Founders
How to model inference costs for AI voice agents before launch
AI voice agents: Estimate inference cost from a defined workflow, not a headline price for a model.
Estimate inference cost from a defined workflow, not a headline price for a model. Count requests, input and output sizes, retries, tool calls, and the mix of configurations. Keep vendor prices dated and separate them from fictional exercise assumptions.
Start with the decision
Build a per-completed-task cost bridge. Add failure and retry scenarios, then compare the estimate with observed logs once available. Include related hosting and human delivery costs in a broader margin model rather than renaming inference cost as total cost.
The illustrative scenario
This is an educational workbook. The company or fund and all numerical inputs below are fictional. They are not ADIN results, customer claims, market benchmarks, or a forecast.
The fictional product handles appointment requests for a multi-location service business. A caller may change their request mid-conversation, and the scheduling system is the final record of an available slot.
A short call is not useful if it books the wrong appointment. Separate successful scheduling from hangups, transfers, and calls where the caller never confirmed the details.
A monthly cost bridge with visible assumptions
For this exercise, assume 850 completed units in a month, 6 minutes of human work per unit, and a fictional labor rate of $30 per hour. The labor rate, service cost, and revenue are invented inputs. They are not current vendor prices, customer results, or a suggested commercial offer. A short call is not useful if it books the wrong appointment. Separate successful scheduling from hangups, transfers, and calls where the caller never confirmed the details.
The bridge subtracts only the costs shown. The remainder is not an accounting gross margin or profit figure: other expenses, taxes, timing differences, and classifications are outside this simplified model. A negative result is useful information about the assumptions, rather than evidence that an entire product category cannot work.
Now change one input. If the workflow requires twice the review time, the human-work line doubles even when the provider's unit price stays constant. If volume rises, distinguish a genuinely fixed cost from a cost that will step up. Reconcile the model with observed delivery logs before using it to support a pricing or fundraising claim.
| Item | Illustration |
|---|---|
| Completed workflow units | 850 |
| Assumed revenue for this month | $2,200.00 |
| Variable service cost | 850 × $0.21 = $178.50 |
| Listed fixed delivery cost | $420.00 |
| Human work at an assumed $30/hour | $2,550.00 |
| Remainder after these listed costs | -$948.50 |
Use this worksheet
Answer the questions with a source, date, owner, and next action. Mark an unknown as unknown. Use public or permissioned information and involve a qualified specialist when a question requires expertise beyond the team’s competence.
- What is one completed task?
- How many calls does it require?
- What retries occur?
- Which prices are current?
- What costs are outside inference?
Questions specific to this scenario
- Who confirms the booking?
- What happens when a caller interrupts?
- How is a failed transfer recovered?
Collect permissioned test calls, interruption cases, confirmation records, and failed booking attempts. Keep different languages and noisy environments visible in the sample.
A concrete next test
Run permissioned tests with an interruption, an unavailable appointment, and a failed transfer. Ask the scheduling owner to verify the final records, then inspect what the caller was told. A correct transcript alone does not establish a correct booking.
Interpret the result
The model is a planning tool whose uncertain inputs should remain visible. A cheaper model can cost more per completed task if it increases retries or correction work, so compare the full workflow before switching.
Compare completed, correct bookings with the existing phone process. Include repeat calls and staff repair time instead of measuring call duration alone.
Sources and method
These primary resources provide background for the concepts. The fictional scenario, calculations, and worksheet are ADIN educational examples; the sources do not report or validate those invented results. Reviewed October 11, 2026.