This question-and-answer guide helps AI SaaS founders choose a credit conversion schedule and compare subscriptions, prepaid packs, top-ups and refunds. All prices and costs below are illustrative USD amounts, not Ganivra plans, provider quotes or industry benchmarks. The 60% target is a planning assumption, not a recommended universal margin.
Already have a pricing page? Use the free AI pricing strategy review. To test your own delivery costs and margin target, open the AI Unit Economics Calculator.
Offering credits before customers pay? See how to offer an AI free trial without spending too much for trial budgets, usage limits and cost per converted customer.
What is an AI credit, and how is it different from a token?
An AI credit is a selling unit you define; a token is a model-usage unit. A customer might spend credits on a summary, image or research report. Your delivery cost can include model calls, paid tools, retrieval, retries and human review behind that action.
Publish how many credits each action consumes, what changes that amount, and what happens when the balance runs out. For variable work, show an estimate or agreed maximum before starting. See tokens, credits and outcome-based billing for the wider comparison.
Product credits are also distinct from monetary billing credits. Stripe’s billing-credit documentation describes prepaid or promotional grants applied to eligible invoices. Your application must define how an abstract action credit maps to that billing system.
How do you calculate a sustainable price per AI credit?
Start with the least you earn per credit, then calculate the credits each action needs. Include discounts and promotional grants when finding that effective rate.
Effective price per credit = amount allocated to the grant / credits issued
Action revenue allocation = credits consumed × applicable grant rate
Contribution = allocated revenue − included delivery cost
Minimum whole credits per action = ceil(cost / (credit price × (1 − target margin)))Use the same cost boundary throughout. In this example, a basic action costs $0.02 and a premium action costs $0.35, including assumed model, tool and retry costs. Taxes, payment fees, fixed infrastructure, general support and company overhead are excluded. Add the applicable categories to your own model; the example is contribution after included costs, not net profit or full gross margin.
Suppose a $100 monthly allowance grants 1,000 credits and all $100 is assigned to those credits for this operational analysis. Each credit carries $0.10. With a 60% target, the cost budget is $0.04 per credit. A $0.35 premium action needs at least ceil(0.35 / (0.10 × 0.40)) = 9 credits. Start by testing a simple schedule of one credit per basic action and ten per premium action.
If part of the subscription price buys platform access or support, allocate that part explicitly rather than assigning the full price to both access and credits. The minimum formula requires a positive credit price and a target below 100%; free promotional credits require a separate acquisition or retention budget.
What happens when light and heavy users buy the same subscription?
Heavy usage becomes risky when the cost per consumed credit changes. Here every customer buys the same $100 allowance. A basic action uses one credit; a premium action uses ten.
| Usage profile | Credits used | Allocated revenue | Included cost | Contribution | Margin on allocation |
|---|---|---|---|---|---|
| Light: 200 basic actions | 200 | $20 | $4 | $16 | 80% |
| Full basic: 1,000 actions | 1,000 | $100 | $20 | $80 | 80% |
| Full premium: 100 actions | 1,000 | $100 | $35 | $65 | 65% |
| Premium cost spike: 100 actions | 1,000 | $100 | $80 | $20 | 20% |
For the light user, cash collected is $100, but this consumption view allocates only $20 to the 200 credits used. The remaining $80 is unconsumed grant value in this model, not additional revenue assigned to those actions. This is an operational allocation, not an accounting revenue-recognition rule.
The full premium user meets the 60% target at a $0.35 action cost. If longer context and retries raise that cost to $0.80, margin falls to 20%. At $1.20 per action, 100 premium actions would cost $120 against $100 allocated revenue, producing a $20 loss.
Why not charge one credit for every action?
At one credit per premium action, a customer could buy 1,000 premium actions for $100. At $0.35 each, delivery costs $350 and contribution is −$250. A credit cap limits quantity; it does not fix an underpriced conversion schedule.
How do subscriptions, credit packs and top-ups compare?
Compare the same action at each grant’s actual selling rate. A discounted pack lowers revenue per action even if the product uses the same credit schedule.
| Offer | Credits | Price per credit | Revenue per action | Contribution | Margin |
|---|---|---|---|---|---|
| $100 subscription allowance | 1,000 | $0.10 | $1.00 | $0.65 | 65% |
| $80 prepaid pack | 1,000 | $0.08 | $0.80 | $0.45 | 56.25% |
| $30 top-up | 250 | $0.12 | $1.20 | $0.85 | 70.83% |
The subscription meets the illustrative 60% target; the discounted pack misses it. Prepayment changes cash timing, not the cost of delivering a premium action.
Should you change the pack price or the credit schedule?
With ten credits per premium action, the minimum credit price is $0.35 / (10 × 0.40) = $0.0875. A 1,000-credit pack therefore needs at least $87.50 at these costs to meet the target. Alternatively, at $0.08 per credit, charge at least 11 credits per premium action. A 12-credit schedule allocates $0.96 and leaves $0.61, or 63.54%, after the $0.35 cost.
That buffer is small: 12 discounted credits support only 12 × $0.08 × 0.40 = $0.384 cost at the target. If a premium action costs $0.80, it needs 25 credits at the same discounted rate and target. Consider a separately priced high-cost action, an agreed size limit, or a verified cost reduction rather than assuming 12 credits covers every workload.
What if a customer spends subscription credits and a top-up in one action?
Use the recorded grant-consumption order. An action drawing eight subscription credits at $0.10 and two top-up credits at $0.12 receives $1.04 of allocated revenue. At $0.35 cost, contribution is $0.69, or 66.35%. Do not revalue every historical credit at the newest top-up price. Version grant rates and action schedules so old executions remain explainable.
