All articles

Metrics & Pricing

SaaS pricing models: per-seat, usage-based, tiered and freemium compared

Per-seat, usage-based, tiered and freemium pricing compared on the same fictional product, with the numbers at 10, 100 and 1,000 customers.

By Published

Isometric 3D illustration of a round acid lime green price tag hanging from a black cord above a stack of four grey discs, on a pale grey background.

You are about to put a price on a SaaS product, and every founder forum says something different: charge per user, no, charge per usage, no, give it away and upsell. The answer is that there are four models, each behaves differently as customers grow, and the right one depends on how your customers get value from the product. This guide prices one fictional product four ways so you can see the behavior in numbers.

The models are the same ones the 30 products in SaaS examples by category use, and the subscription mechanics behind them are explained in What is SaaS?.

The four models in plain terms

You already pay for things in all four ways. A gym membership is a flat price per person per month, whether you go twice or twenty times: that is per-seat pricing. A taxi meter charges by the distance you travel: that is usage-based pricing. A phone plan offers small, medium and large bundles, and you pick the bundle that fits: that is tiered pricing. A free app that nags you to upgrade for the good features is freemium.

In SaaS terms:

  • Per-seat (seat-based) pricing: a fixed price per user per month or year. Slack and Notion charge this way.
  • Usage-based pricing: a price per unit consumed, such as a message, a gigabyte or an API call. Twilio and Datadog charge this way.
  • Tiered pricing: two to four fixed plans with rising limits and features, often per company rather than per user. QuickBooks Online and Xero charge this way.
  • Freemium: a permanent free plan plus paid plans. Not a free trial, which ends after a set time. GitHub, Trello and Zoom have one.

Most real products mix two of them: a per-seat plan with usage charges above an included allowance, or a tiered plan with a free entry tier. The mixes are easier to judge once you can see each pure model on its own.

The same product, priced four ways

Example: a fictional product, Ledgerly, helps small teams track expenses. Round numbers throughout. The assumptions are the same for every model so that only the pricing changes:

  • The average customer has 5 users.
  • The average customer records 600,000 expense events a month.
  • Customers come in three sizes: 60% small, 30% medium, 10% large.
  • For freemium, “customers” means sign-ups, and 4% of sign-ups upgrade to a paid plan.

The four price lists:

  • Per-seat: $12 per user per month. Average customer: 5 x $12 = $60.
  • Usage-based: $0.10 per 1,000 events. Average customer: 600 x $0.10 = $60.
  • Tiered: Small $29, Medium $79, Large $199 per month. Weighted average: 0.6 x $29 + 0.3 x $79 + 0.1 x $199 = $17.40 + $23.70 + $19.90 = $61.
  • Freemium: free plan, then the per-seat price of $12 per user. Per sign-up: 4% x $60 = $2.40.
Monthly revenue 10 customers 100 customers 1,000 customers
Per-seat ($60 per customer) $600 $6,000 $60,000
Usage-based ($60 per customer) $600 $6,000 $60,000
Tiered ($61 per customer) $610 $6,100 $61,000
Freemium ($2.40 per sign-up) $24 $240 $2,400

The first three rows are set to land in the same place on purpose. The point is not which earns more at the average; it is what happens when customers stop being average.

When a customer grows. Under per-seat pricing, a customer that goes from 5 to 15 users triples its bill. Under usage-based pricing, a customer that triples its events triples its bill, even with the same 5 users. Under tiered pricing, that customer pays nothing extra until it hits the next tier, then jumps from $79 to $199. Under freemium, growth inside the free plan earns nothing until the upgrade trigger fires.

When a customer shrinks. Per-seat revenue falls with layoffs. Usage revenue falls with a quiet quarter. Tiered revenue holds until the customer downgrades, which customers do more slowly than they cut seats. That stickiness is why finance teams like tiers and why customers resent them.

