AIO Apex
en

Vertical AI Agent Startups Are Killing the Per-Seat SaaS Pricing Model

Share:
Vertical AI Agent Startups Are Killing the Per-Seat SaaS Pricing Model

For twenty years, software pricing followed a simple logic: count the humans who touch the product, charge per seat, and let usage grow with headcount. That logic breaks the moment the thing using the software isn't a human sitting at a desk but an AI agent handling thousands of customer conversations a day. Vertical AI agent companies — Sierra, Decagon, and the agent layer inside Salesforce's Agentforce — are quietly proving that per-seat pricing has no answer for a product that doesn't have seats.

What outcome-based pricing actually looks like. Sierra, the customer-service agent startup founded by former Salesforce co-CEO Bret Taylor, charges primarily per resolved interaction. Industry estimates put that figure around $1.50 per resolution, and Sierra's own materials reportedly say that if a conversation goes unresolved, in most cases there's no charge at all. Intercom's Fin agent publishes a cleaner number: $0.99 per successful resolution. Decagon takes a different angle — several sources describe its most common contract as billing per conversation handled, regardless of whether the agent actually resolved the customer's issue. Three vendors, three different definitions of what you're actually paying for, and none of them map to a seat count.

Salesforce's hybrid shows the tension directly. Agentforce didn't abandon per-seat pricing — it bolted a consumption model onto it. Published rates include $0.10 per action or $2 per conversation (the latter reportedly limited to existing Conversations customers), while Agentforce add-ons still run $125 per user per month. That's a company trying to protect its existing per-seat ARR base while hedging against a future where enterprise buyers refuse to pay a human-sized license fee for a system that scales nothing like a human.

Why the old model can't survive this. Per-seat pricing was never really about seats — it was a proxy for usage, because usage scaled with headcount in a world where software assisted humans rather than replaced their output. An AI agent breaks that proxy completely. One Sierra deployment can handle the conversation volume of fifty support reps without needing fifty licenses, which means a vendor charging per seat for an agent is either wildly underpricing a system doing the work of fifty people, or a buyer paying per-seat is unwittingly paying far less than the value delivered — neither is sustainable once a competitor offers to price on outcomes instead.

The budgeting problem this creates for buyers. Consumption and outcome pricing solves the seat-mismatch problem but creates a forecasting one. A support team that resolves 40,000 tickets a month has a bill that moves with ticket volume, not with a number procurement set in a contract negotiation. That's a harder line item to budget against than a flat per-seat fee, and it means finance teams evaluating AI agent vendors need usage-volume models, not headcount models, before signing. The contracts also hide real variation in what counts as a chargeable event — an unresolved Sierra conversation may be free, while an unresolved Decagon conversation may still bill. That distinction is worth more negotiating leverage than most buyers currently spend on it.

What this means for incumbent SaaS. Companies that built valuation multiples on predictable per-seat ARR growth now face a genuine strategic choice: bolt consumption pricing onto existing contracts (Salesforce's path), risk a vertical AI-native competitor undercutting them on a per-outcome basis, or accept that some categories of software — especially support, sales development, and other high-volume conversational work — are moving to a pricing model that doesn't reward per-seat incumbents at all.

Takeaways for buyers evaluating AI agent vendors: model total cost against your actual conversation or task volume, not your current headcount — a cheap per-resolution rate can still cost more than a human team at sufficient scale. Get explicit contract language on what counts as «unresolved» or «non-billable,» since that definition is where vendor margin hides. And treat per-seat add-ons stacked on top of consumption pricing, as Agentforce does, as a signal the vendor hasn't fully committed to outcome-based economics yet — which may mean better negotiating room for you today, before the model fully shifts.

Share:
Why AI Agent Startups Are Killing Per-Seat SaaS Pricing (Sierra, Decagon, Agentforce) | AIO APEX