Price the whole operating service, not just the connected minute. Use public rates to construct a bill of materials, then replace them with the selected configuration and contract.
What do published prices establish?
| Published item | Public price observed | What it does not establish |
|---|---|---|
| Retell AI voice-agent headline range | $0.07–$0.31 per minute | Not a universal all-in ceiling; the page separately lists components and add-ons. (retellai.com) |
| Vapi hosting | $0.05 per minute | Model and other provider costs are additional; support packages and add-ons can change the total. (vapi.ai) |
| Vapi HIPAA-eligible handling add-on | $2,000 per month | Does not replace configuration review, agreements or your own HIPAA obligations. (vapi.ai) |
| Twilio US local inbound calling | $0.0085 per minute | Connectivity, not an entire voice agent. (twilio.com) |
| Twilio US local outbound calling | $0.014 per minute | Check destination-specific pricing and the actual call architecture. (twilio.com) |
| Twilio US local number | $1.15 per month | Number rental only. (twilio.com) |
Do not add both platform and component charges without checking inclusions. Equally, do not remove carrier costs because a platform labels its own SIP transport free.
Retell’s pricing FAQ states that silence and hold time are billable and that, after transfer, its AI-agent fee stops while telephony charges continue. Model those rules against your workflow rather than assuming every billed minute ends when the bot stops speaking. (retellai.com)
What implementation budget should you reserve?
The following is an illustrative 160-hour resource plan, not a supplier quote or market-rate survey. Replace the hours and loaded rates with your own staffing and scope.
| Workstream | Assumed hours | Assumed loaded rate | Modelled cost |
|---|---|---|---|
| Operations discovery, rules and coordination | 40 | $50/hour | $2,000 |
| Engineering, integration and recovery | 80 | $100/hour | $8,000 |
| QA, test execution and documentation | 30 | $50/hour | $1,500 |
| Specialist legal/privacy/security review allowance | 10 | $250/hour | $2,500 |
| Labour subtotal | 160 | — | $14,000 |
| Assumed setup and test-usage allowance | — | — | $500 |
| Illustrative contingency: 20% of $14,500 | — | — | $2,900 |
| Planning provision | — | — | $17,400 |
Internal staff time may not create a new invoice, but it is still a resource commitment. If you cannot staff the work, obtain scoped external quotes rather than describing the project as free to implement.
This budget excludes any unquoted production minimums, substantial enterprise procurement, major legacy-system work or additional regulated-workflow requirements.
What does cost per completed task look like?
Consider this fully illustrative monthly scenario:
- 1,000 unique eligible tasks, each with one four-minute AI call.
- 600 autonomous completions.
- 150 human-rescued completions.
- 250 unresolved tasks.
- No repeat calls in the assumed dataset; actual repeat contacts must be added.
- An assumed blended AI-and-telephony rate of $0.18 per minute.
- Twenty total intervention hours across successful and unsuccessful rescue work.
| Cost item | Calculation | Monthly cost |
|---|---|---|
| AI and telephony usage | 4,000 minutes × $0.18 | $720 |
| Human intervention | 20 hours × $35 | $700 |
| Monitoring and QA | 12 hours × $45 | $540 |
| Hosting and miscellaneous allowance | Assumption | $100 |
| Implementation allocation | $17,400 ÷ 12 months | $1,450 |
| Total modelled workflow cost | $3,510 |
There are 750 verified completions, including human rescue. Therefore:
- Direct AI usage per attempted task: $0.72.
- Direct AI usage divided by autonomous completions: $1.20.
- Modelled fully loaded workflow cost per verified completion: $4.68.
- Recurring modelled cost excluding the implementation allocation: approximately $2.75 per verified completion.
These figures do not prove savings. Compare with the same workload under the alternative, include any omitted shared labour, credits, repeat contacts and retained answering-service fees, and use the same completion definition.
Which hidden costs should you check?
Request line items for:
- Testing and simulation usage.
- Failed attempts, voicemail and billing increments.
- Transfer legs, waiting and hold time.
- Concurrent-call capacity.
- Knowledge retrieval, redaction and quality analysis.
- Support packages and response commitments.
- API, middleware and business-system access.
- Minimum commitments and prepaid balances.
- Monitoring, corrections and on-call coverage.
- Security review, contract changes and exit assistance.
For sensitivity analysis, a $0.05-per-minute saving over 4,000 minutes is $200. Five additional monitoring hours at an assumed $45 per hour cost $225. Use that arithmetic to test whether a cheaper configuration remains cheaper after operating effort.
Keep three benefit categories separate: actual cash savings, avoided future hiring and incremental fulfilled-job contribution. Released staff time is not automatically a payroll saving, and a booking is not automatically incremental revenue.