Your AI Cost Is Not Your Price
A report costs you tokens, retries, storage, and model calls. The customer buys none of them. Build a price around the result without losing control of margin.
Updated 2026-08-15
The customer sees one result. You pay four small bills.
This is the unit the customer understands.
One report may trigger tokens, OCR, retrieval, storage, and three retries. Your infrastructure sees five costs. The customer sees one finished report.
If you expose the machinery, the bill becomes homework. If you hide it completely, one heavy account can erase the margin. AI pricing lives inside that tension: sell the result customers understand without losing sight of the cost they never see.
Customers buy the ending, not the machinery
Supplier cost helps set the floor, but it is not automatically the unit you should sell.
| Inside your system | What the customer buys |
|---|---|
| Tokens and retries | A resolved conversation |
| OCR pages and model calls | An extracted document |
| GPU time and generation steps | A finished image |
| Retrieval, storage, and calls | A completed report |
Price the result when it is clear and measurable. Expose a technical unit only when the customer is already technical and needs direct control over it.
Find the cost hiding behind one click
Do not count only the successful request. Include input, output, external tools, storage, failed work, and retries. Then compare cost by feature, customer, and period.
If a report costs SAR 0.22 on average, that does not make SAR 0.22 its price. Cost defines the floor. The value of the report helps define the price.
Let behavior choose the model
| Model | Use it when | Decision to settle |
|---|---|---|
| Subscription with an allowance | Usage is reasonably consistent | What happens after the limit? |
| Usage units | Actions have different costs | How many units does each action consume? |
| Direct usage | The customer controls volume | How will they predict the bill? |
| Hybrid | You need a fixed base with room to grow | What is inside the plan and what is outside it? |
| Outcome-based | Success is clear and provable | When does the outcome count as successful? |
For AI reports, a plan can open the feature and include a monthly amount. Above that limit, the customer pays per report or buys additional units.
Credits should remove complexity, not rename it
Usage units help combine different actions: 5 units for a report, 2 for an image, and 1 for document extraction.
But they add another layer customers must understand. Make clear:
- how many units each action needs;
- how many the plan includes;
- when they renew or expire;
- whether they roll over;
- what happens at zero.
If customers must convert units back to currency before every action, you did not simplify pricing. You only renamed the complexity.
Access and consumption are different promises
A plan may open a higher-quality model, more concurrency, or enterprise controls. Those are entitlements. Reports, images, and minutes are usage.
Keeping them separate lets you change a limit or action cost without rebuilding feature availability inside the product.
A healthy model protects margin and trust
Review cost per action, cost-to-revenue, unused units, overages, and billing questions. A healthy margin with a confusing calculation is still a weak model. A clear calculation where one customer consumes the margin needs a better limit or conversion rule—not a price increase for everyone.
Tirdad brings entitlements, meters, usage units, subscriptions, and invoices into the same model. Customers see a clear plan while your team keeps the detail needed to operate it.
If you choose usage units: implement them step by step in Turn Different AI Actions into One Credit Balance