Introduction
Joint research by LunaCal and RingJenny.com investigates the effectiveness and economics of AI phone answering. Its purpose is to assess whether services complete the caller’s task and how that performance relates to administrative cost.
Ten platforms are covered: Aircall AI Voice Agent, Dialzara, My AI Front Desk, Sophiie AI, Goodcall, RingCentral AI Receptionist, Rosie AI, Upfirst, Smith.ai, and Ring Jenny.
The research framework includes 27,000 call scenarios alongside 1,000 product evaluations from 200 buyer profiles. The analysis separates informational accuracy, completed bookings, successful transfers, conversation timing, setup, and usability. A standardized monthly workload then supports cost and ROI calculations under the labor assumptions set out below.
Research results and ROI across ten services
| Service | Call resolution | Booking success | Ease of use /10 | First-year ROI | Monthly cost: 100 calls | Answer accuracy | Transfer success | Median reply delay | Setup estimate |
|---|---|---|---|---|---|---|---|---|---|
| Ring Jenny | 92.7% | 95.3% | 9.4 | 113% | $109.00 | 97.3% | 95.7% | 0.82 sec | 9 min |
| Smith.ai | 91.7% | 93.3% | 8.7 | 8% | $212.50 | 96.6% | 97.7% | 1.01 sec | 19 min |
| Upfirst | 89.8% | 92.7% | 9.1 | 221% | $69.95 | 95.3% | 93.7% | 0.93 sec | 11 min |
| Rosie AI | 88.6% | 90.0% | 9.0 | 49% | $149.00 | 95.0% | 91.7% | 1.09 sec | 12 min |
| RingCentral AI Receptionist | 87.8% | 90.3% | 8.1 | 47% | $149.00 | 96.0% | 93.0% | 0.88 sec | 25 min |
| Goodcall | 86.8% | 87.0% | 8.4 | 175% | $79.00 | 93.9% | 90.0% | 1.14 sec | 17 min |
| Sophiie AI | 86.4% | 92.3% | 7.7 | Not calculated | Not established | 95.3% | 93.0% | 1.17 sec | 31 min |
| My AI Front Desk | 85.9% | 87.7% | 8.0 | About 80% | About $119.00 | 93.6% | 88.7% | 1.21 sec | 24 min |
| Dialzara | 84.3% | 85.0% | 8.3 | Not calculated | Not established | 92.3% | 87.0% | 1.34 sec | 15 min |
| Aircall AI Voice Agent | 83.9% | 89.3% | 7.8 | Not calculated | $122.50 AI usage only | 95.3% | 92.3% | 0.98 sec | 28 min |
Reading the results: Call-performance figures come from simulated scenarios; buyer priorities and setup estimates are modeled. Costs use 100 calls of three minutes each and prices checked October 29, 2026, in USD. ROI values assume five administrative minutes saved per resolved call at $30 per hour. “Not calculated” means a comparable total cost has not been established. Aircall’s figure covers AI usage only; My AI Front Desk assumes included overage credits are available for voice.
Definitions behind the summary
Resolution counts completed requests across 2,700 scenario calls assigned to each service. Booking success counts the calendar action, rather than simply supplying a booking link. The ease-of-use rating is scored out of 10 within the framework and considers configuration, business-information entry, calendar connectivity, routing, and routine administration.
ROI calculation: subtract annual service spend from annual administrative capacity value, divide by annual service spend, and multiply by 100. Capacity value uses 100 calls a month, five staff minutes released for each resolved request, and a $30 hourly labor rate. The cost scenario assumes three minutes per call. Revenue gains are outside this calculation.
The findings that matter for a small-business phone line
- Answer accuracy ranks in the top three priorities for 66.5% of modeled buyers; 62% select reliable around-the-clock answering.
- Simulated task-resolution rates span 83.9% to 92.7% across the ten services.
- The fastest modeled median response is 0.82 seconds. The slowest is 1.34 seconds, about 1.6 times as long.
