Modern Pricing & UBB
February 26, 2026 ai-pricing

Salesforce's Agentforce Is a Pricing Experiment in Real Time

In September 2024, Salesforce announced Agentforce with a pricing structure that sent shockwaves through enterprise software pricing circles: $2 per conversation. Not per seat. Not per month. Per conversation that an AI agent handles. It's the most high-profile deployment of outcome-adjacent pricing in enterprise software history, and it's happening at a company with $35 billion in annual revenue. Everyone is watching.

To understand why this matters, you need to understand what Salesforce is betting.

The $2 Conversation Bet

Agentforce agents handle customer service, sales development, commerce, and HR workflows autonomously. A "conversation" is a session between the AI agent and a customer, employee, or prospect — from initiation to resolution. At $2 per conversation, a company running 500,000 AI-handled service interactions per month is paying $1M/month to Salesforce for its AI agents. That's replacing human agents who might have cost $3-5M/month in fully-loaded salary and overhead. The ROI case is straightforward on paper.

What's novel is what Salesforce is not charging. They're not charging per user for Agentforce access. They're not charging per API call to their LLM. They're charging for the output — the conversation completed. This is as close to outcome-based pricing as any major enterprise software company has shipped at scale. The conversation is not purely an outcome (it doesn't verify that the customer problem was solved), but it's far closer to value delivery than a seat or a token.

What This Signals About Enterprise AI Pricing

BCG's 2025 report on agentic AI adoption identifies Salesforce's per-conversation model as the inflection point that "legitimizes outcome-adjacent pricing for enterprise AI." Before Agentforce, per-conversation pricing existed in niche AI tools. After Agentforce, every enterprise software procurement team has a mental model for AI agent pricing that isn't seat-based. That's a category-defining shift.

The signal is clear: the major enterprise software vendors are betting that AI capability is distinct enough from their existing seat-based products to justify a separate pricing motion. Salesforce isn't replacing Einstein AI seats with conversation pricing — they're adding conversation pricing on top of, and eventually potentially replacing, their existing per-seat model for service cloud. ServiceNow has moved similarly with their Now Assist features. Microsoft Copilot moved from seat-only to a hybrid model with consumption elements. The direction is consistent across the major platforms.

The economic logic for vendors

From Salesforce's perspective, per-conversation pricing solves the adoption problem elegantly. Customers don't need to commit to a fixed number of agent seats before they know how many conversations their agents will handle. They can start small, prove the ROI at low volume, and scale up — with Salesforce's revenue scaling in proportion to the customer's AI adoption. This is the classic usage-based adoption advantage, applied to enterprise AI.

The risk that comes with it

The same risk that hit Snowflake in 2023 applies here. Customers will optimize their Agentforce conversations — handling only the cases where AI deflection is effective, routing complex cases back to humans, engineering their conversation flows to minimize $2/conversation charges. Salesforce's revenue from Agentforce is fundamentally correlated with how many conversations their customers choose to route to AI agents, not with their total customer service volume. Customers who get good at routing only appropriate conversations to Agentforce could underspend Salesforce's initial projections significantly.

Can Outcome-Based Pricing Scale to $35B?

Here's the honest question: can outcome-based or outcome-adjacent pricing carry the revenue load of a major enterprise software company? Or is it necessarily a supplement to, rather than a replacement for, seat-based subscription revenue?

The current answer from Salesforce's own financials is: it's a supplement. Agentforce conversations are additive to existing Salesforce subscriptions. Customers still pay for Sales Cloud seats, Service Cloud licenses, and Einstein AI add-ons — and then pay additionally for Agentforce conversations. The per-conversation model is growing Salesforce's total revenue, not replacing their existing subscription base.

Whether outcome-based pricing can ever replace subscription revenue at $35B+ scale is an open question. BCG's analysis suggests that at sufficient maturity — when AI agents are handling the majority of workflows that human users once handled — the revenue base must follow the AI agent volume, not the human headcount. At that point, per-conversation or per-outcome pricing isn't a supplement anymore. It's the business.

Salesforce is the canary in that coal mine. The rest of enterprise software is watching.

Sources

← AI & Agentic Pricing · All posts