Enterprise software providers are navigating uncharted waters as they experiment with different monetization methods
AI has completely upended the software pricing model.
Prior to the artificial-intelligence boom, most software-as-a-service companies made money by charging a fixed fee per user. Today, things are a lot more complicated.
Now, in addition to seat licenses, software vendors are offering a diverse menu of pricing schemes - including prepurchased pools of AI credits, flat per-conversation rates, fees for successfully resolved support tickets and other assorted add-ons. It's causing confusion for providers and users of the software alike.
AI has completely upended the software pricing model, challenging both traditional software companies and leading AI companies. For established software companies, charging a client solely based on how many of its employees use the technology is no longer a viable solution, because AI-powered efficiency gains could lead firms to lay off workers. Customers looking to adopt agentic workflows are also discovering the hard way that they don't always know what they're paying for: In May, Uber Technologies (UBER) capped employee spending on tools like Anthropic's Claude Code after burning through its full-year AI budget by April.
"Both customers and vendors are trying to figure out the formula that makes the most sense," Citizens analyst Patrick Walravens told MarketWatch.
The stakes are high as companies experiment with different pricing methods. Rapidly improving AI capabilities are resulting in software products that are constantly evolving, making pricing software an even bigger challenge.
Software providers are dedicating more resources to this puzzle, and some are initiating new roles to meet the task at hand. Last year, Salesforce appointed Craig Shull to become its first-ever chief pricing officer. Shull's mandate is to define the company's monetization strategy as AI offerings evolve, incorporating feedback from sales representatives on the ground.
While ServiceNow (NOW) doesn't have a chief pricing officer, it is actively growing its pricing team in what Robin Manherz, a senior vice president at the software company, called an enterprise "inflection point" in a LinkedIn post earlier this year.
"A lot of the deployment of AI and AI-driven applications is supposed to replace human labor, right?" Walravens said. "We need to move from a world that was primarily just 'per month, per seat,' to a world that looks more like how we pay humans."
Tokenomics
While everyone searches for an AI pricing model that works, companies and customers are quickly discovering what doesn't. Metering usage by tokens - the individual units of data processed by an AI model - is becoming a major headache.
In a CNBC interview last month, Palantir Technologies (PLTR) CEO Alex Karp pushed back against the token-based pricing models of AI labs. "At every single enterprise I deal with, these people are livid. They're like, 'I am paying for tokens that create no value,'" Karp said.
"Nobody has much of a sense of what tokens even mean and how you translate that into costs," Tom Davenport, information-technology and management professor at Babson College, told MarketWatch. "It's one of the more opaque pricing approaches."
Customers who purchase bundles of AI tokens or other AI credits often find it difficult to figure out how many they need to buy, added Davenport, a business analyst who advises corporations on their AI strategies.
Additionally, as AI use cases evolve beyond simple queries, complex agentic workflows are becoming much more token-intensive.
"The transition from an all-you-can-eat subscription to one that's led to quite high costs in a short time is challenging for a lot of executives," Davenport said. "There is a desire to move toward something where you pay for outcomes that are useful and valuable, rather than paying for inputs that people don't understand."
Paying for outcomes
Now, software vendors are eyeing outcome-based monetization, charging customers only when the software delivers a specific, measurable business result. Last month, Salesforce launched Agentforce Help Agent, a prebuilt customer-service agent that charges $2 per resolution. Enterprises only pay when the agent autonomously resolves an issue from start to end. If a customer leaves negative feedback or asks to escalate to a human, there is no charge.
On Oracle's $(ORCL)$ June earnings call, the company shared that it was rolling out outcome-based commercial models. For example, interview agents would be priced on the number of candidates screened. "All of this helps our customers control their costs and align their spending with a value being generated," co-CEO Mike Sicilia said.
AI-native companies such as Sierra are pushing the industry standards for pricing. Sierra, which builds customer-service AI agents, is largely credited for pioneering the outcome-based pricing model. The company charges a platform setup fee then charges per resolution.
Ali Esfahani, CFO of Unconventional AI and a veteran technology investment banker, believes software companies will need to completely overhaul existing seat-based pricing.
"A lot of SaaS companies are becoming specialized data repositories," Esfahani told MarketWatch. "I think the right way to do it is to actually give away the seat at cost - so zero gross margin - and basically charge for the work."
With vibe coding lowering the barriers to entry, customers are paying for the unique data instead of the software user interface, Esfahani said. Cheap seats will attract more users and generate more data, leading to more opportunities for deploying agentic workflows.
Lowering AI costs
Even as the focus shifts toward outcome-based pricing, Citizens' Walravens believes there's no uniform pricing model applicable to AI solutions.
"There will be per seat, there will be platform fees, there will be outcome-based pricing and there will be consumption-based pricing," Walravens said. "Just like in the real world, [where] humans get paid in a lot of different ways."
Pricing AI solutions on their outcomes could win over skeptical CIOs and ease worries about token costs, Constellation Insights editor in chief Larry Dignan wrote in a June note. However, outcome-based models face challenges: "Both parties need to agree on the outcome and metrics," Dignan said. "Who bears the risk of costs to deliver the outcome?"
Vendors risk taking a hit to margins if they charge a fixed outcome fee while using underlying AI-model APIs that charge by token consumption. This setup could result in vendors losing money on each resolution if the AI agent's compute costs exceed the resolution price. For customers, outcome-based pricing removes the predictability of monthly IT budgets.
"The customers are very clear they want hybrid pricing," ServiceNow CEO Bill McDermott told MarketWatch. "The reason for that is they want the predictability around the seat-based licensing model, but they also want usage-based flexibility."
ServiceNow offers tiers of AI-native bundles with hybrid and consumptive meters across its product portfolio. Meters include human and nonhuman identities for its cybersecurity offerings, assets for IT asset management, and "AI Assists" for generative AI capabilities. Enterprise customers are moving up to premium tiers and expanding their software spend, and ServiceNow is anticipating a pricing uplift of 20% to 30%.
As vendors offer more AI-powered workflows and customers spend more on AI capabilities, both are prioritizing return on investment. That's leading to the rise of model routing - a technique that directs queries to the most suitable AI model.
For example, ServiceNow customers can choose to route their AI workloads through ServiceNow's platform-native large language models, or third-party providers such as OpenAI, Anthropic and Google $(GOOGL)$ $(GOOG)$. The model-agnostic platform allows customers to mix and match models for specific use cases, steering less intensive tasks to smaller, cheaper models while reserving complex tasks for more powerful models.
"You should hire the right model for the right job," McDermott said. "You do not need to take a Lamborghini to deliver the mail."
-Christine Ji
Comments