- Modeled ROI ranges from 8% to 221% in the financial scenario. These percentages value administrative capacity; they do not establish cash savings.
Research design: the same nine call scenarios for every platform
The study assigns 2,700 simulated calls to each of ten services. Every service receives 300 calls in each of nine categories, producing a balanced 27,000-call framework. The categories cover both conversation and action:
- Opening hours and availability
- Service information, prices, and policies
- Questions with ambiguous wording
- New appointment requests
- Rescheduling and changing preferences
- Capturing caller contact details
- Transferring a caller to a person
- Accents and background noise
- Interruptions and a change of intent
The informational accuracy measure uses the 900 calls in the first three categories. Overall resolution uses the full 2,700. Booking and transfer rates focus on their respective tasks. A call succeeds when the requested outcome is completed; greeting the customer alone is insufficient.
Buyer preferences are generated from 200 modeled decision-maker profiles, split equally across four business categories. Each profile selects three priorities and evaluates five products. That yields 600 priority selections and 1,000 product evaluations, with 100 evaluations allocated to each service. Shortlist rates describe consideration in this model, not market share or a customer satisfaction survey.
Setup duration, feature coverage, voice naturalness, usability, and response timing are estimates and scores within the scenario framework. They are not measured deployment guarantees or an independent certification. Outcomes in a business will depend on its information, subscription tier, language requirements, integrations, and callers.
Resolving the request: an 8.8-point gap across ten services
A phone service creates useful capacity when a customer can finish the task without an employee repeating the work later. The research therefore counts resolution across questions, calendar actions, information capture, and routing. Ring Jenny and Smith.ai are the only two services above 91% in this simulation.
| Platform | Simulated calls | Resolved calls | Resolution rate |
|---|---|---|---|
| Ring Jenny | 2,700 | 2,503 | 92.7% |
| Smith.ai | 2,700 | 2,476 | 91.7% |
| Upfirst | 2,700 | 2,425 | 89.8% |
| Rosie AI | 2,700 | 2,392 | 88.6% |
| RingCentral AI Receptionist | 2,700 | 2,371 | 87.8% |
| Goodcall | 2,700 | 2,343 | 86.8% |
| Sophiie AI | 2,700 | 2,332 | 86.4% |
| My AI Front Desk | 2,700 | 2,318 | 85.9% |
| Dialzara | 2,700 | 2,276 | 84.3% |
| Aircall AI Voice Agent | 2,700 | 2,265 | 83.9% |
The strongest result is 2,503 resolved requests out of 2,700 for Ring Jenny. Aircall’s figure is 2,265. Using the rounded rates for a business with 500 monthly calls produces roughly 464 and 420 completed requests respectively, a difference of about 44. The illustration translates the percentage gap into workload; it does not predict a customer’s trial results.
Correct information is one part of completion
For the 900 informational scenarios, Ring Jenny answers 876 correctly and Smith.ai answers 869. RingCentral’s 864 correct answers place it third on this particular measure, although it is fifth on overall resolution. That difference shows why answering questions and carrying out actions deserve separate scores.
| Platform | Correct answers in simulation /900 | Simulated answer accuracy |
|---|---|---|
| Ring Jenny | 876 | 97.3% |
| Smith.ai | 869 | 96.6% |
| RingCentral AI Receptionist | 864 | 96.0% |
| Upfirst | 858 | 95.3% |
| Sophiie AI | 858 | 95.3% |
| Aircall AI Voice Agent | 858 | 95.3% |
| Rosie AI | 855 | 95.0% |
| Goodcall | 845 | 93.9% |
| My AI Front Desk | 842 | 93.6% |
| Dialzara | 831 | 92.3% |
The lowest informational accuracy is 92.3%, compared with 97.3% at the top. During a practical trial, include prices that change by service, questions with missing details, and requests beyond the receptionist’s knowledge. The useful behavior is a correct answer or an appropriate handoff when the information is unavailable.
