AS | Ankit Sarawagi|Founder, CFOmatrix·July 2026·11 min read | SaaS Finance |
- Pricing is moving in three stages. Per-seat gives way to consumption (already mainstream) and then to outcome-based (emerging, roughly two to three years from widespread).
- Build consumption price from cost. Per-unit cost of delivery plus a target margin gives the list price; volume discounts come only as committed volume rises.
- The commitment chain protects margin. Customer commits volume, you commit that volume to your compute or LLM vendors, you earn a vendor discount, and you pass part of it back.
- Anchor outcome pricing to their cost-to-serve. If a ticket costs them $10 to resolve, charging $5 to $7 per resolved ticket is an easy value case.
- Bill from a prepaid wallet, not in arrears. Usage draws down credit the customer has already paid, so cash stays ahead of your variable compute cost.
| 3 Stages: seat, consumption, outcome | $5-7 Per resolved ticket vs their $10 cost | 20-30% Annual discount that actually moves buyers |
| 1. Why AI broke per-seat pricing 2. Building a consumption price 3. The vendor-commitment chain | 4. Outcome pricing: anchor to their cost 5. Wallet billing, not arrears 6. Frequently asked questions |
Why AI Broke Per-Seat Pricing
Traditional SaaS is a high-margin business with a simple pricing answer. Your cost to serve one more user is close to zero, so you charge per seat: ten users, ten licences, one predictable bill. It works because cost is fixed per user and your gross margin sits comfortably above 80 percent whether that user logs in once a month or all day.
AI SaaS breaks that logic in one place: cost is no longer fixed per user, it is variable per use. Every request runs compute, often calls a paid model, and hits databases and pipelines. A single “seat” that fires a thousand AI tasks a day costs you far more to serve than a seat that fires ten. Charge both the same per-seat price and you either overprice the light user or lose money on the heavy one. That is why, in the AI SaaS companies I have worked with, pricing steadily moves off seats and onto what the product actually does.
This migration is the single biggest reason AI SaaS finance differs from classic SaaS finance. It links your revenue directly to variable compute cost, which is also why per-customer margin becomes the number to watch. For the full picture of how these pieces fit together, read the SaaS finance pillar guide.
Building a Consumption Price That Protects Margin
Consumption pricing sounds complicated but the build-up is disciplined and simple: start from your cost, add your margin, then discount only for volume. The mistake founders make is pricing off a competitor’s per-seat number or a gut feel, which disconnects the price from the variable cost that AI actually incurs. Do it the other way round.
Because AI gross margins currently run around 65 percent or more, versus 80 percent or more for traditional SaaS, a price set off a competitor’s seat number can silently run at a loss on heavy users. Build the price off your loaded per-unit cost and meter margin per customer so repricing and upsell decisions are grounded in real unit economics, not averages.
To do this well you need to know your true cost of delivery, which means getting cloud COGS and per-customer metering right. Read SaaS gross margin and cloud COGS for what belongs in cost of goods sold, and per-customer and per-unit margin for the metering discipline that makes consumption pricing safe.
The Vendor-Commitment Discount Chain
Here is the mechanism that makes volume discounts sustainable rather than margin-destroying. A discount you give a customer should be funded by a discount you receive, not carved out of your own margin. That is the commitment chain, and it runs in one direction.
The lesson is to never give a volume discount you have not first earned upstream. Enterprise volume commitments are the trigger for the whole chain, which is another reason to push for commitment rather than open-ended usage. That naturally leads to how you bill it.
Outcome Pricing: Anchor to Their $10 Cost
Outcome-based pricing is the frontier. Instead of charging for usage, you charge for the result the product delivers, and you price it against what that result already costs the customer to produce themselves. The anchor is their current cost-to-serve, and it turns pricing into an obvious value case.
The power of this model is that the customer sees the saving directly, and you are paid only when you deliver. The catch, and the reason it is still two to three years from being the default, is measurement: you have to define the outcome cleanly, prove attribution, and both sides must trust the count. Get that right and outcome pricing is the strongest value story in AI SaaS; get it wrong and you argue over every invoice. Until the measurement is airtight, most companies run consumption pricing and pilot outcome pricing on a well-defined unit.
Before you promise outcome pricing, ask whether you can count the outcome without dispute. “Resolved ticket” needs a shared definition of resolved. “Qualified lead” needs an agreed qualification bar. If the unit is fuzzy, stay on consumption pricing and revisit outcome pricing once the data is trustworthy.
Bill From a Wallet, Never in Arrears
You can get the pricing model right and still be undone by how you collect. With variable AI compute, the billing mechanism decides whether cash stays ahead of cost or falls behind it. The rule from operating these businesses is blunt: avoid usage billing in arrears, use a prepaid wallet.
