Salesforce Expands Its Agentic AI Strategy at Dreamforce, and BMO Raises Its Price Target

Salesforce (NYSE: CRM) used Dreamforce this week to share a platform vision built around a straightforward idea: rather than asking customers to come to Salesforce to use its AI agents, bring the agents to wherever customers already work. That approach, presented under the AIforce banner, anchors much of what the company shared at the conference, from a new reasoning model developed with NVIDIA to a range of pricing options that even supportive analysts acknowledge may make Salesforce's AI revenue somewhat harder to model in the near term.

The AIforce platform strategy

AIforce reflects Salesforce's effort to make its underlying platform — the data, business logic, security model, and permissions behind its CRM — available outside of Salesforce's own interface. Under the strategy, that platform layer can be reached through Claude, Slack, ChatGPT, and Salesforce's own Agentforce Coworker, so users need not log into Salesforce directly.

The company is pairing that access layer with a broader goal: enabling AI agents to work not only across Salesforce but also across other enterprise systems such as SAP and Workday, all from a single environment. If this works as described, it could position Salesforce less as one CRM application among several and more as the trust and permissions layer beneath a customer's entire AI-agent workflow, a meaningfully larger role than the one it plays today.

Flexible pricing and what it means for modeling

Alongside the platform news, CEO Marc Benioff described the range of ways customers will be able to pay for Salesforce's AI agents going forward: per-user, per-agent, consumption, usage, transaction-outcome, or business-outcome pricing.

That balance between greater deal flexibility and reduced forecasting clarity is the trade-off Wall Street is now weighing, and it is reflected in how BMO framed its own price-target increase below.

Koa: Salesforce's first CRM reasoning model

The most technical announcement of the week was Koa, Salesforce's first CRM-specific reasoning model for Agentforce, developed in partnership with NVIDIA. Rather than simply generating a response, Koa is designed to reason through the steps behind a complex, multi-step business task, such as updating an opportunity, routing a support case, or determining the next best action in a service escalation.

• Built on NVIDIA Nemotron: Koa was created by post-training NVIDIA's Nemotron 3 Super model, keeping the training and inference process inside Salesforce's own infrastructure.
• Trained on synthetic data, not customer data: Salesforce says Koa's training corpus was built from synthetic CRM scenarios modeled on 27 years of enterprise CRM knowledge, and that no customer data was used in training.
• Benchmark results: on Salesforce's own CRM Benchmark, the company reports that Koa matched or exceeded leading model performance on CRM actions, with three times fewer errors.
• Government and regulated deployments: Salesforce and NVIDIA are extending Nemotron-based models and accelerated computing into Missionforce, aimed at government and regulated customers that need private cloud or fully air-gapped environments.

It may be helpful to note that the CRM Benchmark result is Salesforce's own, self-administered comparison rather than an independent one. This is a relevant distinction, as outside reporting, including a Bloomberg investigation cited by several outlets covering Dreamforce, has raised questions about how closely Agentforce's real-world enterprise deployments have matched Salesforce's marketing claims so far.

The Street's response: BMO raises its target

BMO's Bachman said he left Dreamforce "more constructive" on Salesforce's ability to benefit from enterprise AI adoption, crediting new products and partnerships, including AIforce, with easing adoption and building more durable value. He also acknowledged the pricing consideration described above, that new AI pricing models could introduce more volatility, while observing that AI-related annual recurring revenue remains small relative to Salesforce's total ARR today, which limits the near-term impact of that volatility. On balance, Bachman said he comes away with "greater confidence in FY28 revenue potential," which supported the target increase.

Guidance remains unchanged

Salesforce did not raise its own financial targets at Dreamforce. The company reaffirmed its fiscal 2030 revenue target of at least $63 billion, a target it had already raised earlier this year, at its fiscal fourth-quarter earnings call, from an original goal of $60 billion-plus set at last year's Dreamforce. This week's announcements therefore centered on strategy and product rather than a new guidance increase, and management appears comfortable that AIforce, Koa, and the new pricing options fit within the growth plan it has already outlined.

What to watch

Several near-term signals may be helpful to follow: how quickly Koa moves from pilot customers to general availability (Salesforce has previously targeted this winter for U.S. regions), whether AIforce's cross-platform access through Claude, Slack, and ChatGPT translates into measurable usage, and whether independent, third-party benchmarks corroborate Salesforce's own CRM Benchmark results for Koa.

Key takeaways

• This week centered on platform and product rather than guidance. Salesforce reaffirmed its existing $63 billion FY30 target rather than raising it, which places the emphasis on AIforce and Koa to help deliver the growth already reflected in that figure.
• Pricing flexibility brings both benefits and considerations. More ways to pay may help win more deals, though it can also make Salesforce's AI revenue somewhat harder to forecast in the near term, a trade-off BMO acknowledged while raising its target.
• Koa's benchmark results are self-reported. The three-times-fewer-errors result from Salesforce's own CRM Benchmark is a meaningful data point, though it has not yet been independently verified, and it arrives alongside outside reporting that raises questions about Agentforce's real-world return on investment so far.

Frequently asked questions