Per-Token Pricing Is Dead. The Outcome Tax Is How AI Companies Actually Charge in 2026.
Two years of per-token billing produced unpredictable customer invoices and razor-thin SaaS margins. The 2026 pricing reset is moving the AI category onto outcome-based models — and changing which companies survive the transition.
By Maya Lin Chen, Product & Strategy · May 20, 2026
Per-token AI pricing is collapsing under unpredictable invoices and thin margins. The 'outcome tax' replaces it in 2026 — analysis of the pricing reset reshaping AI SaaS economics.
Frequently Asked Questions
What is the 'outcome tax' pricing model?
The outcome tax is an emerging AI SaaS pricing model that charges customers based on measurable business outcomes the AI delivers, rather than on the underlying input or token consumption. Examples include charging per resolved customer support ticket (Intercom Fin), per closed sales deal where the AI contributed material work (HubSpot Breeze, Outreach), per shipped pull request (Cursor and several enterprise AI coding products), and per agent-completed task (Sierra, Decagon, multiple vertical AI agents). The 'tax' framing reflects that the customer pays a defined percentage of the value the AI generates rather than a usage-based meter that may or may not produce value. For the customer, the outcome tax is more predictable than per-token pricing and aligns vendor success with customer success. For the vendor, the outcome tax shifts margin from input costs to output value, which is generally a healthier economic position as inference costs continue to fall.
Why is per-token pricing failing in 2026?
Per-token pricing has produced three structural failures in 2026. First, customer-side unpredictability: enterprise customers cannot budget against AI features that produce variable monthly invoices, and procurement teams have started rejecting purchases of AI products that cannot offer fixed-cost or capped-cost models. Second, vendor-side margin compression: per-token resale models generate gross margins of 30-50% versus the 70-85% margins of traditional SaaS, which makes it nearly impossible to build venture-scale AI businesses on a pure resale model. Third, customer trust erosion: per-token billing has produced multiple public incidents of surprise invoices ranging from 5x to 50x expected costs, which has damaged the broader category's trust position with enterprise buyers. The pricing model that defined 2023-2025 AI commercialization has reached the end of its useful life as the primary commercial wrapper for AI features.
Which AI companies have already moved to outcome-based pricing?
By Q1 2026, an estimated 30-40% of enterprise AI SaaS companies had shifted their primary pricing motion to some form of outcome-based or hybrid outcome-and-usage model. Specific named examples include Intercom Fin charging per resolved customer support conversation, Cursor charging enterprise tiers per shipped feature or per pull request in certain plans, Sierra AI charging per autonomous customer agent interaction with defined success criteria, Harvey charging legal services firms per matter completed rather than per query, and Abridge charging healthcare providers per clinical note finalized. HubSpot's Breeze AI launched in 2024 with a hybrid credit and outcome model and shifted toward more outcome-weighted pricing through 2025. Among vertical AI agents, the outcome-based pricing transition has been most rapid because vertical agents have clearer measurable outcomes than horizontal agents.
How do you actually structure outcome-based AI pricing?
Five design decisions drive outcome-based AI pricing implementation. First, define the outcome precisely with measurable success criteria — 'resolved support ticket' must be defined by specific resolution signals (customer-confirmed resolution, no follow-up within X days, sentiment threshold), not by AI confidence. Second, set the price per outcome relative to customer alternatives — what does the same outcome cost without AI, and what percentage of that value can you capture. Third, build telemetry that observes the outcome rather than relying on AI self-reporting, which creates the trust foundation customers need to accept outcome billing. Fourth, design fail-safes for AI errors — what happens when the AI reports an outcome that was not actually achieved, and how do you handle the customer experience. Fifth, structure the contract to mix outcome billing with platform fees so the vendor has predictable revenue even when individual outcome volumes vary. The companies that get all five right will define the next decade of AI commercialization.
Does outcome-based pricing kill the open-source AI ecosystem?
Outcome-based pricing has a more complex relationship with open-source AI than per-token pricing did. In a per-token world, open-source models compete directly with closed models on cost-per-token, which favors open-source for cost-sensitive deployments. In an outcome-based world, the customer pays for measurable business outcomes regardless of which underlying model produces them, which decouples the model layer from the commercial layer. The implication is that open-source models can become infrastructure for outcome-priced products without competing for customer dollars directly. Companies like Sierra, Cursor, and Harvey can use whichever underlying model produces the best outcome per dollar — open-source or proprietary — without changing their customer-facing pricing. This is structurally favorable to open-source models because it removes the per-token commercial competition while preserving the technical contribution. The open-source AI ecosystem may actually expand under outcome-based commercial models, even as the per-token resale economics that previously favored it weaken.
Related Articles
Topics: Pricing Strategy, AI, SaaS, Product Management, Usage-Based Pricing
Browse all articles | About Signal