Mastering Usage-Based Pricing for AI-era SaaS
Why AI is pushing SaaS toward usage-based pricing, what you need in place before you launch, and three mistakes to avoid.
Why don't seats alone work for AI-era SaaS?
Per-seat pricing assumed that value grew with headcount. AI breaks that link. One person with a good AI feature can now do the work of five, so customers get more out of your product without buying more seats.
At the same time, every prompt and agent run costs you compute. Your heaviest users cost the most to serve but pay the same seat price as everyone else, so margin shrinks on exactly the accounts using you most. And because seats follow headcount, a customer that cuts staff cuts your revenue even if their usage stays flat.
- Value per seat rises with AI, but the seat price doesn't
- Compute costs grow with every request, so heavy users squeeze margin
- Seat revenue drops when customers cut headcount, even if usage holds
Why do B2B products need seats and usage together?
Most B2B software should keep seats and add usage. Seats still control who has access, and they're how buyers plan budgets. What's new is that every seat, and every AI agent, now consumes something.
So the model Jon sees most often is a subscription with seats that draw from one shared pool of credits. The account gets a predictable base, and usage above it grows with value. Pricing like credit burndown is one common way to run that pool.
- Seats: access control and the budget line buyers recognize
- Shared credit pool: what every user and AI agent draws from
- Add-ons and top-ups: how the pool grows when the team needs more
How do you pick a value metric for usage-based pricing?
Your value metric is the unit you charge for. For an AI product it's usually tokens or credits; for other products it might be API calls, documents processed, or reports generated.
A good one passes three tests. If it fails the third, finance teams can't estimate the bill, and deals slow down. For most products Jon recommends starting with a subscription that includes an allowance, plus overage or credit top-ups, so buyers know their minimum and revenue still grows with use.
- It grows as the customer gets more value
- You can measure it accurately, event by event
- The buyer can predict it before they sign
What do you need in place before you launch usage-based pricing?
Changing a pricing page is easy. Running usage-based pricing reliably takes six pieces working together, from the moment a customer does something billable to the invoice they receive.
- Metering: every billable action becomes an event, counted exactly once
- Rating: the rules that turn usage into money, including rates, allowances, resets, and rollover
- Limits: one for the shared account pool and one per user, so a single person can't burn the team's allowance
- Real-time enforcement: an instant answer to whether this user or agent can do this right now
- Visibility: in-product usage views and alerts before a customer hits a limit
- Billing handoff: rated usage flows into the billing platform you already use
Why does real-time enforcement matter so much?
Billing tools can count usage after the fact, but they can't stop it as it happens. If your product can't check entitlements the instant a request comes in, customers blow past their limits before anyone notices.
That leaves three bad options: absorb the cost, write off the revenue, or send a surprise invoice and start a dispute. Enforcing limits at request time, and keeping metering, rating, and enforcement in one place, keeps what the customer sees in your product in step with what they're billed.
What are the most common usage-based pricing mistakes?
Three mistakes come up again and again after launch, and each has a simple fix.
- Charging for too many things: if buyers have to watch four meters, they use your product less, so start with one metric or one credit balance
- Moving every customer at once: start with new customers, then those who opt in, and move enterprise accounts at renewal
- Hard-coding prices and limits: your first pricing is a best guess and AI costs keep changing, so keep pricing out of your code and avoid an engineering release for every change
Video transcript
Auto-generated from the video and lightly edited for readability.
AI has changed the economics of SaaS. Every prompt and every agent run costs you money to deliver. So more software companies are asking: should we charge for usage?
I'm Jon Gillespie-Brown, CEO of Nalpeiron. We've helped software companies price and package what they sell for twenty years. In the next few minutes I'll cover why usage-based pricing is taking off, what you need in place before you launch it, and three mistakes to avoid.
For years, B2B SaaS charged per seat. That model is under pressure.
With AI, one person can do the work of five, so customers get more value without buying more seats. Meanwhile your compute costs rise with every request. Your heaviest users cost you the most, but they pay the same as everyone else.
With usage-based pricing, revenue grows as customers get more value, and your price keeps pace with your costs.
But most B2B software needs seats and usage together. Seats still control access, and they're how buyers budget. What's new is that every seat, and every AI agent, now consumes something. So the model we see most is a subscription with seats, drawing from a shared pool of credits.
Start by deciding what you charge for. For an AI product that's usually tokens or credits.
A good value metric grows as the customer gets more value. You can measure it accurately. And the buyer can predict it. If their finance team can't estimate the bill, the deal slows down.
For most products, I'd start with a subscription that includes an allowance, plus overage or credit top-ups. Buyers know their minimum, and your revenue still grows with use.
Changing your pricing page is easy. Running usage-based pricing reliably is not.
First, metering. Every billable action becomes an event, counted exactly once. Lose one and you give away revenue. Count one twice and you send a wrong invoice.
Second, rating. The rules that turn usage into money: rates, allowances, resets, and whether unused credits roll over.
Third, limits. In B2B you need them at two levels: the shared pool for the account, and per user, so one person can't burn the whole team's allowance.
Fourth, real-time enforcement. When a request comes in, your product must answer one question instantly: can this user, or this agent, do this right now? If it can't control this, customers blow past their limits, and you either eat the cost or send a surprise invoice.
Fifth, visibility. Show customers their usage inside your product, and alert them before they hit a limit. The invoice should never be a surprise.
And last, the handoff to billing. Rated usage should flow into the billing platform you already use, so the invoice matches exactly what the customer saw.
One. Charging for too many things. If buyers have to watch four meters, they'll use your product less. Start with one metric or one credit balance.
Two. Moving every customer at once. Start with new customers, then those who opt in, and move enterprise accounts at renewal.
Three. Hard-coding prices and limits into your product. Your first pricing is a best guess, and AI costs keep changing. If every price change needs an engineering release, you'll always be slow.
That's what we built Zenmeter for. It meters every event, rates it against your plans, and enforces seats, shared pools and per-user limits in real time, for people and AI agents. Then it hands rated usage to your billing platform, so what customers see always matches what they're billed.
To see your own pricing model running in Zenmeter, book a demo. The link is below.
Thanks for watching.
Decouple your
monetization today
Join the enterprises scaling their revenue without rebuilding their stack every year.
Unleash the monetization potential of your software / SaaS / Hardware