How to calculate cost per resolved call: voice AI vs human agent

Quick answer: To calculate cost per resolved call, divide the full cost of a service by the customer tasks it resolved and verify each resolution after an agreed observation window. Compare voice AI with human agents only when both figures cover the same intent, operating hours, customer cohort, repeat contacts, and downstream work.

What counts as the full cost?

For voice AI, include platform and telephony charges, implementation cost allocated across the measurement period, integration maintenance, support, quality review, and the human work that follows a transfer. For the human-agent baseline, include the labour or BPO cost attributable to the same calls, telephony, supervision, quality review, repeat contacts, and any downstream work needed to finish the task. Finance should document how it allocates shared costs in both figures. Keep released agent capacity separate from cash savings because the operation may retain the same staffing cost. Use actual invoices and finance records whenever they exist. Label every forecast or assumed input as modelled. Preserve the calculation sheet so another reviewer can reproduce the result.

How do you calculate cost per resolved call?

Use this formula:

Cost per resolved call = total attributable cost ÷ verified resolved calls

Define a resolution before the pilot starts. The definition should name the target outcome, the evidence that proves completion, and the observation window used to catch a repeat contact or reversal. Count a transferred call as a voice AI resolution only when the agreed definition says the automated service completed the task before transfer. Record transfers, abandoned calls, unresolved calls, and repeat contacts separately so the denominator doesn't hide them.

The calculation has four steps:

Select one call intent and the measurement period.
Add every cost attributable to that cohort.
Count the calls that meet the agreed resolution definition after the observation window.
Divide the attributable cost by the verified resolutions.

What does a modelled example look like?

Assume a monthly cohort contains 20,000 eligible calls. The team verifies 12,000 resolutions after its observation window, and the total attributable service cost is £72,000. The modelled cost per resolved call is £6 because £72,000 divided by 12,000 equals £6.

This example illustrates the formula. It is not a Butter Labs result, customer result, price, or ROI claim. Replace every figure with finance records and measured pilot data before making a buying decision.

How should voice AI be compared with human agents?

Calculate the human-agent baseline with the same intent, hours, customer cohort, outcome definition, and observation window. A blended cost across the whole contact centre can distort the comparison when the automated cohort contains simpler or harder work. Apply the same treatment to repeat contacts, escalations, and downstream fulfilment in both figures.

Then show the raw counts beside each rate: eligible calls, verified resolutions, transfers, abandoned calls, unresolved calls, and repeat contacts. The contact-centre AI metrics guide explains how to keep those measures distinct. The containment versus resolution guide explains why a contained call may still leave the customer's task unfinished.

Why add a downside case?

Recalculate the model with fewer verified resolutions, more transfers, and a longer implementation-cost allocation. Add the cost of human handling after transfer and any quality review needed during the measurement period. This downside case shows finance how the result changes when live traffic differs from the central assumptions.

Discuss a measured voice AI pilot with Butter Labs.

Reviewed 14 October 2026