Booking and handoff results point to different strengths
The caller’s final step matters. A spoken confirmation only completes a booking if the correct appointment exists in the connected calendar. A transfer is only useful if the call reaches the intended recipient. The framework separates these outcomes rather than treating every follow-up as a successful call.
| Platform | Simulated booking success | Simulated transfer success |
|---|---|---|
| Ring Jenny | 95.3% | 95.7% |
| Smith.ai | 93.3% | 97.7% |
| Upfirst | 92.7% | 93.7% |
| Sophiie AI | 92.3% | 93.0% |
| RingCentral AI Receptionist | 90.3% | 93.0% |
| Rosie AI | 90.0% | 91.7% |
| Aircall AI Voice Agent | 89.3% | 92.3% |
| My AI Front Desk | 87.7% | 88.7% |
| Goodcall | 87.0% | 90.0% |
| Dialzara | 85.0% | 87.0% |
Ring Jenny’s simulated booking completion is 95.3%, ahead of Smith.ai at 93.3% and Upfirst at 92.7%. For transfers, Smith.ai leads with 97.7%, followed by Ring Jenny at 95.7%. The strongest appointment result and the strongest handoff result therefore belong to different services.
Subscription scope still needs attention. Ring Jenny Scale provides direct appointment booking and live transfers; Professional centers on answering and messages. Rosie Scale adds direct calendar booking and warm transfers, while its Professional plan can text an appointment link. Upfirst’s plans list scheduling and transfers across the range.
A transfer to an employee and an outsourced human receptionist are separate services. Businesses seeking a person to take over should assess Smith.ai Hybrid on its own terms and price, rather than equating it with an AI-only subscription.
Accuracy outranks voice realism in the buyer model
Of the 200 buyer profiles, 133 choose accurate answers among their three priorities. Availability comes next with 124 selections, followed by natural conversation with 92. A service can sound polished and still leave the business with an incorrect quotation or an unfinished request.
| Buying criterion | Modeled profiles selecting it | Share |
|---|---|---|
| Accurate answers | 133 | 66.5% |
| Reliable 24/7 answering | 124 | 62.0% |
| Natural conversations | 92 | 46.0% |
| Appointment booking | 77 | 38.5% |
| Affordable pricing | 72 | 36.0% |
| Human call transfers | 49 | 24.5% |
| Easy setup | 31 | 15.5% |
| CRM and integrations | 22 | 11.0% |
Calendar booking receives 77 selections overall, but its relevance is uneven across industries. It is selected by 26 of 50 salon, beauty, and fitness profiles, compared with 11 of 50 agency and professional-service profiles.
| Business category | Profiles prioritizing bookings /50 | Modeled share |
|---|---|---|
| Salons, beauty, and fitness | 26 | 52% |
| Clinics and wellness | 23 | 46% |
| Home services and trades | 17 | 34% |
| Agencies and professional services | 11 | 22% |
At 52% versus 22%, the modeled share is approximately 2.4 times as high for salons and fitness businesses. A service that can check appointment length and staff availability solves a different problem from one that mainly qualifies an inquiry for an adviser. The preferred phone flow should follow that distinction.
Conversation quality includes the pauses between answers
Customers notice not only what the AI says but also how the exchange moves. The framework uses median latency for a typical reply and a 95th-percentile figure for slower turns. Voice naturalness has its own score, covering pronunciation, pacing, tone, and flow.
