Seat-based pricing charges per person with access, so the bill tracks headcount. Usage-based pricing charges for what customers consume (messages, API calls, tasks, transactions), so the bill tracks activity. Seats give predictable revenue and simple buying; usage matches price to value and cost. Choose seats when value grows with team size, usage when value grows with volume, and a hybrid when both matter.
Key takeaways
- Seats track headcount; usage tracks activity. Pick the one your customer's value follows.
- Seat pricing can leave heavy users underwater on margin when each action costs you money.
- Usage pricing can undercharge big teams that use the product lightly.
- A seat fee with an included allowance and overage is now the most common answer for AI products.
Procore sells construction project management software. When I was writing The Subscription Playbook, the line on its pricing page stopped me cold:
"Explore unlimited plans for every builder and budget."
Unlimited users? For a fixed fee?
If I ran a building company, I'd love that. I'd also assume there was a catch. Nobody gives away unlimited anything. (Ask anyone who's ever been on an "all you can eat" buffet with a teenage boy.)
There was a catch, and it was a clever one. Procore doesn't charge per user at all. It charges an annual fee based on how much construction you do each year.
That's the whole seat-versus-usage debate in one pricing page. Do you charge for the people, or for the work?
What is the difference between seat-based and usage-based pricing?
Seat-based pricing charges for each person with access. Ten users, ten fees. The bill follows headcount. (Full guide: seat-based pricing.)
Usage-based pricing charges for how much the customer uses. Messages sent, minutes transcribed, tasks run, transactions processed. The bill follows activity. (Full guide: usage-based pricing.)
Here's how they compare side by side:
| Seat-based | Usage-based | |
|---|---|---|
| Customer pays for | People with access | What they consume |
| Bill grows when | The customer hires | The customer does more |
| Your revenue is | Predictable month to month | Variable, can spike or dip |
| Customer's bill is | Easy to budget | Harder to predict without caps |
| Fairness to light users | Poor: they pay full price | Good: they pay little |
| Margin on heavy users | At risk if each action costs you | Protected: they pay more |
| Easy to explain? | Very | Depends on the metric |
| Encourages adoption across a team? | No: each new user costs more | Yes: adding people is free |
| Exposed to AI shrinking headcount? | Yes | No |
| Typical 2026 examples (US$) | Slack $7.25–$15, Linear $10–$16, Microsoft 365 Copilot $30 per user | Twilio SMS from $0.0083, Deepgram from $0.0043 per minute, Fin $0.99 per outcome |
Neither column is "better." They're built for different value.
How Procore made "unlimited users" pay
Procore's model has lasted (here's the full Procore case study). For full-year 2025, it reported revenue of $1,323 million, up 15%, with gross revenue retention of 95% and net revenue retention of 106%. It had 2,710 customers paying more than $100,000 a year.
Look at why it works.
Construction firms have site managers, subcontractors, architects and clients who all need to see the plans. Charging per seat would make every one of those people a cost, so the builder would ration logins. By charging on construction volume, Procore made adding people free. The software spread through every project, which made it harder to remove. That's land and expand built into the price.
And because the price follows the customer's business, a small builder can afford it and a giant pays more. As I said in the book, small businesses pay less and big corporations pay more.
There's a trade-off. When construction slows, so does Procore's expansion. Usage pricing ties your growth to your customer's industry, in good years and bad.
Worked example: the same two customers on three models
Let's make this real. Say you sell an AI tool that drafts client reports. Each report (a "task") costs you about $0.03 in AI model fees.
You have two customers:
- Customer A: a 5-person agency that runs 8,000 tasks a month. Small team, heavy use.
- Customer B: a 50-person firm that runs 1,500 tasks a month. Big team, light use.
Seats vs usage vs hybrid, in dollars
Seat model ($30 per user): A pays $150 and costs you $240 to serve, a 60% loss. B pays $1,500 and costs you $45, a 97% margin.
Usage model ($0.10 per task): A pays $800 (70% margin). B pays $150 (70% margin).
Hybrid ($15 per user, 100 tasks included per user, then $0.08 per task): A pays $75 plus 7,500 extra tasks at $0.08 = $675 (64% margin). B pays $750 and stays inside its allowance (94% margin).
Total monthly revenue from the pair: seats $1,650, usage $950, hybrid $1,425.
Look what happened. Pure seats made the most money in total, but only because the light user was subsidizing the heavy one. Customer A is losing you money every month and will happily stay forever. Pure usage fixed the margin but gave the 50-person firm a bargain it didn't ask for. The hybrid got close to the seat revenue with healthy margins on both.
Illustrative numbers. Run your own with the Credit Pack Designer if you meter AI actions.
Which is more predictable?
It depends whose predictability you mean.
Seats are predictable for you. You know next month's revenue to within a few percent.
Seats are also predictable for the customer, which is why buyers like them. Decagon offers its AI support agents on both per-conversation and per-resolution pricing and says "the vast majority of our customers choose per-conversation pricing." Salesforce, which now runs several meters at once, saw fixed-price enterprise agreements grow from 16 to more than 120 in a quarter, and one executive said European customers were "afraid of all these metrics."
