Usage-based billing means charging customers for what they actually used, such as the number of messages sent, the minutes of audio processed or the gigabytes stored, instead of a fixed fee for each user. It is also called metered billing or pay-as-you-go pricing. It is a part of subscription billing.
With a fixed fee, billing is a calculation done once a month: the number of users multiplied by the price. With usage, the bill depends on thousands or millions of small events, each recorded as it happens. So billing becomes a calculation that runs all the time.
The software works in four steps. First comes metering: the product sends a record of each use, called a usage event, to the billing engine. Second comes aggregation: the engine adds up the events for each customer, for example the total minutes in the month. Third comes rating: the engine applies the price plan, with any free allowance, price tiers and minimum fee. Fourth comes invoicing: the engine produces the bill. The customer can see the running total at any time during the month.
Imagine a speech-to-text company in Bengaluru that turns recorded customer calls into text for call centres. Its plan has a base fee of ₹20,000 a month, which includes 50,000 minutes. Each further minute costs ₹0.40. One call centre processes 3,40,000 minutes in August, and each processed call sends a usage event. At month-end, the engine adds them up: 3,40,000 minutes, of which 2,90,000 are above the allowance. It charges ₹20,000, plus 2,90,000 × ₹0.40 = ₹1,16,000, a total of ₹1,36,000 before GST. On 20 August, the customer's dashboard had already shown 2,10,000 minutes used, so the bill was no surprise.
AI software made this kind of pricing common. Running an AI model costs the provider money for every request. So a flat fee for each user is risky for the provider, because one heavy user can cost more than the fee. Many AI products now charge per request, per unit of text processed or per task completed. The page on SaaS pricing models compares charging per seat, by usage and by outcome.
The idea is older than software. Telecom operators have always charged by minutes and data, through telecom billing and real-time charging. Printer makers charge per page under managed print services (MPS). Electricity is billed by the unit.
The hard parts are accuracy and trust. If usage events are lost, the company loses money. If they are counted twice, the customer is overcharged. So the engine removes duplicate events, and it keeps every event, so that a customer can check the bill. Billing providers are now also adding AI models that recommend the price, and that predict which customers are likely to cancel.