Add to Google Preferred Sources
Artificial intelligence is upending the subscription pricing model that has defined the enterprise software industry for more than two decades. Salesforce CEO Marc Benioff announced that the company will allow enterprise customers to choose how they pay for its AI product Agentforce, including pricing based on revenue growth generated by AI-assisted deals or cost savings from automated customer service. OpenAI, Sierra, and Fin are among the companies following suit with pay-on-completion models. The shift marks a fundamental challenge to the SaaS subscription model that Salesforce itself pioneered, reflecting a broader transition in AI commercialization from capability delivery to business outcomes. Yet outcome-based pricing carries inherent attribution risks—Stripe has already issued guidance warning that sales conversions may stem from product adjustments, marketing campaigns, or seasonal factors rather than the software itself. Financial data from Palantir and DeepZero show that AI companies embedded in core enterprise decision-making are demonstrating operating leverage with profit growth outpacing revenue growth, but the success or failure of these pricing experiments will ultimately determine whether legacy software giants can reinvent themselves in the AI era.
Key Elements
Artificial intelligence is shaking the pricing foundation that has underpinned the enterprise software industry for more than two decades. Legacy players led by customer relationship management giant Salesforce (CRM) are being forced to shift away from the fixed per-seat subscription model toward usage-based pricing—and even pricing tied to actual business outcomes. The transition presents both opportunity and deep uncertainty, and the latest remarks from Salesforce CEO Marc Benioff have brought this pricing experiment squarely into the open.
According to an August 30 report from tech publication The Information, Benioff told investors on a conference call last week that the company is allowing enterprise customers to choose their own payment structure for its AI product Agentforce, including customized contracts that bill based on revenue growth generated by AI helping salespeople close more deals, or on cost savings achieved through automated customer service.
“Customers want to buy and price in different ways—that’s something I’ve come to appreciate deeply recently,” Benioff said. The remarks reflect a period of profound experimentation in software pricing in the AI era. Salesforce shares have climbed roughly 23% since its earnings release, signaling that investors are giving initial approval to the strategic pivot.
Twilight of the Subscription Era
Salesforce’s transformation amounts to rewriting the software-as-a-service (SaaS) business model it established twenty-five years ago. The company led the industry’s shift from one-time perpetual licenses to per-employee subscription fees, dramatically lowering upfront costs for small and mid-sized businesses while shifting the burden of software upgrades and maintenance to vendors—ushering in two decades of SaaS industry prosperity.
But the rise of AI is inverting that logic. As enterprises increasingly deploy advanced AI agents like Anthropic’s Claude to handle complex tasks that previously required operating applications such as Salesforce, the frequency with which employees directly interact with these applications is declining. The per-seat subscription model is losing its original foundation.
Benioff acknowledged that software pricing is in a period of deep uncertainty, and that Salesforce is now following the lead of startups rather than driving change itself.
Converging on Outcome-Based Pricing
The new model Salesforce is exploring bears a striking resemblance to the approach long employed by data analytics software company Palantir (PLTR). Palantir signs highly customized contracts with enterprise clients that combine fixed fees with usage- and outcome-based charges. Benioff said this flexible pricing approach has helped Salesforce “sign very large deals,” noting that vendors can “command extremely high pricing” for their products under this model.
He elaborated on his understanding of outcome-based pricing: “We don’t just want to say ‘you completed this many calls, so that’s $2.’ We want to be able to say ‘we helped you grow revenue by this much, so that’s $2—because we helped you earn $20 or $40.'” This means Salesforce aims to tie its own revenue deeply to customer business outcomes, rather than merely metering task completion.
Industry Chain Reaction
The shift is already triggering ripple effects across the industry. People familiar with the matter say OpenAI has in recent months begun offering select large customers the option to pay only after AI completes tasks; customer management startups Sierra and Fin—the latter being acquired by Salesforce for $3.6 billion—have likewise adopted pay-on-completion models. Coding assistant Cognition has pledged up to $10 million in credits if it fails to deliver engineering outcomes at least equal in value to what enterprise customers paid.
Facing pressure from AI-native competitors like Anthropic, Salesforce last week launched Claudeforce, a service that allows customers to use Claude directly to complete a wide range of tasks involving Salesforce applications without needing to operate those applications themselves. People familiar with the matter say Salesforce plans to build a monetization mechanism through Claudeforce: whenever a third-party AI calls data within Salesforce applications, Salesforce can capture value, and customers must upgrade to higher subscription tiers to enable the feature.
The strategic intent behind this move is clear: even if users no longer interact directly with the Salesforce interface, the company can maintain its core position at the data layer within the AI ecosystem, converting potential user attrition risk into a new monetization gateway.