How should refunds and returned credits affect your margin calculation?
Keep cash refunds, unused-credit cancellations and restored product credits separate. None automatically reverses supplier costs already incurred.
Consider the $80 pack with 1,000 credits. The customer consumes 600 credits on 60 premium actions under the original ten-credit schedule. Allocated revenue is $48 and included cost is $21. The remaining 400 credits carry $32 of unconsumed value.
- Refund all unused value: refund $32 and cancel the corresponding 400 credits under the agreed policy. Cash retained is $48; contribution after the $21 cost is $27, or 56.25%. Do not also subtract that $32 from the $48 already allocated to consumed credits.
- Also refund $8 for completed work: net amount retained becomes $40. With the same $21 cost, contribution is $19, or 47.5%, on this net operational basis.
- Restore 100 credits instead of refunding cash: there is no new payment. If redeemed for ten more premium actions at the original schedule, they add $3.50 of delivery cost. Record the restoration and its future cost without inventing new revenue.
These examples compare operational economics, not refund entitlement or accounting recognition. Apply the terms agreed with customers and your finance policy. Record the original grant, consumption, cash adjustment and credit correction separately to prevent double counting.
How do you prevent heavy usage from creating unexpected losses?
Combine transparent pricing with reliable balance handling and cost monitoring. Pricing alone does not prevent an application from delivering work after credits run out.
- Estimate or reserve credits before long-running work and settle against the documented rule afterward.
- Handle concurrent jobs and duplicate events so the same balance cannot fund unlimited simultaneous work.
- Keep paid retries in your cost records even when the customer receives credits back.
- Show balances, consumption history, rollover and expiry rules clearly; test full redemption rather than relying on unused balances.
- Require an explicit customer choice for top-ups or an agreed automatic-top-up policy.
- Track premium-heavy customers separately from averages and flag unpriced usage as incomplete.
Ganivra can compare reported credits and payment context with recorded costs. Its credit and transaction integration supports that economics analysis; your application remains responsible for balances and enforcement. For the review process, see which customers cost more than they pay.
What pricing decision follows from this worked example?
Do not launch the $80 pack with the ten-credit premium schedule if 60% is your required contribution target. At the assumed $0.35 cost, either raise the pack to at least $87.50 or increase premium consumption to at least 11 credits. A 12-credit schedule adds some room, but the $0.80 stress case still requires a separate decision.
Before changing an existing offer, evaluate customer impact and the terms attached to already-purchased credits. Test the proposed schedule with realistic heavy-user task sizes and completion quality, then monitor actual costs after launch.
How can you test the example in Ganivra’s calculator?
Open the AI Unit Economics Calculator, choose “Other,” and enter 100 attempted and successful premium actions, $35 total included delivery costs counted once, $80 workload revenue, and a 60% target. The expected margin is 56.25% and required revenue is $87.50. For the cost-spike case, replace $35 with $80: required revenue becomes $200. The calculator evaluates workload economics; it does not maintain a credit ledger.
Use the free pricing strategy review to examine your published offer, and the calculator to test private cost assumptions. A website-only review cannot establish your actual delivery costs.
Frequently asked questions about AI credit pricing
How do I price AI credits without losing money on heavy users?
Set credits per action from delivery cost, your lowest effective selling price per credit, and a target margin. Test heavy usage, expensive workflows, discounts, retries and refunds. Enforce balances or agreed usage limits in your application, and review actual customer contribution regularly.
How much should one AI credit cost?
There is no universal AI credit price. Divide the amount allocated to a credit grant by the credits issued, then check whether each action earns enough at that rate to cover its delivery cost and target margin. A $100 grant of 1,000 credits has an effective price of $0.10 per credit.
How many credits should an AI action consume?
For whole-number credits, divide the included action cost by the effective price per credit times one minus the target margin, then round up. For a $0.35 action, $0.08 per credit and a 60% target margin, the minimum is 11 credits before adding a cost-variation buffer.
Are AI credits the same as tokens?
No. Tokens measure model input or output. Credits are a product-defined selling unit that can represent an action using models, tools and other services. Publish the conversion schedule and explain when an action can consume more credits.
Should I sell subscriptions, credit packs or top-ups?
Subscriptions provide recurring access with an allowance, packs let customers prepay for usage, and top-ups add capacity after an allowance is exhausted. Compare all three at full consumption and at the same cost boundary. A discounted pack can earn less per action than the subscription.
How do refunds affect AI credit margins?
A cash refund reduces the amount retained, while provider costs already incurred remain unless the provider also credits them. An unused-credit refund and compensation for completed work need separate treatment. Returning product credits creates additional future usage capacity; it does not reverse the cost of the original execution.
Should unused AI credits count as profit?
Do not rely on unused credits to make the unit economics work. Track cash collected, consumed-credit revenue allocation and unused balances separately. Test full redemption and rollover. Accounting treatment of unused or expired credits follows your contracts and accounting policy, not this operational pricing model.
How can I calculate AI credit profitability with Ganivra?
Use Ganivra’s AI Unit Economics Calculator to test workload costs, revenue and target margin, or its pricing strategy review to examine your published pricing. For ongoing analysis, report customer usage, credits consumed and payment context so Ganivra can compare allocated revenue with recorded costs. Your application remains responsible for credit balances and usage enforcement.
Does your credit pricing survive heavy usage?
Review the offer customers see, then check it against the cost of delivering their work.
Review my AI pricing →Method: original illustrative calculations by Ganivra Editorial. Product and billing references reviewed November 10, 2026. No third-party provider rates or customer outcomes are claimed.