AI Phone Answering Service Research: Costs, ROI, and What to Measure

An AI phone answering service should answer questions accurately, complete the caller’s request, and reduce the work left for your team. Evaluating those outcomes requires more than listening to a convincing voice or checking a feature list.

The 2026 joint research by LunaCal and RingJenny.com on AI answering services, covers ten platforms through 27,000 simulated call scenarios, 200 modeled buyer profiles, and 1,000 product evaluations. It reports task completion, booking, usability, conversational timing, costs, and administrative ROI. Ring Jenny has the highest simulated resolution rate at 92.7%; Upfirst has the highest calculated percentage ROI at 221% under the study’s financial assumptions. Understanding the difference helps businesses choose an answering service for the work their calls actually require.

What does the AI phone answering service research show?

Task completion, usability, and response timing

Source: Joint Research by LunaCal and RingJenny.com on AI Answering Services (2026), published October 29, 2026. Call outcomes and timing are simulated; ease-of-use ratings belong to the study’s scenario framework.

Ring Jenny leads simulated resolution, booking success, and ease of use. Smith.ai follows on resolution at 91.7%, while Upfirst is next on usability at 9.1/10. The source research separately reports Smith.ai’s transfer success at 97.7%, ahead of Ring Jenny’s 95.7%. That distinction matters for a business whose main requirement is connecting callers with a person.

Costs and first-year administrative ROI

These are the research’s October 29, 2026 cost scenarios in US dollars, using 100 calls a month averaging three minutes. They are workload calculations rather than entry-plan prices. ROI assumes five administrative minutes saved per resolved call at $30 an hour. The original cost analysis explains allowances, overages, regional currencies, and credit use.

Upfirst’s $69.95 scenario produces the highest calculated ROI at 221%, followed by Goodcall at 175% and Ring Jenny at 113%. My AI Front Desk’s approximately 80% estimate depends on the availability of included credits. Aircall’s AI usage charge excludes the complete phone-system cost, so the research does not calculate its ROI. Sophiie and Dialzara also lack comparable total-cost figures in this workload.

What counts as a successfully handled call?

A call can sound successful while leaving the caller’s task unfinished. A receptionist might correctly describe opening hours but give an unavailable appointment time. It might capture a name yet send the message to the wrong person. Research needs a separate definition for each requested outcome.

Answer accuracy: was the information correct?

Accuracy checks the substance of an answer against the business’s information. Prices, service areas, availability, cancellation rules, and opening hours should agree with the approved business rules. When the answer is unavailable, the service needs a useful fallback, such as taking a question for staff or transferring the call.

The joint study by LunaCal and RingJenny.com uses 900 informational scenarios per platform for this measure. Ring Jenny records 876 correct answers, or 97.3%, and Smith.ai records 869, or 96.6%. These percentages have a different denominator from overall resolution, which covers 2,700 scenario calls per service.

Call resolution: was the requested task completed?

Resolution combines the outcome appropriate to the caller’s intent. A pricing question needs the correct information; an appointment request needs the calendar action; a transfer needs the intended connection. A greeting, a long conversation, or a positive-sounding closing is insufficient evidence that the task was completed.

Ring Jenny resolves 2,503 of 2,700 scenarios in the source research. Smith.ai resolves 2,476, and Upfirst resolves 2,425. For an evaluation inside your business, record the initial request and inspect the resulting message, calendar entry, or transfer record. This gives the resolution rate a traceable basis.

Booking success: did the appointment reach the calendar?

A completed booking requires more than a time mentioned during the call. Check the service, date, time zone, duration, staff member, caller details, and actual event creation. Rescheduling should update the intended appointment and avoid leaving duplicate events.

Sending a scheduling link can be useful, but it creates another step for the caller. Track it separately from an appointment completed during the conversation. Upfirst’s appointment-setting service, for example, describes connecting to a calendar in real time. During a trial, verify the event itself and the confirmation received by the caller.

Transfers: who receives the caller, and what happens if they cannot answer?

A transfer to your employee and escalation to a provider’s staffed receptionists are different services. Establish whether the receptionist introduces the caller, passes along the reason for calling, and uses a fallback destination when the first person does not answer.

