Use the same workload, count all delivery costs and keep autonomous success separate from human rescue. The following is the desk’s proposed accounting method, not an industry-standard accounting rule.
Cost per verified outcome = total attributable workflow cost ÷ unique, correctly completed, policy-compliant tasks.
Include, without double counting:
- Platform, speech/model and telephony charges.
- Transfer legs, numbers, recording, storage and relevant messaging.
- Minimum commitments, capacity and support packages.
- Implementation and integration maintenance.
- Monitoring, quality review and supervisor time.
- Human rescue, corrections and repeat-contact handling.
- Exit and migration costs when material to the decision.
Use one consistent measurement boundary. If spam and out-of-scope calls incur charges, retain those costs even if they are excluded from the eligible-task denominator. Deduplicate repeat attempts at the same task rather than awarding several successes for one booking.
Can a higher minute price produce a lower completed-task cost?
Yes in arithmetic; whether it does so in production requires evidence. This example demonstrates the mechanism without making a supplier claim.
Every input below is assumed. Each option receives 1,000 unique eligible tasks monthly. Both AI options consume 4,000 AI minutes. Human-rescue attempts take six minutes at an assumed loaded labour cost of $30 per hour. Telephony allowances include transfer-related usage.
| Monthly input or result | A: Lower-priced AI implementation | B: Higher-priced, more supported AI implementation | C: Human answering-service baseline |
|---|---|---|---|
| AI software usage | $480: 4,000 × $0.12 | $880: 4,000 × $0.22 | — |
| Human answering-service fee | — | — | $3,000 |
| Telephony allowance | $120 | $120 | $120 |
| Support, monitoring and buyer QA | $600 | $400 | $400 |
| Implementation allocated over 12 months | $200 | $300 | $100 |
| Additional human-rescue cost | $750: 250 attempts | $360: 120 attempts | Included in assumed service scope |
| Total monthly workflow cost | $2,150 | $2,060 | $3,620 |
| Autonomous verified completions | 600 | 780 | Not applicable |
| Additional verified completions through rescue | 200 | 100 | Not separately modelled |
| Total verified completions | 800 | 880 | 900 |
| Cost per verified completion | $2.69 | $2.34 | $4.02 |
The higher-priced AI option is cheaper per verified completion in this constructed example because its assumed support and rescue burden is lower. But if the buyer’s agreed completion floor is 90%, neither AI option qualifies: their assumed completion rates are 80% and 88%, versus 90% for the human baseline.
The point is not that option B wins. It is that unit price, total cost and acceptable service are three separate decisions.
Also prepare a cash-flow view: allocating implementation over 12 months does not mean the supplier accepts monthly payment. Separate actual cash savings, avoided hiring and released staff capacity; do not count unchanged payroll as money saved.
How should home-services buyers value recovered demand?
Use a controlled recovery funnel rather than multiplying every missed call by average job revenue:
Missed calls → unique enquiries → eligible jobs → genuinely incremental bookings → completed jobs → contribution after fulfilment costs.
This is our recommended attribution method. Subtract bookings that the existing callback process would have recovered, respect available technician capacity, and show uncertain conversion assumptions separately from measured cost reductions.