Usage Billing Review

Guide

SaaS pricing models, and what each one asks of your billing system

Flat-rate, per-seat, feature tiers, usage-based, tiered and volume rates, prepaid credits, hybrid and outcome-based pricing: how each one charges, where it fits, and the billing work it creates.

By Usage Billing Review Editors · Updated

A SaaS pricing model is the rule that turns what a customer does with your product into what they owe. Most lists of pricing models stop at the label. The label is the easy part: the harder question is what each model makes your billing system count, store and explain on the invoice, because that is where a model that looked right in a pricing meeting either works or quietly leaks revenue.

This guide covers the eight models SaaS and AI products actually ship, in order from the simplest to bill to the hardest. For each one: what the customer pays for, where it fits, where it breaks, and the billing work it creates. Most products end up combining two or three of them, so the last sections cover hybrids and how to choose.

The eight models at a glance

ModelCustomer pays forRevenue predictabilityWhat billing has to handle
Flat-rateAccess to the product, one priceHighA recurring charge and renewals
Per-seatEach user or licenceHighSeat counts, proration on mid-cycle changes
Feature tiersA package of features and limitsHighPlan entitlements, upgrades and downgrades
Usage-basedUnits consumed: calls, tokens, GB, minutesLowEvent metering, aggregation, rating each period
Tiered and volume usageUnits, at a rate that changes with volumeLowBreakpoints, per-tier or whole-volume rating
Prepaid creditsA balance drawn down by usageMediumWallets, real-time balance checks, expiry, top-ups
HybridA base fee plus usage beyond what is includedMediumBoth subscription and metering on one invoice
Outcome-basedA result: a resolved ticket, a completed taskLowDefining and verifying the outcome, disputes

Flat-rate pricing

One product, one price, every billing period.

Every customer pays the same amount for the same product, monthly or yearly. It is the easiest model to sell and to bill: there is nothing to count, so the invoice is the same every period.

  • Fits when: Customers get roughly the same value from the product and serving a heavy customer costs about the same as serving a light one.
  • Breaks when: Usage, and the cost of serving it, varies widely between customers. Light users subsidise heavy ones, and there is no way to charge more as an account grows.
  • Billing work: A recurring charge, renewal dates, failed-payment retries and annual-versus-monthly terms.

Per-seat (per-user) pricing

The price scales with the number of people using the product.

The customer pays a fixed price for each user or licence. Revenue grows as the customer adds people, which made per-seat the default for collaboration and workflow software. Some vendors charge only for active users rather than provisioned ones, which is fairer but means the bill has to know who was active.

Per-seat strains when the value a product delivers stops tracking headcount. An AI feature that does work on a user's behalf can deliver more value with fewer seats, and the seat price starts to cap revenue instead of growing it.

  • Fits when: Value grows with the number of people who use the product, and each user's cost to serve is similar.
  • Breaks when: Automation, agents or API access let a few seats do the work of many, or one seat's compute cost dwarfs another's.
  • Billing work: Seat counts, true-ups when seats are added mid-term, proration on plan changes, and active-user rules if you charge only for active seats.

Feature-tiered pricing

Good, better, best: packages with more features and higher limits at each step.

The customer picks a package, and each step up unlocks features, higher limits or support levels. Tiers let one product serve a small team and an enterprise at different prices, and the top tier is often a custom contract rather than a list price.

Tiers are packaging, not a charge metric, so they usually sit on top of another model: per-seat with feature tiers, or a platform fee per tier with usage on top. A free tier at the bottom is how freemium works.

  • Fits when: Customer segments want clearly different things, such as SSO, audit logs or higher limits for larger accounts.
  • Breaks when: The limits between tiers become the real price. Customers who outgrow a limit mid-month need an overage or an upgrade path, or they hit a wall.
  • Billing work: Entitlements per plan, enforcement of limits, upgrade and downgrade timing, and versioning plans without moving existing customers by accident.

Usage-based (pay-as-you-go) pricing

The customer pays for what they consume, measured in a unit you define.

Also called metered or consumption-based pricing. The charge follows a usage unit: API calls, tokens, GB stored, compute minutes, messages sent. Cloud infrastructure and LLM APIs are the familiar examples: the large model APIs bill by input and output tokens, priced per model.

The appeal is that price tracks both value and cost to serve, so a customer can start small and grow without renegotiating. The cost is predictability: the customer cannot always forecast the bill, and you cannot always forecast revenue.

Usage-based pricing is also the first model on this list where billing becomes an engineering system. Every billable action has to be captured as an event, deduplicated, attributed to the right customer, aggregated and rated before an invoice can exist, and a gap anywhere in that chain is revenue that never gets billed.

  • Fits when: Cost to serve varies with consumption, usage is a fair proxy for value, and customers want to start without a commitment.
  • Breaks when: The usage unit is something customers cannot predict or relate to value, or the metering behind it is not accurate enough to defend on an invoice.
  • Billing work: Event ingestion, idempotency, late-arriving events, aggregation, rating, usage visible to the customer mid-period, and spend alerts.

Tiered and volume usage pricing

Usage pricing where the per-unit rate changes as consumption grows.

Most usage pricing is not one flat rate per unit. With graduated (tiered) pricing, each band of usage is charged at its own rate: the first block of units at one price, the next block at a lower one. With volume pricing, the band the total lands in sets one rate for every unit. The two produce different invoices from the same usage, so the rate card has to say which one applies.

