Per-Outcome AI Pricing Is Emerging: Enterprise Contracts Shift Beyond Per-Token

Per-Outcome AI Pricing Is Emerging: Enterprise Contracts Shift Beyond Per-Token

For most of the past decade, documented across industry reports like Bessemer Venture Partners State of the Cloud, enterprise software pricing was simple enough that a procurement officer could explain it to a board in one sentence: you pay per seat, per year, and the vendor locks you in for three. Then AI arrived, and the sentence got complicated. Per-token.

Per-conversation. Per-action. Per-resolution. By 2026, the enterprise software industry is in the middle of a third pricing revolution that is redefining modern AI pricing models in five years — and this time, the vendors are selling outcomes, not access.

The Three-Generation Shift in AI Pricing

The trajectory is well-documented inside the analyst community. Per-seat licensing defined enterprise software from the 1990s through roughly 2022. Every user of Salesforce, ServiceNow, or Microsoft Office 365 cost a fixed monthly fee, billed annually, with a multi-year true-up. It was predictable for both sides: vendors built reliable ARR forecasts; finance teams could amortize costs in a spreadsheet.

Generative AI broke the legacy seat model, forcing a fundamental reassessment of AI pricing. The underlying economics of AI inference — compute costs that scale with tokens processed, not headcount — made per-seat pricing a poor proxy for vendor costs. Between 2023 and 2025, the industry experimented with per-token and per-conversation billing.

Salesforce launched Agentforce in Fall 2024 at $2.00 per conversation. GitHub Copilot charged per active seat. OpenAI and Anthropic priced their APIs in fractions of a cent per thousand tokens. (For a deeper treatment of how per-token economics reshaped enterprise AI contracts, see NWI’s earlier coverage of AI consumption pricing and the per-token enterprise shift.)

But per-token and per-conversation pricing exposed a new problem: it billed for AI activity, not AI results. Salesforce’s $2/conversation model carried a hidden sting — at a 60% resolution rate, the effective cost of a successful resolution was $3.33, because customers were paying for the 40% of conversations in which Agentforce failed. Enterprise procurement teams paying for failed AI interactions found themselves back in the ROI quagmire they thought AI was supposed to solve.

Per Deloitte’s 2026 State of Generative AI in the Enterprise, 72% of enterprise AI projects exceeded their original budget by at least 30%. That figure, circulating through CFO offices across the Fortune 500, is accelerating the move toward a third pricing generation: per-outcome.

What Outcome-Based Pricing Looks Like in Practice

The cleanest articulation of the outcome model is Intercom’s Fin agent, priced at $0.99 per outcome. The logic is unambiguous: a “resolution” occurs when, following Fin’s last response, the customer either confirms the answer solved their problem or exits the conversation without requesting further help. If Fin doesn’t resolve the conversation, the customer doesn’t pay. At 5,000 resolutions per month, that maps directly to support tickets avoided and human agent hours saved.

Zendesk moved in the same direction, becoming the first major CX platform to formalize outcome-based pricing for AI agents: $1.50 per automated resolution on committed volume, or $2.00 pay-as-you-go. Zendesk’s definition is strict — the AI agent must handle the interaction end-to-end, with no human intervention. HubSpot’s Customer Agent, launched in April 2026, undercut both at $0.50 per resolved conversation, a competitive price point that signals a race to establish outcome-pricing floor rates across the mid-market.

The pure-play enterprise AI vendors are structuring even larger contracts around outcomes. Sierra AI, which builds AI agents for large enterprise deployments, has published its rationale for outcome-based pricing explicitly: vendors should be compensated in proportion to business results generated, not compute cycles consumed.

Sierra’s year-one enterprise contracts run $200,000 to $350,000 or more, with implementation fees of $50,000 to $200,000 negotiated separately. Decagon, a competing enterprise AI agent platform, structures contracts on per-conversation or per-resolution terms negotiated directly — annual contracts range from roughly $95,000 to more than $590,000 based on volume and integration complexity, according to external contract reviews rather than published price lists.

Vendors Not Converging on a Single Formula

The transition is far from uniform. Salesforce has introduced three distinct pricing models for Agentforce: a standard consumption model at $2.00 per conversation; an Flex Credits structure for enterprise clients with variable volume; and committed-tier pricing negotiated into custom master services agreements. ServiceNow has resisted pure outcome pricing, arguing through leadership statements that customer-environment integration and workflow complexity make binary resolution attribution too contentious for multi-tenant enterprise software.

