What is this workflow actually costing you?
Most businesses underestimate by a factor of three, because they count the salary and ignore the interruption, the error rate and the revenue that walked.
The short version: most businesses underestimate what a repetitive workflow costs by a factor of three, because they count the salary and ignore the interruptions, the error rate and the revenue that walked. The model below is shown in full so you can audit it.
Last reviewed August 2026 · Sources cited inline
Use the calculator
No email gate.
Nothing leaves your browser. We default to 70% task capture, not 100%, because the last stretch of any workflow is judgement and should stay human.
Build it. This is the easy decision.
How the calculation works
- Direct labour recovered. Hours per week × loaded hourly cost × 52, multiplied by the proportion of the task an agent can genuinely take. We default to 70%, not 100%, because the last stretch of any workflow is judgement and should stay human.
- Error reduction. Current error rate × average cost per error × annual volume. Include rework, credits, penalties and the hours spent fixing it. In compliance-heavy work this line frequently exceeds the labour line.
- Recovered revenue. For anything customer-facing: missed calls, unanswered enquiries, quotes never followed up. Volume × conversion rate × average job value. In trades and clinics this is usually the largest number on the page and the one nobody tracks.
- Cost side. Build cost, amortised over 24 months, plus usage costs (AI voice agent usage typically runs $0.05–$0.15 per minute on mainstream platforms; Retell AI, 2026), plus the AgentOps retainer. We include all three. A calculator that omits ongoing costs is a sales tool, not a model.
What the output means
| Payback | What we’d tell you |
|---|---|
| Under 6 months | Build it. This is the easy decision. |
| 6–12 months | Build it, but start with one workflow and prove it. |
| 12–24 months | Marginal. Look for a second workflow to share the integration cost. |
| Over 24 months | Don’t. We’ll tell you this in the audit too. |
Manual vs agent: what actually changes
| Manual | With an agent | |
|---|---|---|
| Coverage | Business hours | Always |
| Response time | Minutes to days | Seconds |
| Consistency | Varies by person and by mood | Identical every time |
| Capacity at peak | Queues or breaks | Flat |
| Audit trail | Partial, reconstructed | Complete, timestamped |
| Cost curve | Linear with volume | Near-flat with volume |
| Judgement calls | Native | Escalated to a human |
That last row is the one to read carefully. Agents don’t replace judgement. They remove the work surrounding it.
Frequently asked questions
Aren’t these numbers just marketing?
The model is shown above so you can audit it. We use 70% task capture rather than 100%, we amortise build cost, and we include usage and retainer costs. Most vendor calculators do none of those things.
What if we don’t know our error rate?
Estimate low. If the case works at a conservative error rate it works at the real one.
Does this replace the audit?
No. This gives you an order of magnitude. The audit tests whether your systems will actually permit the integration, which is the thing that decides feasibility.
Why don’t you ask for my email?
Because you’d give us a fake one. If the number is compelling you’ll book.