Smith.ai’s AI Receptionist describes intake, qualification, appointment booking, and live-agent support. Check which handling path and charges apply to the configuration you are considering. A successful connection to an unavailable voicemail box may still leave your customer waiting for help.

How should you interpret ease-of-use and setup scores?

The study’s usability score covers configuration, business-information entry, calendar connectivity, routing instructions, and everyday administration. Ring Jenny scores 9.4/10, Upfirst 9.1/10, and Rosie AI 9.0/10. These ratings describe the scenario framework’s assessment of operating the service.

Initial setup and ongoing maintenance deserve separate attention. A service can be quick to activate yet difficult to update when your hours change, a staff member leaves, or a location introduces a new booking rule. Ask the person who will maintain the receptionist to make those changes during the trial.

For businesses with multiple branches, evaluate whether information stays specific to the selected location. A correct answer for one branch can be wrong for another. ElevenLabs’ receptionist documentation describes separate receptionists for departments, locations, languages, or brands. Location-specific information and routing are useful evaluation criteria even when a different provider handles your calls.

Check daily administration too: can staff find a failed booking, correct an answer, change an escalation destination, and see whether the correction took effect? Those routine tasks determine how much supervision the service needs after launch.

Why response speed needs more than one number

Median response latency describes a typical conversational turn. A 95th-percentile figure describes the slower end of the distribution. In the source study, Ring Jenny’s modeled median is 0.82 seconds and its 95th percentile is 1.92 seconds. Dialzara’s corresponding figures are 1.34 and 3.28 seconds.

Those measures answer different questions. A service can respond promptly to a simple FAQ but pause while checking an appointment or retrieving customer information. Listen for long unexplained silences, interruptions, repeated questions, and whether it resumes correctly when the caller changes a detail.

Separate a quick acknowledgement from a useful answer. “Let me check” can keep a conversation moving, but it does not complete the calendar lookup. For a booking trial, measure both the first response and the time until the caller receives a confirmed appointment.

How to calculate AI answering service ROI for your business

The study calculates administrative ROI from the value of staff time released. Its workload is 100 calls each month, with five minutes of administrative work saved for each resolved call and labor valued at $30 per hour. Each resolved request therefore contributes $2.50 of administrative capacity.

Monthly administrative value = monthly calls × resolution rate × minutes saved per resolved call ÷ 60 × hourly labor value.

First-year ROI = (annual administrative value − annual service cost) ÷ annual service cost × 100.

A worked example using the research’s Ring Jenny scenario

Using 2,503 resolved scenarios out of 2,700, the monthly administrative value is approximately $231.76. At the research’s $109 monthly service assumption, annual administrative value is about $2,781.11 and annual subscription spend is $1,308. The difference is $1,473.11, producing approximately 113% first-year administrative ROI.

The calculation values staff capacity. To turn released time into a budget benefit, decide what staff will do with it: process work that would otherwise wait, reduce overtime, or handle more customer requests. Additional sales and retained customers need a separate calculation based on your own records.

The time-saving assumption can change the result

Keep the same call volume, resolution assumption, hourly rate, and $109 cost, but reduce time saved from five minutes to two. Each resolved request is then worth $1 of administrative time. Monthly capacity value falls to about $92.70, and the calculated annual ROI becomes approximately −15% before implementation or supervision costs.

This is a sensitivity calculation using the study’s assumptions, not another platform performance result. Measure the administrative work your staff currently does after a call and the work that remains after automation. That difference provides a more useful time-saving estimate than the call’s duration alone.

Add cost per completed request to the budget

Divide total monthly service spend by completed requests to see what successful handling costs. In the source’s 100-call scenario, that is approximately $1.18 per resolved request for Ring Jenny, $0.78 for Upfirst, and $0.91 for Goodcall, calculated from the unrounded resolution counts.

Use the complete bill for this calculation: subscription, overages, required phone seats, integration fees, and paid escalation. Track staff review time separately. A low headline subscription can become expensive if many calls still require staff to reconstruct the conversation or finish the task.

How to compare call, minute, and customer-based billing

Start with the workload rather than the cheapest plan. Record monthly calls, average duration, peak periods, repeat callers, and the share needing appointments or transfers. Then apply each provider’s allowance and overage rules to that same set of calls.