- Customer loads credit upfront; usage draws it down
- You are never out of pocket for compute already paid
- Cash sits ahead of cost, protecting working capital
- Auto top-up keeps service uninterrupted
- Customer runs up usage, you invoice after the fact
- You have already paid the compute cost, waiting to be repaid
- Collection risk and disputes on a large variable bill
- Working capital gap widens as usage grows
Layer commitment on top and the picture is complete. The best case is an annual upfront commitment, which is always good because it boosts runway and improves retention: a locked-in customer stays long enough to see value. Make the annual discount meaningful, 20 to 30 percent, because 5 to 10 percent will not move the decision. For an enterprise on usage, take a volume commitment (no commitment means the highest rate). And where there is no annual commitment at all, put usage on a prepaid wallet that draws down, so you are billing against money already in the bank. Set the plumbing up properly with the SaaS subscription billing stack.
“In AI SaaS the price is not a marketing number, it is a function of your compute cost. Build it off cost, discount only what a vendor first discounts to you, and collect it before you spend it.”
Ankit Sarawagi, CFOmatrixFrequently Asked Questions
How should an AI SaaS company price its product?
AI SaaS pricing migrates along three stages. Traditional SaaS uses per-seat pricing, which works when cost is fixed per user. Because AI compute is variable and grows with usage, pricing moves to consumption-based (charge per unit of work, such as per API call, token or task) and, increasingly, to outcome-based (charge per result delivered, such as per resolved ticket or qualified lead). Consumption pricing is already mainstream; outcome pricing is emerging and likely two to three years from widespread. The right model depends on how directly your cost scales with usage and how measurable the customer’s outcome is.
What is consumption-based pricing?
Consumption-based (or usage-based) pricing charges for the volume of work the product does rather than the number of users. In AI SaaS this is usually per API call, per token, per document, per minute or per task. You build the price from the per-unit cost of delivering that work (compute, LLM calls, data pipelines) plus a target margin, then offer volume discounts as usage grows. It aligns your revenue with both the value and the cost of what the customer actually uses, which matters when AI compute cost is variable.
What is outcome-based pricing?
Outcome-based pricing charges for a result the product delivers, not for usage or seats: a resolved support ticket, a qualified lead, a completed task, a successful match. You anchor the price to the customer’s current cost to produce that outcome themselves. If a support ticket costs them $10 to resolve with a human agent, charging $5 to $7 per resolved ticket is an easy value case. The billable unit varies by industry. Outcome pricing is the emerging frontier for AI SaaS but needs clean measurement and attribution, which is why it is still a few years from being the default.
How do I set an AI SaaS price that protects my margin?
Start from the fully loaded per-unit cost of delivery: all compute, LLM or model API calls, databases, pipelines and any third-party providers tied to serving that unit. Add your target margin to get the list price, then layer volume discounts only as committed volume rises. Protect the margin further with the commitment chain: when a customer commits volume, you commit that volume to your compute or LLM vendors, earn a vendor discount, and pass part of it back. Because AI gross margins run lower than traditional SaaS (around 65 percent or more versus 80 percent or more), pricing off cost rather than off a competitor’s per-seat number is what keeps unit economics safe.
Should AI SaaS bill annually or by usage?
Both, and ideally with commitment. Annual upfront billing is always good: it boosts runway and improves retention because a locked-in customer stays long enough to see value, and a meaningful annual discount of 20 to 30 percent is what actually moves the decision (5 to 10 percent will not). For usage-based AI products, an enterprise customer should give a volume commitment; with no commitment they pay the highest rate. Where there is no annual commitment, use a prepaid wallet or top-up that usage draws down, rather than billing usage in arrears.
How do I bill for usage without running into collection problems?
Avoid billing usage in arrears, where the customer runs up consumption during the month and you invoice afterward: it creates a collection nightmare and puts your variable compute cost at risk. Instead, use a prepaid wallet or top-up model. The customer loads credit upfront and usage draws it down, so you are never out of pocket for compute you have already paid for. For enterprises, pair this with an annual volume commitment. This keeps cash ahead of cost and protects both your margin and your working capital.
This is general educational information for founders, current to mid-2026, and not legal, tax or investment advice. Pricing figures and margin targets are illustrative and vary by product, industry and stage. Vendor terms and model costs change; verify the current position before setting or repricing a plan.
Per-Customer and Per-Unit Margin
The SaaS Subscription Billing Stack
AS | Founder, CFOmatrix | Finance Strategy & Equity Compliance CFOmatrix is a knowledge platform focused on how finance actually works inside growing companies, from SaaS and AI SaaS metrics and unit economics to structure, fundraising and exits. |