Call Center Cost per Call and Cost per Resolution Benchmarks

Cost per resolution is the total cost of handling calls divided by the number of calls that actually solved the caller's problem. Call center cost per call tells you what a conversation costs. Cost per resolution tells you what an outcome costs, which is the number that matters once an AI voice agent and a human team share the same queue.

This page gives the formula, the few published benchmarks worth citing, and a worked comparison of human-only, AI-first and hybrid handling built from sourced wage data and published component prices.

How is cost per resolution different from cost per call?

Cost per call divides spend by every handled call, resolved or not. Cost per resolution divides the same spend by resolved calls only, so every failed or repeated call raises it. Two operations with the same cost per call can have very different costs per resolution if one of them sends a third of its callers back into the queue.

The gap matters most with automation. An AI-handled call is cheap per minute, but a call the agent cannot finish still costs money and then costs again when a person picks it up. Measuring per resolution puts that second cost back in view.

How do you calculate cost per resolution?

Add up everything it cost to handle a period's calls, then divide by the calls that were resolved in that period. For a single channel, that is the same as dividing cost per call by the resolution rate.

Cost per resolution formula: Cost per resolution = Total handling cost ÷ Resolved calls

1,000 calls cost $2,478 to handle, and 900 of them were resolved. $2,478 ÷ 900 = about $2.75 per resolution.

Use the same definition of "resolved" for AI-handled and human-handled calls, or the comparison is meaningless.

Decide up front what counts as resolved. A common definition is "no repeat contact on the same issue within a set window, such as seven days." The first-call resolution benchmarks article covers how teams set that window. Whatever you choose, apply it the same way to AI-handled and human-handled calls.

What do published benchmarks say about cost per contact?

Very few current, public benchmarks exist, and most cost-per-call figures you will find online are unsourced. Two are reliable enough to use as reference points:

• Gartner, 2019: in Gartner's 2019 Customer Service and Support Leader poll, live channels such as phone, live chat and email cost an average of $8.01 per contact, while self-service channels such as company websites and mobile apps cost about $0.10 per contact (Gartner press release). The figure is from 2019 and blends phone with chat and email, so treat it as a ceiling check, not a target.
• US Bureau of Labor Statistics: the median hourly wage for customer service representatives was $21.53 in May 2025 (BLS Occupational Outlook Handbook). Wages were 70.0% of total employer compensation costs for private industry workers in June 2026, with benefits making up the other 30.0% (BLS Employer Costs for Employee Compensation).

There is no neutral, published benchmark for AI voice agent cost per resolution yet. Vendor figures depend on the model, the voice, call length and how each vendor defines "resolved." The reliable approach is to build the number from your own call data, as below.

What does a human-handled resolution cost?

Start from the BLS median wage and load it for benefits: $21.53 divided by 0.70 is about $30.76 per paid hour, or roughly $0.51 per paid minute. Then account for the time agents spend not handling calls (training, breaks, idle time between calls). That share is specific to your floor, so use your own workforce data.

For illustration, assume a 6-minute handle time including after-call work, and that 60% of paid time goes to handling calls. Each call then consumes 10 paid minutes, or about $5.13 in labor. If 75% of those calls are resolved, labor cost per resolution is about $6.84. This leaves out seat licenses, telephony, supervision and facilities, so a real figure is higher.

What does an AI-handled resolution cost?

On Retell AI, a voice agent minute is the sum of its components, listed on the Retell pricing page. One common setup on pay-as-you-go rates as of October 2026:

• Retell voice infrastructure: $0.055 per minute
• LLM (GPT 5.1): $0.04 per minute
• Text-to-speech (Retell platform voices): $0.015 per minute
• Knowledge base add-on: $0.005 per minute
• Telephony on a Retell US number: $0.015 per minute

That totals $0.13 per minute, so a 3-minute AI-handled call costs about $0.39. If the agent resolves 60% of calls on its own, AI cost per AI resolution is about $0.65. A different model or voice changes this quickly: GPT 5.5 alone is $0.16 per minute. The monthly cost calculation guide walks through every line item, including fixed monthly charges.

How do hybrid AI and human workflows change the math?

Most deployments are hybrid. On an inbound support line, the agent handles what it can and transfers the rest. On Retell, once a transferred caller is connected and the agent drops off, the AI fee stops and only telephony continues for the rest of the call. The cost of a transfer is the AI minutes already spent plus the full human cost.

Using the assumptions above, per 1,000 calls with a 3-minute AI segment, 6 more minutes on the line after each transfer and human handling unchanged:

• Human only, 75% resolved: Total cost: $5,130. Cost per resolution: $6.84.
• AI first, 40% transferred: Total cost: $2,478. Cost per resolution: $2.75.
• AI first, 20% transferred: Total cost: $1,434. Cost per resolution: $1.51.

With 40% transferred, AI minutes cost $390, transferred telephony costs $36 and human labor on 400 calls costs $2,052. That is $2,478 for 900 resolutions (600 by the agent, 300 by people), or about $2.75 each. At 20% transferred, the same math gives about $1.51 per resolution. Containment, not the per-minute rate, moves the number most. The guide to measuring and improving call containment covers how to raise it.

A transferred call costs more than a human-only call. The caller has already used AI minutes before reaching a person, so raising containment lowers cost per resolution faster than cutting the per-minute rate.

What resolution rates do production voice agents report?

Published customer results give a sense of the range, though each company defines success differently:

• In Retell's case study, Anker reported an 80.4% case resolution rate on tier-1 post-sales support calls in the US and UK (Anker case study).
• In Retell's case study, Everise contained 65% of internal service desk voice calls that previously went to a person (Everise case study).
• In Retell's case study, Medical Data Systems reports a 70% containment rate across inbound patient billing support, with a 30% transfer rate (Medical Data Systems case study).

Containment and resolution are not the same. A contained call that leaves the caller to ring back tomorrow lowers cost per call and raises cost per resolution. Track both.

How do you measure cost per resolution in your own operation?

Measure it from call records rather than estimating it:

1. Define success per agent. In Retell, the built-in Call Successful field in post-call extraction marks whether a call met the success criteria you write for that agent.
2. Chart cost against outcomes. The analytics dashboard can plot Combined cost and Call Successful Rate side by side, broken down by agent or agent version.
3. Add the human side. Pull handle time and repeat-contact data for transferred calls from your contact center platform and cost them with your loaded wage.
4. Recalculate after every agent change. A prompt edit that lifts success by a few points can matter more than any rate negotiation.

If you want to test these numbers on your own call mix before committing to a contract, talk to the Retell team about a pilot on real calls.