Salesforce is tearing up its own pricing playbook for AI. According to The Information, the company is overhauling how it charges customers for its AI agents, moving away from the per-seat license model that built its empire and toward pricing tied to usage and outcomes. That’s a big deal, and not just for Salesforce.
For two decades, enterprise software ran on a simple formula: pay per user, per month, forever. Salesforce practically wrote that rulebook. AI breaks it. When an AI agent does the work a human used to do, charging per human seat stops making sense. Fewer seats, same work. So the vendors are scrambling to find a new way to get paid.
What’s actually changing
Salesforce has been testing several models for its Agentforce product, and the direction is clear:
- Per-conversation pricing. Early on, Salesforce floated roughly $2 per AI conversation. Simple to understand, but it punishes customers for using the product more, which is the opposite of what you want.
- Consumption-based credits. Customers buy a pool of usage and draw it down as agents run tasks. This mirrors how cloud infrastructure and API access already work.
- Outcome-based pricing. The most ambitious version: charge only when the AI resolves a case, closes a ticket, or books a meeting. You pay for results, not activity.
The Information’s reporting shows Salesforce isn’t settling on one answer yet. It’s feeling out what customers will actually accept, because the wrong model can stall adoption or leave money on the table.
Why this matters now
This is the pricing question hanging over the entire AI industry. Salesforce isn’t alone. OpenAI has been testing pay-only-when-it-works pricing. Microsoft, Google, and a wave of startups are all wrestling with the same tension. When your product replaces labor instead of assisting it, per-seat billing quietly caps your own upside.
Outcome-based pricing sounds great in a pitch deck. In practice it’s messy. Who decides when an outcome counts? If an AI agent handles a support ticket but the customer follows up angry, did it work? Attribution gets complicated fast, and complicated pricing scares off buyers who just want a predictable line item on the budget.
What stands out here is the admission baked into the move. Salesforce is effectively conceding that AI changes the unit of value from “who uses the software” to “what the software gets done.” That’s a philosophical shift dressed up as a billing change.
The catch for Salesforce
There’s real risk in this transition. Salesforce’s revenue and its sky-high margins are built on predictable, recurring per-seat subscriptions. Consumption and outcome models introduce volatility. Revenue can swing with how much customers use the product, which makes forecasting harder and can spook investors who prize the steady growth Salesforce is known for.
There’s also a trust problem to solve. Customers have watched cloud bills balloon from usage-based pricing before. Sell them AI agents on a “pay per outcome” promise, then hand them a surprise invoice, and you’ve bought yourself a churn problem.
Practical takeaways
If you’re a business buying AI tools, this is the moment to read the fine print:
- Model your costs at scale, not at pilot size. Per-conversation or per-outcome pricing that looks cheap in a trial can get expensive when you roll it out company-wide.
- Nail down how “outcome” is defined before you sign. Get it in writing. Ambiguity always favors the vendor.
- Watch for pricing lock-in. Consumption credits that expire or don’t roll over are a classic way to inflate spend.
For software vendors, the signal is louder: the per-seat era is bending. The companies that figure out AI pricing customers actually trust will win the next decade of enterprise software.
Salesforce is making its bet in public, and everyone in the industry is watching to see if it holds. More detail is available in the original reporting from The Information.