The cynical take on enterprise AI: every software company is slapping "AI" on their product, nobody is actually making money, and the whole thing is a speculative bubble that will pop when the CFOs start asking about ROI. The cynical take has evidence. Microsoft Copilot revenue is opaque — buried in the "Azure and other cloud services" line and never broken out. Google's AI monetization story is "trust us, it'll work." OpenAI's enterprise revenue is real but the margin structure is negative at current pricing.
And then there's Salesforce.
Agentforce launched in September 2024 at $2 per conversation. Not per seat. Not bundled into an existing SKU at no additional charge. Per conversation — a consumption-priced AI product with a direct revenue line. In Q4 FY2025, Salesforce reported that Agentforce had handled over 380,000 conversations in its first full quarter of availability. In Q1 FY2026, that number was 2.1 million. In Q2 FY2026, the most recent quarter, it was 5.8 million. The trajectory is not linear.
At $2 per conversation, 5.8 million conversations is $11.6 million in quarterly revenue — small relative to Salesforce's $9.3 billion total revenue. But the compounding math is what matters. Salesforce has 150,000 customers. The average enterprise customer runs thousands of customer service interactions per month. If 20% of Salesforce's customer base deploys Agentforce across their service and sales operations at scale, the annual revenue from this one product line runs into the billions — and it's recurring, growing with customer volume, and tied to a platform that has two decades of CRM data and workflow configuration behind it.
The CRM data moat is the thing nobody talks about. Every other AI agent company — and there are dozens — starts from zero. They have a model. They do not have 20 years of your company's sales pipeline history, customer interaction logs, support case resolutions, and workflow automation rules. Salesforce does. When Agentforce generates a response to a customer inquiry, it's drawing on the actual history of that customer's interactions with that company. When it recommends a next-best-action for a sales rep, it's working from the opportunity data that has been accumulated across years of deals. This is not replicable by an LLM alone.
Microsoft could theoretically compete on distribution. But Microsoft's CRM data is in Dynamics, which has single-digit market share versus Salesforce's roughly 24% of the CRM market. The data advantage is structural and growing — every Agentforce conversation generates training data that improves the model, every improvement makes the product more useful, more usage generates more data. It is a flywheel, not a feature launch.
The contrarian position: Salesforce Agentforce has the unit economics, the data moat, and the distribution to become the first enterprise AI product line that delivers billions in standalone, identifiable revenue. The $2 per conversation is not expensive — it's the mechanism by which the moat compounds, and no competitor has a comparable foundation to build on.