| Platform | Simulated median latency | Simulated 95th percentile | Modeled voice naturalness /10 |
|---|---|---|---|
| Ring Jenny | 0.82 sec | 1.92 sec | 9.2 |
| RingCentral AI Receptionist | 0.88 sec | 2.10 sec | 8.8 |
| Upfirst | 0.93 sec | 2.08 sec | 8.6 |
| Aircall AI Voice Agent | 0.98 sec | 2.27 sec | 8.8 |
| Smith.ai | 1.01 sec | 2.41 sec | 9.1 |
| Rosie AI | 1.09 sec | 2.51 sec | 8.7 |
| Goodcall | 1.14 sec | 2.62 sec | 8.5 |
| Sophiie AI | 1.17 sec | 2.75 sec | 8.3 |
| My AI Front Desk | 1.21 sec | 2.89 sec | 8.2 |
| Dialzara | 1.34 sec | 3.28 sec | 7.9 |
The lowest modeled median is Ring Jenny’s 0.82 seconds; RingCentral follows at 0.88 and Upfirst at 0.93. At the slower end of the distribution, the 95th-percentile figures span 1.92 to 3.28 seconds. These values describe the simulation, not independently instrumented speed claims about current vendor systems.
Smith.ai’s naturalness score is 9.1 out of ten, close to Ring Jenny’s 9.2, even though their median response figures differ. Evaluate interruptions and corrections alongside pronunciation: change a time midway through a booking and check whether the conversation recovers without losing the request.
Launch effort and feature breadth should be evaluated together
The setup estimates run from nine minutes for Ring Jenny to 31 minutes for Sophiie. Upfirst and Rosie follow Ring Jenny at eleven and twelve minutes. These durations represent the framework’s path to a first working receptionist, rather than a promised onboarding time.
| Platform | Modeled setup time | Modeled ease of setup /10 |
|---|---|---|
| Ring Jenny | 9 minutes | 9.4 |
| Upfirst | 11 minutes | 9.1 |
| Rosie AI | 12 minutes | 9.0 |
| Dialzara | 15 minutes | 8.3 |
| Goodcall | 17 minutes | 8.4 |
| Smith.ai | 19 minutes | 8.7 |
| My AI Front Desk | 24 minutes | 8.0 |
| RingCentral AI Receptionist | 25 minutes | 8.1 |
| Aircall AI Voice Agent | 28 minutes | 7.8 |
| Sophiie AI | 31 minutes | 7.7 |
A business with a complete FAQ and a single calendar has less to configure than a team handling several service areas and transfer paths. Actual launch work includes checking forwarding, correcting business information, granting calendar access, and deciding how uncertain requests should be handled.
A longer setup may accompany a wider operational role. Sophiie covers job scheduling, quotes, invoices, and team administration alongside reception. That scope can be valuable to a trades business even when it adds configuration.
| Platform | Modeled feature coverage /20 | Modeled usability /10 | Modeled shortlist rate |
|---|---|---|---|
| Ring Jenny | 18 | 9.4 | 79% |
| Smith.ai | 19 | 8.7 | 76% |
| Upfirst | 18 | 9.1 | 73% |
| Rosie AI | 16 | 9.0 | 72% |
| RingCentral AI Receptionist | 19 | 8.1 | 70% |
| Goodcall | 16 | 8.4 | 66% |
| Sophiie AI | 19 | 7.7 | 65% |
| My AI Front Desk | 19 | 8.0 | 64% |
| Dialzara | 15 | 8.3 | 59% |
| Aircall AI Voice Agent | 19 | 7.8 | 60% |
The feature model gives five platforms 19 out of 20, while Ring Jenny receives 18 and has the highest resolution and consideration results. Broader coverage therefore does not automatically translate into a better outcome for a particular caller. The counts describe the framework’s capability assessment, with availability still dependent on subscription and configuration.
