Gross margin is an accounting ratio derived from recorded revenue and cost of revenue under your accounting policy. AI contribution margin is a narrower operational ratio: processor-net observed revenue minus attributable AI cost, divided by that same processor-net observed revenue. Use it to inspect AI product economics, never as a substitute for gross margin or company profit.
Two margins, two scopes
Gross margin = gross profit ÷ recorded revenue
Processor-net observed revenue = observed payment revenue − refunds − processor fees
AI contribution profit = processor-net observed revenue − attributable AI cost
AI contribution margin = AI contribution profit ÷ processor-net observed revenue
The first formula depends on the company’s financial statements and accounting classification of cost of revenue. The second is a deliberately limited product-operations metric used by Grow or Die. It uses processor-net observed revenue as the denominator and includes only the payment and model-cost evidence named in the formula.
Do not rename AI contribution margin “gross margin.” SEC staff guidance warns that a non-GAAP measure can be misleading when its label does not reflect its calculation or when it uses the same label as a differently calculated GAAP line item.
Worked example: reconcile the gap instead of hiding it
This hypothetical company uses one complete month and one currency. Its accounting policy classifies payment fees, model cost, hosting, and allocated support inside cost of revenue. A different company may classify costs differently, so the gross margin calculation must come from its own books.
| Line | Amount | AI contribution metric | Gross profit example |
|---|---|---|---|
| Gross payments | $10,000 | Included | Starting evidence |
| Refunds | ($500) | Included | Assumed reduction of recorded revenue |
| Processor fees | ($300) | Included | Included in cost of revenue for this example |
| Processor-net observed revenue | $9,200 | Denominator | Reconciliation subtotal |
| Attributable AI cost | ($1,200) | Included | Included in cost of revenue for this example |
| Other hosting | ($700) | Excluded | Included in cost of revenue for this example |
| Allocated support | ($500) | Excluded | Included in cost of revenue for this example |
| Result | — | $8,000 | $6,800 |
AI contribution margin: $8,000 ÷ $9,200 = 87.0%
Hypothetical gross margin: $6,800 ÷ $9,500 = 71.6%
The 15.4 percentage-point gap is not an error. It is the cost that the narrower operational metric intentionally leaves out. The useful action is to name and reconcile the gap, not force the two numbers to match.
Choose the metric that matches the decision
| Question | Use | Why |
|---|---|---|
| What do our financial statements report? | Gross margin from the accounting system | It follows the company’s revenue and cost classification. |
| Does payment revenue cover model usage? | AI contribution margin | It isolates payment, refund, fee, and AI-cost evidence. |
| Which model or feature threatens unit economics? | AI contribution margin plus cost breakdown | The aggregate alone cannot identify the driver. |
| Is the company profitable? | Neither metric alone | Operating expenses, taxes, and other costs remain outside. |
| Should we publish this metric externally? | Finance and legal review | Non-GAAP presentation and reconciliation rules may apply. |
What these margins cannot prove
- Neither margin proves that a model call or feature caused a customer to purchase.
- Site-wide AI contribution margin cannot identify a loss-making customer without exact payment and usage identity links.
- SDK-priced model cost may not equal the provider invoice until usage coverage and pricing are reconciled.
- A positive margin for one month does not establish lifetime value, retention, cash flow, or company net income.
- A payment processor ledger is not a replacement for the company’s accounting records.
Implementation rules that keep the comparison honest
- Use one complete time window and one reporting currency.
- Keep refunds and processor fees explicit.
- Price input, output, and cache usage with the applicable model schedule.
- Label missing fees as a before-fee metric instead of estimating them silently.
- Reconcile the operational metric to the nearest accounting subtotal before sharing it externally.
- Display unallocated revenue, unallocated cost, and identity coverage beside customer results.
Start with why token cost is not profit , review the complete AI unit economics formula , and then learn how to classify customer contribution without averaging .