Accounting treats money differently based on its context or 'color'. Cash from a customer only becomes revenue after the business fulfills its obligations. This distinction is crucial for accurately perceiving a business's health and is formalized in the process of revenue recognition, which has three core tests.

AI companies are selling large, seat-based contracts based on hype and experimental budgets, inflating current ARR. Investors are skeptical because, like early SaaS, customers will eventually demand usage-based or outcome-based pricing, challenging the long-term revenue stability of these startups.

On a $2M ARR, Dresma’s largest customer pays $500,000 annually, representing 25% of their total revenue. This validates that a pure usage-based pricing model, without seat-based or feature-gated upsells, can successfully land and expand large enterprise accounts, demonstrating a clear path to significant customer lifetime value.

Unlike typical SaaS where revenue from a monthly subscription is recognized ratably over the month, revenue from pay-as-you-go AI APIs is much simpler. Because the service—token consumption and inference—is delivered almost instantly, the revenue can be recognized as soon as the API call is complete.

Headlines about massive government contracts are misleading. Anduril's $20B deal is not obligated money but a pre-approved spending ceiling. It acts as a 'fast track' by removing initial procurement friction, but revenue is only recognized as individual orders are placed and products are actually delivered.

At scale, a one-size-fits-all pricing model fails. Salesforce CEO Mark Benioff explains that they must offer a mix of seat-based, all-you-can-eat enterprise agreements (ELAs), and consumption-based models. For nearly every significant customer, a custom pricing agreement is crafted to meet their specific needs and circumstances.

Standard SaaS pricing fails for agentic products because high usage becomes a cost center. Avoid the trap of profiting from non-use. Instead, implement a hybrid model with a fixed base and usage-based overages, or, ideally, tie pricing directly to measurable outcomes generated by the AI.

In Snowflake's consumption model, a salesperson's job isn't done at signing. They have separate quotas for bookings (the commitment) and consumption (actual usage). This structure forces them to act as a long-term business partner, ensuring the customer successfully adopts and uses the platform.

Revenue from virtual goods depends on whether they are consumable ('potions') or durable ('swords'). Consumable revenue is recognized upon use. Durable item revenue must be recognized ratably over the item's 'economically useful life,' which is often simplified to the player's expected lifetime with the game.