As software companies scale up their AI product sales, the industry is moving away from fixed subscription fees and toward usage-based pricing tied to the actual business value AI delivers, a shift vividly illustrated by Salesforce's recent moves.

The customer relationship management software maker now allows enterprises to choose how they pay for Agentforce, its AI agent platform. Options include negotiable custom contracts where billing is based on either the revenue growth AI generates by helping sales teams close more deals, or the cost savings achieved through automated handling of high-volume customer service interactions.

"Customers want diverse purchasing options and diverse pricing structures. That's something I've felt very deeply recently," Chief Executive Marc Benioff told investors on a Wednesday conference call.

Benioff's comments reflect the widespread uncertainty pervading software pricing in the AI era. Software vendors hold sharply divergent views on the matter, and many companies are under pressure as their AI offerings have failed to meaningfully boost overall revenue growth. Salesforce and other incumbents also face competitive pressure from startups that pioneered "pay-for-outcome" models, where customers only pay when AI actually delivers results.

According to people familiar with the matter, OpenAI has in recent months offered some large customers the option to be billed only when AI completes tasks like客服 conversations, a previously unreported development. An OpenAI spokesperson declined to comment. Customer management startups Sierra and Fin, which Salesforce is acquiring for $3.6 billion, also employ this model, charging customers only when AI completes tasks without human intervention. Code assistant Cognition, meanwhile, promises to compensate clients if engineering output fails to justify the fees paid.

Several established software companies, including Adobe, HubSpot, and Zendesk, have also begun piloting "pay only when AI delivers results" mechanisms.

Where the pricing puzzle stands

Companies are grappling with how to commercialize AI products while managing high supply costs, even as a proliferation of AI tools such as Anthropic's Claude squeezes enterprise IT budgets, further complicating monetization efforts. The rapid rise of Anthropic and its AI agents presents Salesforce with a fresh pricing dilemma. When enterprises deploy advanced AI agents like Claude to handle complex tasks involving Salesforce software, employees interact directly with the original software less frequently, potentially eroding the influence of legacy enterprise software vendors.

For now, Salesforce is embracing this new paradigm: last week the company launched Claudeforce, enabling customers to leverage Claude's large language models to access Salesforce applications and complete substantial business workflows without manually navigating the software interface. Salesforce will likely design multiple pricing options for Claudeforce, tied to a broader strategy of monetizing whenever external vendors' AI accesses data within Salesforce applications. According to people familiar with the sales strategy, customers must upgrade to higher subscription tiers to enable this capability.

Following Palantir's playbook

Salesforce's approach of charging based on whether AI drives sales or reduces costs echoes the pricing logic long employed by software company Palantir. Palantir signs highly customized agreements with large enterprise clients, helping them integrate data and build business applications, with fees structured as a hybrid of fixed base charges plus usage-based and outcome-based components.

Benioff said this flexible AI pricing strategy has helped the company secure "extremely large customer orders." Payment processor Stripe appears to have anticipated future disputes between vendors and customers, having already published operational guidelines for outcome-based pricing.

Salesforce disclosed last week that sales of Agentforce and data management services grew more than twofold year over year. While this growth has not yet significantly lifted overall company revenue, investors have responded positively, with the stock rising roughly 23% since the announcement.

"We're not simply doing outcome-based billing like 'complete this many calls, pay us $2,'" Benioff said. "What we want to achieve is: 'We helped you increase revenue by this much, so pay us $2, because we generated $20 or $40 for you.'" He noted that this model allows software vendors to command premium pricing.

Palantir serves as a prime example, having adopted this model for some clients and experiencing explosive revenue growth over the past year.

Benioff understands well how pricing models can build competitive moats. More than two decades ago, Salesforce led an industry transformation: instead of purchasing business software outright, enterprises began paying per-user subscription fees. This eased the upfront procurement burden for small and mid-sized businesses while allowing easy scaling as companies grew. Another major advantage: software upgrades were handled entirely by Salesforce on the backend, eliminating the traditionally expensive and time-consuming upgrade processes. That shift inaugurated two decades of prosperity for the SaaS software industry.

Yet the transition was not always smooth: systems monitoring software vendor Splunk experienced short-term revenue declines during its shift from perpetual software licenses to a subscription model.

The attribution dispute concern

But AI is now disrupting the traditional subscription model, and Salesforce finds itself a follower in software pricing. The outcome of this experimental phase will determine whether legacy enterprise software giants like Salesforce can withstand the seismic changes AI brings to the industry.

As outcome-based pricing gains momentum, a thorny question emerges: attributing cost savings and revenue growth to either the customer's own operations or the software product itself becomes increasingly murky. Stripe has already published guidance on outcome-based pricing, proactively flagging the risk of vendor-customer disputes. Stripe points out that sales conversions or other business outcomes "may stem from product tweaks, marketing campaigns, seasonal fluctuations, rather than the software itself."

"Without clearly defined attribution rules, customers will argue over whether a particular business outcome should be credited to the software vendor," the guidance warns.