• Per-call billing: confirm what counts as a billable call, including spam, hangups, and very short conversations. Longer calls may have a different economic effect from short ones.
• Per-minute billing: check rounding increments and whether hold time, transfers, or other parts of the conversation are included.
• Unique-customer billing: identify how repeat callers are counted and how the service recognizes an existing customer.
• Credit-based billing: identify how voice, messages, and chat consume the allowance. Several activities may draw from the same credit balance.
• Phone-system dependencies: add required licenses, numbers, telephony usage, or integration costs before comparing totals.

The study’s 100-call workload equals 300 minutes only because average duration is fixed at three minutes. If the same 100 calls average six minutes, usage becomes 600 minutes. A per-minute plan may move into overages while a per-call plan behaves differently. Recalculate the bill at your expected workload and at a busier month.

A practical evaluation plan before routing customer calls

Use identical business information and call requests across the services on your shortlist. Define the expected outcome before making the call, then inspect the result afterwards.

1. Prepare the business information. Include hours, locations, service areas, approved prices, appointment rules, and staff escalation destinations. Identify questions that need a human answer.
2. Test ordinary requests. Ask a business-hours question, request a service price, leave a callback request, book an appointment, and ask for a named employee.
3. Introduce realistic complications. Change the requested date, interrupt an answer, correct a name, use a noisy connection, and ask a question missing from the business information.
4. Inspect the action. Check calendar entries, captured contact details, staff notifications, transfer destinations, and what the caller was told. Record partial completion separately from full completion.
5. Calculate the workload cost. Apply the observed call duration and handling pattern to allowances, overages, and any required add-ons.
6. Assign an owner. Decide who reviews exceptions, updates information, checks missed actions, and manages fallback routing during service interruptions.

Include data handling in the evaluation: ask where recordings and transcripts are stored, who can access them, how retention works, and how records can be deleted. Check the provider’s actual terms and controls for the information your callers will share.

For another view of product capabilities and trade-offs, see the Sophiie AI alternatives comparison. Use the original AI answering service research methodology when checking how the reported scores were defined.

Research scope and calculation basis

The October 2026 joint research by LunaCal and RingJenny.com assigns 2,700 simulated calls to each of ten services: Ring Jenny, Smith.ai, Upfirst, Rosie AI, RingCentral AI Receptionist, Goodcall, Sophiie AI, My AI Front Desk, Dialzara, and Aircall AI Voice Agent. Nine scenario categories receive 300 calls each per service.

The categories cover business hours, services and policies, ambiguous questions, appointment booking, rescheduling, contact capture, transfers, accents and background noise, and interruptions. The modeled buyer component uses 200 profiles, each evaluating five products, for 1,000 evaluations. Figures in this article refer to that published framework; the worked budget examples identify their assumptions separately.

Frequently asked questions

What should AI phone answering service research measure?

It should distinguish correct answers, completed requests, successful bookings, transfers, conversational timing, administration, and total cost. Each metric needs a clear definition and a checkable outcome, such as the calendar entry or transfer destination.

Which services lead the joint research?

Ring Jenny has the highest simulated call-resolution rate at 92.7%, booking success at 95.3%, and ease-of-use score at 9.4/10. Smith.ai leads transfer success at 97.7%. Upfirst leads modeled administrative ROI at 221% under the study’s 100-call monthly assumptions.

Can an AI answering service book appointments?

Some services can connect to a calendar and create an appointment during the call; others send a booking link or take a message. Check the relevant subscription, calendar integration, and resulting event before counting a call as a successful booking.

How do I estimate what an AI answering service will cost?

Apply monthly call volume, duration, repeat-caller frequency, and required actions to each plan’s allowances. Add overages, required phone licenses, paid escalation, and implementation costs. Compare the resulting bill with the number of requests completed and the staff work still required.

Can AI replace a human receptionist?

AI can handle defined answering, message-taking, booking, and routing tasks. Decide which conversations require staff judgment and give those callers a clear escalation path. Measure the follow-up work before deciding how responsibilities should change.

Where can I read the full AI answering service research?

The full joint AI phone answering service research by LunaCal and RingJenny.com contains the platform tables, scenario definitions, buyer-priority figures, pricing references, and ROI methodology.