The competitive dynamics driving Salesforce and ServiceNow’s divergent approaches are also examined in NWI’s coverage of the emerging Salesforce-ServiceNow pricing war, which maps how each vendor’s architecture dictates its pricing flexibility.

The Contractual Mechanics: SLAs and the Clawback Clause

As outcome pricing enters production contracts, the legal and financial terms governing it have become intensely negotiated. In traditional SaaS contracts, service-level agreements covered uptime, latency, and support response times. In outcome-priced AI contracts, SLAs govern resolution definitions, false-positive dispute procedures, and minimum accuracy thresholds.

The most consequential contractual mechanism to emerge in early 2026 is the clawback clause — sometimes framed as an outcome true-up or resolution dispute reserve. Under standard language appearing in CX and ITSM contracts, a customer has a specified window (typically 14 to 30 days) to audit and dispute automated resolutions. If a conversation classified as resolved by an AI agent resulted in a reopened ticket, an escalated complaint, or a customer churn event within that window, the fee for that resolution is credited back against the customer’s next invoice.

Enterprise procurement attorneys are also negotiating tier-adjusting performance floors: if an AI agent’s verified first-contact resolution rate falls below an agreed baseline — commonly 55% to 65% in production pilots — the per-outcome price drops automatically for the subsequent billing period. That structure transfers underperformance risk back to the vendor, reversing the dynamic that made early per-conversation pricing so unpopular with finance teams.

What Enterprise Buyers Should Be Negotiating Now

The Bessemer Venture Partners 2026 AI Pricing and Monetization Playbook tracks a decisive shift: per-seat as a share of SaaS vendor pricing has fallen from 21% to 15% in twelve months, while hybrid models (base plus usage or outcome overage) have risen from 27% to 41%. A Futurum 1H 2026 survey found 43% of enterprise buyers prefer consumption-based models; 27% favor outcome-based structures. Gartner’s longer-range projection: by 2030, at least 40% of enterprise SaaS spend will have migrated to usage-, agent-, or outcome-based pricing.

For enterprise buyers currently in AI contract negotiations or approaching renewals, four practices will be consequential. First, demand definitional specificity: the phrase “per resolution” has no universal meaning across vendors, and any contract with outcome-based pricing should include a precise, agreed-upon definition with examples.

Second, negotiate performance floors: if the AI resolves fewer than an agreed percentage of interactions, the per-outcome price or contract terms should adjust automatically, not at the vendor’s discretion. Third, preserve audit rights: the gap between what the vendor’s system classifies as a resolution and what the customer considers resolved can widen as AI systems grow more complex. Fourth, watch the hybrid trap: credit-based models such as Salesforce’s Flex Credits create committed spend regardless of how many outcomes are actually delivered — buyers should calculate effective per-outcome costs under realistic resolution-rate assumptions, not vendor best-case projections.

Software Pricing Converges on the Labor Market

Research in PitchBook analyst reports on enterprise software frames the pricing evolution in its broadest context: enterprise software is converging with the global knowledge-worker labor market. Vendors are no longer selling tool access — they are selling completed work. The price of a successfully resolved support ticket, a correctly classified regulatory document, or a closed sales opportunity is beginning to be set by the same forces that set the price of outsourced labor: competitive supply of AI-capable vendors, measurable output quality, and buyer leverage on volume.

That framing has disruptive implications for the existing SaaS pricing stack. A market where you pay $0.99 per resolved ticket has a natural ceiling — it competes directly with the loaded cost of a human support agent handling the same ticket. When AI pricing converges on labor economics, the enterprise software vendors that win will be those whose AI actually delivers the outcomes they are billing for, at prices below the human alternative.

The third generation of enterprise AI pricing has arrived. The question is no longer whether enterprise AI pricing will be based on outcomes. It is whether enterprise buyers are ready to negotiate contracts that enforce them.

Sarah Chen covers enterprise technology and the economics of AI adoption for NextWave Insight.

This article was produced with AI assistance and reviewed by the editorial team.

Sarah Chen

Sarah Chen

Sarah Chen covers the economics of the technology industry for Next Waves Insight — venture funding, enterprise AI adoption, and the business models emerging around frontier technology. Her work traces capital flows and corporate strategy to explain not just what companies are building, but why, and what it signals about where the industry is heading. She has a background in technology and financial journalism.

Leave a Comment