Two-thirds of the companies on the Forbes AI 50 now use some form of usage-based pricing. At the second annual Monetize conference, Metronome CEO and cofounder Scott Woody characterized how these businesses operate as "alien" relative to traditional SaaS: they rework pricing far more often, favor market share over ARR, and run several go-to-market strategies at the same time. More than 300 leaders from companies including Replit, Fin, Snowflake and Clay gathered to compare notes on pricing, cost management and scaling AI products.
Subscriptions, usage and credits combine
The expectation that usage-based pricing would simply replace subscriptions has not held. Instead, subscriptions increasingly serve as packaging for variable usage, often through credits.
Fin's director of product for monetization, Aisling O'Reilly, traced her company's path from a platform subscription to a hybrid model. Outcome-based pricing arrived with Fin AI Agent, initially charging per support conversation resolved. As the product took on actions such as recommending products, qualifying leads and booking meetings, a single billable metric no longer captured customer value, so Fin added billable outcomes that customers select per conversation type. Customers commit to a pool of credits and draw down against it across outcomes without forecasting each one separately. The resulting layered model uses subscriptions for predictability, usage-based pricing for agentic work and credits to unify multiple units.
Making spend predictable is part of the product
Usage-based billing creates a predictability problem: customers must understand what drives usage, anticipate spend, set limits and connect consumption to value. Companies have answered with in-product dashboards, predictions and related tooling.
Snowflake senior director Ryan Campbell described in-product forecasting, calculators, scenario builders and usage visibility that help customers monitor and control consumption. Stripe's head of global pricing and commercialization, Sam Lee, called the usage control pane—alerts, adjustable spending controls and consumption management—a new standard. Commercial terms add further predictability while keeping charges tied to usage: annual usage buckets, discounted overages, and first-year fixed contracts with follow-on true-ups. As bills grow less predictable, usage controls become part of what customers are paying for and help build trust in the vendor and the product.
Flexible packaging as costs fall and workloads grow
Better AI economics let the same product be packaged in different ways—subscriptions for lower-cost, rate-limited experiences and usage-based billing for resource-intensive agent work. Replit uses both as model costs and agent workloads shift, according to president and head of AI Michele Catasta. When Replit's agents ran for only a few minutes, the company charged 25¢ per agent checkpoint; as runtimes lengthened, some jobs cost Replit several dollars while users still paid 25¢. Usage-based billing closed that gap by tying price to the work performed.
The change brought unpredictable spending and abrupt paywalls for users accustomed to a flat monthly fee. As inference costs fell—partly because capable open-source models pressured frontier-model prices—Replit added a rate-limited Free Mode so users could keep building on a simple subscription. The company now changes elements of its pricing several times a month, passing savings to customers. This supports versions of one product at different speeds, prices and service levels, with the subscription/usage mix adjusted as underlying economics move.
Finance teams measure outcomes, not tokens
Tokens may drive costs, but they rarely represent the value customers believe they are buying. Clay head of finance Karan Parekh said the company avoids "upcharging and taxing tokens," treating model tokens and third-party data as pass-through inputs distinct from the product's differentiated value.
Clay applies the same lens internally: sales tracks productivity, product tracks development time, and support measures throughput and customer satisfaction. Parekh's team accepts AI spending growing faster than revenue only when it produces something distinctive the company could not otherwise achieve. Because multiple tools and workflow changes contribute to the same result, token-level ROI calculations mislead; Clay instead measures AI tooling against function-specific targets reviewed quarterly.
Pricing must be legible to agents as well as people
Agents already influence how products are discovered and evaluated, even before autonomous enterprise purchasing is common. Kyle Poyar, founder of Growth Unhinged, tested how six AI engines presented pricing for Cloud 100 companies across 7,600 generated responses. For one unnamed company, negative pricing language appeared in 88% of responses; without accessible pricing, LLMs turned to informal sources such as Reddit and reported $100–$2,400 per user per month—a 24x spread.
Pricing visible to a human visitor but blocked from crawlers or rendered in ways agents cannot interpret is effectively invisible. Pricing now serves two audiences: the person accountable for the purchase and the machine helping them evaluate it.
Pricing as an adaptive system
AI products, model economics and usage patterns change too quickly for static pricing. The recurring pattern across sessions was to identify what customers value, give them more control over spending, and adjust pricing and packaging as products and unit economics evolve.



