Zendesk and Sierra say AI agents should be paid per result, not per seat

Zendesk CEO Tom Eggemeier and Sierra co-founder Clay Bavor said traditional per-seat software pricing is being displaced by outcome-based pricing for AI agents. Both executives argued customers should pay only when AI agents successfully complete tasks, signaling a potential shift in SaaS monetization models toward usage and performance-based revenue.

Analysis

Outcome-based AI pricing is economically attractive only where the customer can measure resolution, conversion, or containment cleanly; customer support is one of the few software categories that qualifies. That creates a potential revenue-per-interaction expansion for public CX vendors such as NICE, Five9 and Salesforce, but it also shifts risk from customer budgets to vendor model performance, inference costs and service-level penalties. Gross-margin dispersion—not headline AI adoption—should become the key valuation differentiator over the next 2-4 quarters.

The near-term risk for incumbent SaaS is cannibalization: replacing a paid human seat with an agent can reduce conventional subscription ARR before usage revenue scales. Vendors with large installed bases and proprietary workflow data should retain an advantage, while horizontal copilots and point-solution chat vendors face commoditization as pricing converges toward measurable business outcomes. Public markets may initially reward any “agentic” packaging announcement, but investors should demand disclosure of effective price per resolved case, model/infrastructure cost per case, and net revenue retention after seat displacement.

There is no standalone trade from executive rhetoric. The investable catalyst is upcoming earnings guidance: a vendor showing higher automation penetration alongside stable or rising gross margin and NRR can earn a multiple premium; a vendor reporting AI attach rates without unit economics is vulnerable to a post-earnings reversal. The thesis is falsified if enterprise buyers retain seat commitments as the primary procurement unit through 2026, or if agent error rates force broad human escalation that prevents meaningful labor substitution.

Key Decisions for Investors

• Maintain a 1-3 month watchlist on NICE and FIVN for earnings disclosures of automated-resolution volume, AI gross-margin impact and NRR; initiate longs only if management demonstrates that AI revenue is incremental rather than replacing higher-margin seat revenue. Target 10-15% upside on a credible re-rating; exit on gross-margin guidance cuts or rising implementation costs.
• Use CRM as the liquid large-cap read-through for outcome-priced enterprise AI, but avoid chasing AI-product announcements absent disclosed paid usage and margin data. A post-results long is justified only if Data Cloud/Agentforce growth supports durable subscription expansion rather than discounting of core seats.
• Consider a selective pair after earnings: long NICE versus short a lower-scale contact-center/communications software basket proxy such as ZM, only if NICE shows measurable automation economics. The intended payoff is multiple divergence from proprietary data/workflow advantages; cover if the spread widens 10% without confirming fundamental data.
• Treat elevated AI-agent adoption claims as an alert for customer-support labor exposure rather than an immediate software beta trade. Monitor outsourcing firms and BPO-sensitive names for contract-volume commentary over the next 6-18 months, when lower ticket volumes—not software bookings—would validate structural displacement.