Package pricing, where units are sold in blocks (per thousand calls, per million tokens), is a variant that rounds usage up to the next block.

  • Fits when: You want to reward larger customers with lower unit prices without a separate negotiated contract.
  • Breaks when: Volume pricing creates cliffs, where crossing a breakpoint makes the whole bill cheaper, and customers time their usage around it.
  • Billing work: Breakpoints, graduated versus whole-volume rating, rounding rules, and invoices that show customers how their total was reached.

Prepaid credit pricing

The customer buys a balance up front and usage draws it down.

Instead of paying for usage after the fact, the customer buys credits, and each action consumes some of them. Credits put a single currency over many different usage units, which is why AI products that bill for several models, tools or agent steps often use them: the customer budgets in credits while each action burns credits at its own rate. Data warehouses have long billed compute this way.

Credits move risk toward predictability for both sides: the customer has a spending cap, and you are paid before you serve. They also bring their own accounting, since prepaid credits are a liability until they are used or expire.

  • Fits when: Usage spans several units with different costs, customers want a hard budget, or you want cash up front.
  • Breaks when: Balance checks are slow or stale, so customers overdraw, or expiry rules feel punitive and push customers to leave.
  • Billing work: A wallet per customer, burn rates per action, real-time balance checks, top-ups, expiry, and revenue recognised as credits are consumed.

Hybrid pricing

A predictable base fee plus usage charges on top.

Hybrid models combine a recurring charge with metered usage. The common shapes are a platform fee with an included allowance and overage beyond it; a minimum commitment that usage is billed against, with a true-up at the end of the term; and seats for people plus usage for what automation consumes.

Hybrid is where many usage-based products end up, because it gives sales a number to forecast and still lets revenue grow with consumption. It is also the hardest model to run on a billing tool built for one side only, since one invoice has to combine the subscription and the usage correctly.

  • Fits when: Customers want a predictable floor, you want revenue to scale with heavy use, or enterprise buyers ask for committed spend.
  • Breaks when: The included allowance, the overage rate and the commitment are set without usage data, so most customers sit far below or far above them.
  • Billing work: Allowances, overage rating, commitment drawdown, true-ups, contract-specific overrides, and revenue recognition across fixed and variable parts.

Outcome-based pricing

The customer pays when the product delivers a defined result.

Outcome-based pricing charges for a result rather than for the work behind it: a resolved support conversation, a qualified lead, a completed task. It has gained ground with AI agents, because an agent's work is often closer to a finished job than to a seat or an API call.

It is the model most closely tied to value, and the hardest to bill. Both sides have to agree what counts as an outcome, the product has to record it in a way the customer can check, and you carry the cost of every attempt that does not succeed.

  • Fits when: The outcome is discrete, easy to verify and clearly valuable to the buyer, and your success rate is high enough to absorb failed attempts.
  • Breaks when: Outcomes are disputed or partly the customer's doing, or the cost of reaching an outcome varies too widely to price.
  • Billing work: Outcome definitions in the contract, event-level evidence for each billed outcome, dispute handling, and margin tracking per outcome.

How to choose a SaaS pricing model

Start from what drives your cost and your customer's value, then pick what billing can support.

  1. Find the value metric. Ask what grows when a customer gets more out of the product: people, records, calls, tokens, finished tasks. Charge on that if you can measure it reliably.
  2. Check your cost to serve. If one heavy customer can cost many times what a light one does, a flat or per-seat price will lose money on the heavy end. Some form of usage pricing protects margin.
  3. Decide who carries the risk of variable bills. Pay-as-you-go puts it on the customer; commitments, allowances and credits share it. Enterprise buyers usually want a number they can budget.
  4. Be honest about your metering. Usage, credit and outcome pricing are only as good as the events behind them. If you cannot count a unit accurately and show it to the customer, do not bill on it yet.
  5. Plan for change. Pricing is rarely right the first time. Pick a billing setup where rates, tiers and plans can be versioned and changed without moving existing customers by accident.
  6. Check the accounting. Prepaid credits, commitments and usage each change when revenue is recognised. Bring finance in before launch, not at the first month-end close.

Where to check how other products price

Real pricing changes often, and many SaaS and AI vendors have reworked their models in the past two years. To see how a specific product prices today, read its public pricing page, its pricing FAQ or documentation (where usage units and overage rules are usually defined) and its changelog or announcements for recent changes. For enterprise plans, the list price is often a starting point for a negotiated contract that is not published.

Questions people ask

What are the main SaaS pricing models?

Flat-rate, per-seat, feature-tiered, usage-based, tiered and volume usage pricing, prepaid credits, hybrid (a base fee plus usage) and outcome-based pricing. Freemium is a free tier placed in front of one of these rather than a model of its own.

What is the difference between usage-based and consumption-based pricing?

In practice, nothing: both mean the customer pays for what they use, measured in a defined unit. Pay-as-you-go and metered pricing are other names for the same model.

Is per-seat pricing going away?

Not entirely, but it fits less well when AI features or agents let fewer people do more work. Many products keep seats for human users and add usage or credits for what automation consumes, which is a hybrid model.

Which pricing model is best for AI products?

Most AI products need some usage component, because the cost of serving a request varies with model, input size and output size. Credits or a hybrid of a base fee plus usage are common because they give customers a budget while keeping price tied to cost.

What is the difference between tiered and volume pricing?

With tiered (graduated) pricing, each band of usage is charged at its own rate. With volume pricing, the band the total falls in sets one rate for all units. The same usage produces a different bill under each.