Cost per resolved call: build a better voice AI business case

Voice AI buying decisions often start with a rate card: cost per minute, subscription tier or implementation estimate. Those numbers matter, but they do not answer the operational question. A business is not paying to produce audio. It is paying to make a customer request easier to resolve, hand off, record and improve. The business case should compare cost per resolved outcome while making assumptions visible.

Dring now works with subscription packages rather than a pay-as-you-go promise. The right package depends on workflow scope, language and channel mix, expected capacity, integrations, quality requirements and support needs. Its pricing page is the starting point for a package conversation; this framework helps a team decide which baseline and outcome to bring into that conversation.

Define “resolved” before calculating cost

Choose one workflow and write the completion event. A support call may be resolved when an approved answer is provided and the case does not reopen in the agreed window. A callback request is resolved when the callback is completed by the correct owner. A sales qualification is resolved when a qualified meeting is booked or the lead is closed with a clear reason. Without a workflow-specific definition, cost comparisons become a contest of optimistic labels.

Keep transfers visible. If a complex case is correctly handed to a human, it may still be a successful outcome even though the AI did not contain it. The cost model should include the human minutes that the handoff consumes and the value of the context passed. Avoid hiding human work to make the automation appear cheaper.

Build the baseline with the real operating costs

A credible human-only baseline can include agent time, supervisor and quality time, scheduling or dispatch work, telephony, overtime, training, repeat contact, rework and the cost of unresolved demand. Include only costs that belong to the selected workflow and period. Do not compare a fully loaded human operation with an AI line that excludes integration, monitoring or human exception handling.

Make the period explicit. A seasonal peak, a monthly average and a new launch month are different baselines. State whether inbound and outbound calls are included, which lines are in scope and whether the comparison covers voice only or voice plus messaging. This is the level of precision investors and operators need when reviewing a cost claim.

Use cost per outcome, not cost per call

A simple model is: total operating cost for the workflow divided by the number of verified resolved outcomes. Total cost can include the Dring subscription package, telephony, integrations, internal oversight, human handoff time, quality review and one-time setup amortised over an agreed period. Resolved outcomes should come from the CRM or an approved event, not from call duration alone.

For example, an agent may answer many calls but create few completed outcomes if data is stale or handoffs fail. Another agent may transfer more cases but resolve the complex work faster because the human receives a clean brief. The second model can have a better cost per resolved call even with lower containment.

Model capacity and subscription fit

Subscription planning should reflect demand shape, not a single average. Look at peak hours, language distribution, weekday and weekend traffic, inbound and outbound campaigns and the number of simultaneous workflows. The capacity planning guide explains why a package should be selected around the operating envelope and the next realistic release.

Ask what happens when demand grows. Does the team need more channels, more languages, more agents, higher concurrency, deeper analytics or a new quality suite? A cheap package that cannot support the next workflow may be more expensive after migration and rework. Conversely, buying capacity that the operation cannot use hides the business case behind unused scope.

Include quality and improvement as value

Human teams spend time sampling calls, updating scripts, coaching agents and finding recurring customer friction. A voice AI operating model should include those activities rather than presenting improvement as free. Dring's Agent Factory and quality controls make the feedback loop visible: calls are reviewed, failures become tests and controlled changes are released.

The value is not only lower unit cost. It can be shorter time to identify a broken policy, fewer repeated explanations, better coverage during peaks and a clearer management view of customer demand. Keep these as separate value hypotheses and validate them with measured data.

Use confirmed claims with careful qualifiers

Dring's current operating signals include approximately 300,000 customer conversations per month across inbound and outbound production workloads, more than 10,000 agent conversation minutes per day and approximately 72% resolution across current production workloads. These are operating signals, not a universal forecast for every customer. Scope, time period and outcome definitions should remain visible when they are used in a business case.

On selected deployments, operating cost has been up to 81% lower than the agreed human-only baseline. “Up to” matters. The result depends on the workflow, volume, human exception policy, telephony and baseline definition. Use the figure as a reference for a scoped comparison, not as a promise that every deployment will achieve the same result.

Build a decision table for the CFO and operator

Show monthly subscription, telephony, integrations, internal oversight, handoff minutes, quality review and expected resolved outcomes. Add a low, base and high scenario for volume and resolution. Include implementation timeline, package upgrade triggers, risks and the evidence needed before expanding traffic. The operator should see what changes in the day-to-day work; the CFO should see which assumptions move the result.

Use Sector Insight to understand the demand behind the numbers. If calls reveal a product or policy problem, automation may lower handling cost while the unresolved root cause remains. A strong business case funds the fix, not just the voice line.

Cost per resolved call checklist

• Define a workflow-specific resolved outcome and follow-up window.
• Scope the baseline by line, channel, period and inbound/outbound mix.
• Include subscription, telephony, integrations, oversight, handoffs and quality.
• Model peak demand and package fit, not only monthly averages.
• Separate containment from resolution and show rework and repeat contact.
• Qualify all performance claims and preserve the agreed baseline.
• Use low, base and high scenarios before expanding the rollout.

A good voice AI business case is not a promise that technology eliminates people or cost. It is a transparent comparison of how a defined customer outcome gets delivered today and how it could be delivered with better coverage, clearer evidence and a more focused human team. That is a decision an operator can defend after launch.

Further reading

• McKinsey on efficiency and customer experience in generative AI care
• Deloitte research on contact centre efficiency and analytics
• Twilio on reducing contact centre costs with AI