Usage is the opposite. Revenue moves with customer activity, and the customer can get a nasty surprise. That's fixable: Supabase ships its Pro plan with a spend cap switched on by default, and an included allowance means most months look the same.
So, if you choose usage, sell predictability back to the customer. Caps, alerts, commitments and allowances all do that.
How do sales teams get paid under each model?
This one catches people out.
In the book I used two salespeople, John and Anna. John made twenty-five sales last month. Anna made five. Under seat pricing with similar-sized customers, John is your star.
But Anna's five customers are bigger businesses that will scale their usage over the next few years. Under usage pricing, her deals are worth far more.
If you move to usage, change commission to reward two things: the customer's expected usage at the point of sale, and how that usage grows. Pay only on the logo count and your Annas will leave.
Seats, usage or both?
The free Pricing Model Picker weighs your costs, buyers and value metric and recommends a model.
When should I choose seat-based pricing?
Choose seats when:
- Value grows with the number of people (messaging, collaboration, design, project tracking).
- Each user costs you about the same to serve.
- Buyers need a fixed annual budget.
- Your product helps people work rather than doing the work for them.
When should I choose usage-based pricing?
Choose usage when:
- Value grows with volume (data, messages, transactions, minutes, outcomes).
- Your cost to serve rises with each action, as it does with AI models.
- Usage varies a lot between customers of the same size.
- You want everyone in a customer's business using the product without a per-person toll.
And choose a hybrid when both lists describe you. That's most AI products sold to teams in 2026. Claude Enterprise is $20 per seat plus usage at API rates. ChatGPT Business sells seats plus credits once allowances run out. Zendesk charges per agent plus a fee per automated resolution. GitHub kept its $10, $19 and $39 seat prices and put a credit allowance inside each one. Read the hybrid pricing guide for the mechanics.
Value Metric FinderScore candidate value metrics and pick the one that grows with your customers.Open the free tool →How do I switch from seats to usage without losing customers?
If you're moving, do it the way the companies that kept their customers did it:
- Lead with a better bundle. WiseTech moved about 95% of CargoWise customers to transactional Value Packs within months, and the packs included AI features. Give people a reason to want the new model.
- Grandfather existing customers. Windsurf kept existing subscribers on their old price when it changed plans in 2026.
- Show the bill before it arrives. Publish what typical tasks cost, and offer a calculator.
- Cap by default. No customer should ever be switched off, or surprised by a bill five times the usual, because of a meter.
- Change the sales plan at the same time. See John and Anna above.
What to do next
Take your last 20 customers and write down two numbers for each: how many people use your product, and how much work they do in it. If the two columns move together, seats will serve you well. If they don't, you've found your answer.
Then try the Pricing Model Picker, and if you're still weighing a flat subscription against pure metering, read usage-based vs subscription pricing.
Frequently asked questions
Is usage-based pricing cheaper for customers than per-seat pricing?
For light users, usually yes. For heavy users, often no. Usage pricing spreads the bill according to activity, so a big team that logs in occasionally pays less than it would per seat, while a small team that runs the product hard pays more. That's why it feels fair, and why heavy users sometimes push back when a company switches.
What is consumption-based pricing?
Consumption-based pricing is another name for usage-based pricing. The customer pays for the amount they consume: gigabytes stored, minutes transcribed, messages sent, credits spent. Snowflake prices compute in credits, Twilio per message and Deepgram per audio minute. The term 'consumption' is more common in enterprise contracts that include a prepaid commitment.
Can I charge per seat and per use at the same time?
Yes, and many companies do. Zendesk charges per agent seat and then $1.50 (committed) or $2.00 (pay-as-you-go) per automated resolution beyond an included pool. Claude Enterprise charges $20 per seat plus usage at API rates. The trick is to make one meter the main price and the other a clearly explained extra.
How do I stop customers sharing logins to avoid seat fees?
Make sharing pointless rather than policing it. Offer cheap or free light-user seats, bill only for active users, or move the main charge to something sharing doesn't change, like projects or volume. Netflix's paid sharing shows the other route: give sharers a fair, official add-on price instead of pretending they don't exist.
Which model do investors prefer?
Investors care less about the meter than about net revenue retention and gross margin. Usage models were prized after Snowflake's 158% net dollar retention at IPO, but seat-based Linear reports 177% NRR. A model that lets good customers grow their spend, without wrecking margin on heavy users, is what gets valued.
Get both books free
The Subscription Playbook (ebook + audiobook) and Feed a Starving Crowd, instantly. In Australia? Claim the free hardcover.
Robert Coorey, MBA
Robert is the author of The Subscription Playbook and Feed a Starving Crowd and a Director at BPAG Accountants in Sydney. More about Robert.
Sources
- Procore Q4 and full-year 2025 results (12 Feb 2026)
- Zendesk pricing (checked 27 Oct 2026)
- Claude pricing (checked 27 Oct 2026)
- Decagon: Pricing AI agents
- Pricing Innovation: Agentforce's pricing journey (6 Oct 2026)
- GitHub: Copilot is moving to usage-based billing
- Supabase pricing (checked 27 Oct 2026)
Facts and prices checked 27 October 2026. Spotted something out of date? Tell us.