What freemium needs. With a 4% conversion rate, 1,000 sign-ups earn what 40 paying customers earn. Freemium only works when sign-ups are cheap to acquire and cheap to serve, and when the free plan has a wall that paying users hit on their own. Without that wall, the free users stay free. Change the conversion rate to 10% in the table and freemium earns $6,000 at 1,000 sign-ups, the same as per-seat at 100 customers. The model is a bet on volume.

How to choose

The rule that survives contact with real products: charge for the thing that grows when your customer gets more value.

  • If value grows with the number of people using the product, per-seat fits. Collaboration tools, CRMs, design tools.
  • If value grows with volume that is independent of headcount, usage fits. Messaging APIs, payments, hosting, monitoring. Twilio prices an outbound US text message at $0.0083 per segment (Twilio pricing, October 2026) because a customer with two developers can send ten million of them.
  • If customers fall into a few recognizable sizes and dislike variable bills, tiers fit. Accounting software, e-commerce platforms.
  • If the product is easy to try alone and spreads inside a company, freemium fits as the entry point, with per-seat pricing above it.

Then test the price. Two numbers tell you whether the model is working: what share of customers upgrade or expand within a year, and how many complain about the bill being unpredictable. The second number is the usage model’s weak point.

AI features are pushing vendors away from pure seats

AI assistants changed one assumption behind per-seat pricing: that value scales with people. An AI agent that answers support tickets does work instead of a person, so vendors have started charging for the work. Three dated, sourced examples:

  • Salesforce moved Agentforce from $2 per conversation to “Flex Credits” metered per action. “Each action costs $0.10 and you only pay for what you use,” the company wrote on its blog on May 23, 2025. Its example: an exchange that would have cost $2 under conversation billing runs 3 to 6 actions, or $0.30 to $0.60.
  • Intercom sells per-seat plans (Essential $29, Advanced $85, Expert $132 per seat per month) and prices its Fin AI Agent separately, “from $0.99 per Fin outcome”, counted when the customer confirms resolution, stops asking, or Fin completes a workflow (Intercom pricing, October 2026).
  • Zendesk includes a number of AI “automated resolutions” per agent in its suites and sells additional resolutions on committed and pay-as-you-go rates, priced separately from the seat (Zendesk pricing, October 2026).

The pattern is a hybrid: seats for the humans, usage or outcomes for the AI. If your product has an AI feature that does work on the customer’s behalf, expect the same question from your customers: why pay per person for something that replaces a person?

Common mistakes

  1. Per-seat pricing on a product that punishes adding seats. If inviting a colleague to a shared budget costs $12 more a month, customers stop inviting colleagues, and the product never spreads. Why it happens: per-seat is the default everyone copies. Instead: make viewers or light users free (Airtable charges only editors; Vercel’s Pro plan includes unlimited viewer seats) or charge by the thing that is used, not by the people who look at it.
  2. Usage-based pricing with bills the customer cannot predict. A finance team that cannot forecast a bill will cap it, or cancel. Why it happens: pure metering looks fair to the vendor. Instead: sell committed usage with a discount, publish a calculator, and alert customers before they cross a threshold. Datadog’s price list shows the shape: a lower rate billed annually ($15 per host per month) and a higher rate on demand ($18), per its pricing page in October 2026.
  3. Freemium with no upgrade trigger. A free plan that does everything a small team needs is a charity, not a funnel. Why it happens: founders fear scaring users away. Instead: pick one limit that a serious user will hit (history, members, volume) and make the upgrade path obvious. Slack’s free plan keeps 90 days of message history (Slack pricing, October 2026); the wall is the archive.
  4. Pricing from your costs instead of the customer’s value. Cost tells you the floor, not the price. Why it happens: costs are the only number a new company knows for sure. Instead: find what the customer saves or earns, and price a fraction of it. Then check the floor.

Sources

In three sentences

SaaS products charge per seat, per unit of usage or per plan tier, often with a free entry plan, and the right model is the one that charges for whatever grows when the customer gets more value. On an average customer the three paid models earn about the same, so the difference shows up only when customers grow, shrink or stay free. AI features that do work on the customer’s behalf are pushing vendors toward hybrids, with seats for people and metered prices for the AI.