Artificial intelligence is forcing some of the biggest software companies in the world to rethink a question that, until recently, seemed pretty settled. How should software actually be priced?
For decades, enterprise software has been built around people – the more employees that needed access to a platform, the more licenses a company bought. But AI agents change that approach. As they’re designed to automate work rather than simply help someone do it, charging per user becomes an awkward fit.
Few companies have embraced that shift more than Sierra, the AI startup founded by former Salesforce co-CEO Bret Taylor. Since launching in 2024, Sierra has reportedly reached $100M in annual recurring revenue (ARR) in just seven quarters. While much of the attention has focused on its technology, one of its biggest differentiators is arguably its commercial model.
Instead of charging for conversations, searches, or token usage, Sierra charges customers for successful outcomes. If an AI agent can’t complete a task and has to hand the work over to a human, the customer doesn’t pay. It’s a fairly simple idea, but one that puts a lot more risk onto the software vendor.
At the same time, Salesforce has been steadily evolving how it prices Agentforce, moving from conversation-based pricing to Flex Credits, and more recently, introducing pay-per-resolution for Help Agent. So, is this simply where the market is heading? Or is Bret Taylor quietly showing the wider enterprise software industry – including his former employer – how AI agents should really be commercialized?
Selling Accountability
The essence of Sierra’s outcome-based model is fairly simple – customers should just pay for the work an AI agent actually completes, rather than the amount of technology it consumes to get there.
This is quite a meaningful departure from how enterprise software has traditionally been sold. Per-seat pricing may have made sense when software was mainly there to help an employee do their job, but agents are being sold a lot differently. They’re not providing information, drafting responses, or helping someone be more productive. Now, they are completing the whole process themselves.
As Bret Taylor explained in a 2025 interview with The Verge: “For most of our customers, that means when the AI agent autonomously resolves the case … there’s a fee for that. If the AI agent has to transfer to a real person, it’s free.”
This model makes things extremely straightforward for customers to buy into. Instead of paying for thousands of conversations, actions, or tokens and hoping they eventually produce a return, the cost is tied to something the business can actually recognize. Did the agent resolve the customer’s issue? Did it process the return? Did it complete the requested change?
Overall AI adoption at the enterprise level has been mixed; some would even say it’s been pretty slow. A big factor in this is cost and understanding the return on investment. Earlier this year, I covered some of the AI mistakes businesses might make that could cost them millions, and the risk in itself is what many buyers remain cautious about when it comes to AI spending.
With the introduction of outcome spending, it should make those different “what’s the ROI?” conversations much easier, but it also shifts a considerable amount of risk back onto Sierra. The company is basically saying “trust the agent, but only pay us when it works”.
Of course, actually defining a successful outcome is not always straightforward either. A support case may look resolved because the conversation has ended, but did the customer get what they needed? What happens if the agent completes 90% of a task before escalating? And who ultimately decides whether the result was good enough to justify the charge?
Intercom and Zendesk have also adopted versions of outcome-based pricing, particularly around customer service. Sierra’s real differentiator may therefore be less about inventing the model and more about placing it at the center of a wider argument: AI vendors should be rewarded when the work actually gets done.
Salesforce Now Moving in the Same Direction
Salesforce’s AI pricing has changed dramatically over the past two years, and it’s difficult to ignore how closely that evolution aligns with the broader shift toward outcome-based pricing.
Agentforce originally launched around conversation-based pricing before moving to Flex Credits, which charges according to the actions an AI agent performs. More recently, Salesforce introduced Help Agent, taking the concept a bit further by charging only when an AI agent successfully resolved a customer issue.
So, you could say that this newfound philosophy is beginning to look very similar to Sierra’s. Rather than paying for activity, customers are paying for results. Salesforce itself describes Flex Credits as aligning costs with “the business value your AI agents create”, while Help Agent only bills when an issue is resolved without human intervention.
That doesn’t necessarily mean Salesforce is following Bret Taylor’s playbook, though, as the company has a much different problem on its hands.
Sierra is primarily focused on customer-facing AI agents, where success might be a bit easier to define. Salesforce, on the other hand, is trying to commercialize AI across Sales, Service, Marketing, Slack, Platform, and Data 360, where use cases don’t have a binary outcome most of the time.
This makes me think that Salesforce probably won’t settle on a single commercial model. Outcome pricing makes perfect sense for customer service, but other AI workloads may continue to rely on Flex Credits or even traditional user licensing. Rather than copying Sierra outright, Salesforce appears to be adapting the same underlying idea to a much broader and more complex platform.
Final Thoughts
Whether Bret Taylor is deliberately showing Salesforce the future or simply responding to the same market forces is almost beside the point. Enterprise buyers are becoming less interested in how AI works behind the scenes and more interested in what it actually delivers. That shift is pushing software vendors to rethink one of SaaS’s oldest assumptions: how value should be priced.
Salesforce’s move toward outcome-based pricing suggests this is more than just an experiment. While different AI workloads will likely require different commercial models, rewarding software for completed work rather than consumed resources feels like a direction the wider industry is steadily moving towards.
