce
- Salesforce
- Qwen
- Anthropic