Attribution Disputes: The Hidden Pitfall
Outcome-based pricing is theoretically appealing, but in practice it can trigger complex attribution disputes. Payment services provider Stripe has already issued guidance on the matter, explicitly noting that sales conversions or other business outcomes “may stem from product adjustments, marketing campaigns, or seasonal factors” rather than the software’s contribution.
“Unless attribution rules are clear, customers may dispute whether outcomes should be credited to the software vendor,” Stripe stated.
This concern is not without precedent. Software monitoring company Splunk experienced a period of revenue decline during its transition from a license model to a subscription model. Analysts believe the results of the current pricing experiments will largely determine whether legacy enterprise software companies like Salesforce can successfully reinvent themselves amid the AI wave.
Financial Validation of Decision AI
Behind the pricing shift lies a divergence into two distinct growth curves in AI commercialization: one that relies on model capability and rapid user scale expansion, and another that depends on continuous penetration into core enterprise operations, transforming AI from “capability delivery” into “business outcomes.” The latter is exemplified by Palantir, as well as China’s DeepZero.
Palantir delivered second-quarter revenue of $1.94 billion, up 93% year over year, with net income of $1.07 billion—more than tripling from a year earlier—and a GAAP net margin of 55%. CEO Alex Karp has declared that this level of growth can continue for at least another 18 months.
DeepZero’s interim report for the first half of 2026 shows similar characteristics. Revenue reached RMB 398 million (approximately $59.2 million), up 43.6% year over year; period profit was RMB 8.2 million (approximately $1.2 million), up 125.2%; operating profit surged 402.2%, with profit growth nearly triple the revenue growth rate. Founder Huang Xiaonan’s “Dual 100 Strategy”—expanding customer scale by a hundredfold on one end and per-customer product count by a hundredfold on the other—is being validated by segment-level business data.
During the reporting period, agent marketing services generated revenue of RMB 370 million (approximately $55.1 million), up 42.9% year over year; agent software revenue reached RMB 28.2 million (approximately $4.2 million), up 52.9%, outpacing overall revenue growth. New orders for agent products and services in the first half grew more than 300% year over year, with 30% of new agent customers purchasing multiple agents in a single transaction and some large enterprises deploying more than a dozen products at once.
On the cost side, while revenue grew 43.6% year over year, selling and marketing expenses rose only 9.9% to RMB 25.27 million (approximately $3.8 million), and per-capita output exceeded RMB 1.2 million (approximately $179,000). R&D investment totaled RMB 33.68 million (approximately $5 million), up 44.3% year over year, roughly in line with revenue growth. Economies of scale have begun to dilute selling and back-office costs, ultimately flowing through to the bottom line.
What Will Decide the Pricing Experiment
In an article published in August 2026, McKinsey argued that AI’s greatest economic benefits rarely come from labor savings alone, but rather from faster decision-making, better asset allocation, and capturing opportunities that would otherwise be missed. Buying Copilot accounts is not much different from buying Office software—per-head pricing—but this budget line is scrutinized repeatedly inside enterprises: Are employees actually using it? How much time is being saved? What value is being created from that saved time?
Decision AI operates on a completely different logic. There is an iron rule in enterprise budget approval: only core business functions that directly impact revenue, profit, risk, or cost can secure long-term, recurring budgets. Decision AI emphasizes mapping technical capabilities to business outcomes, with value that can be anchored to revenue, profit, risk, or cost—enabling a clearer ROI measurement framework.
In its 2025 AI Hype Cycle report, Gartner classified decision intelligence as a “transformational” technology. Cloud computing was designated “transformational” in 2010, and generative AI received the same classification in 2023. At the time of each designation, many questioned the commercial value and practical viability of these technologies, but in the years that followed, both demonstrated the prescience of those judgments through explosive growth.
Wall Street technology analyst Gil Luria recently pushed back directly against the notion that “companies like Palantir are just wrapping a large language model in a shell.” He pointed out that underlying models can be swapped flexibly; what is truly difficult to replicate is the years of accumulated business ontology and customer deployment methodology. Decision AI is not something you can get running by downloading a large model and plugging it into enterprise data.
Salesforce’s pricing experiment is, at its core, answering the same question: when AI directly participates in core enterprise operations, how exactly should software companies charge? By usage volume, by task completion count, or by actual business value created? Benioff’s answer is the latter—but the attribution challenge has only just begun.
Once added, BigGo Finance appears first in Google Search Top Stories, so you get the broadest, most up-to-the-minute, and most comprehensive global financial news first.