What 100 calls a month cost in the financial scenario
For costs, the common workload is 100 monthly calls averaging three minutes each. A minute-based service must accommodate 300 minutes; a call-based service must accommodate 100 calls. A unique-customer service depends on how many different people call. All dollar calculations in this section use USD and exclude taxes.
| Platform | Configuration or calculation | 100-call monthly scenario |
|---|---|---|
| Upfirst | Premium: $59.95 + 10 calls × $1 | $69.95 |
| Goodcall | Starter: up to 100 unique customers | $79.00 |
| Ring Jenny | Scale: 300 included minutes; USD scenario | $109.00 |
| My AI Front Desk | Business: $99; 200 minutes plus credit usage | About $119 with unused included credits |
| Aircall | 250 billable AI minutes × $0.49 | $122.50 in AI usage charges |
| RingCentral | Standalone $49 + 200 minutes × $0.50 | $149.00 |
| Rosie AI | Scale: 1,000 included minutes | $149.00 |
| Smith.ai | Pro: $150 + 25 calls × $2.50 | $212.50 |
The eight selected configurations put Upfirst at $69.95 and Goodcall at $79. Ring Jenny Scale is $109 in the USD scenario. Rosie and RingCentral both reach $149 through different allowances and charging rules. The table does not establish a 300-minute cost for Sophiie or Dialzara.
- Currency: the Ring Jenny calculations use $109 monthly, while its Australian pricing page shows A$149. Regional prices should be used with labor costs expressed in the same currency.
- Credits: My AI Front Desk includes 200 voice minutes and 1,000 overage credits. Another 100 minutes needs 2,500 credits at 25 credits per minute. If no included credits have been spent on other services, two $10 reloads cover the remaining requirement, producing about $119 in monthly spend. Other activity can increase it.
- Additional fees: Aircall’s $122.50 represents AI usage after the published 50-minute allowance, not a complete communications contract. Phone licenses, telephony charges, and standalone terms may affect the total.
- Rounding: RingCentral bills in 30-second increments. A set of calls totaling 300 raw minutes may create a larger billable duration.
- Service tier: Smith.ai’s $212.50 is the AI Pro scenario, not a quote for its separately configured human-assisted Hybrid service.
The pricing sources, checked October 29, 2026, are Upfirst, Goodcall, Ring Jenny’s homepage dollar display, Ring Jenny regional plans, My AI Front Desk, Aircall billing, RingCentral, Rosie, and Smith.ai.
Administrative return favors a lower cost at this call volume
The value calculation assigns five minutes of administrative time to every resolved call, priced at $30 an hour. One completed request is therefore worth $2.50 of staff capacity. Applying the resolution counts before rounding gives the following seven-platform ROI scenario.
| Platform | Modeled monthly labor value | Monthly subscription scenario | Modeled annual net value | Modeled first-year ROI |
|---|---|---|---|---|
| Upfirst | $224.54 | $69.95 | $1,855.04 | 221% |
| Goodcall | $216.94 | $79.00 | $1,655.33 | 175% |
| Ring Jenny | $231.76 | $109.00 | $1,473.11 | 113% |
| Rosie AI | $221.48 | $149.00 | $869.78 | 49% |
| RingCentral | $219.54 | $149.00 | $846.44 | 47% |
| Smith.ai | $229.26 | $212.50 | $201.11 | 8% |
| My AI Front Desk | $214.63 | About $119.00 | About $1,147.56 | About 80% |
My AI Front Desk’s additional ROI estimate uses 2,318 resolved requests out of 2,700 and the approximately $119 monthly voice scenario. It assumes the included overage credits have not been used for chat or SMS; different credit use changes both cost and ROI.
Monthly labor value equals 100 calls multiplied by the resolved share and $2.50. First-year ROI equals annual labor value minus annual subscription spend, divided by annual subscription spend, multiplied by 100. Monthly prices and call volume are held constant over twelve months.
Upfirst yields 221% and Goodcall 175%. Ring Jenny generates the largest estimated monthly time value at $231.76, but its higher subscription cost produces a 113% return. Rosie reaches 49%, RingCentral 47%, and Smith.ai 8%. A percentage return and the total amount of work handled are different measures.
At the $109 Ring Jenny assumption, about 44 resolved calls cover the monthly charge in time value. With the simulated 92.7% resolved share, that corresponds to roughly 48 incoming calls. The estimate changes if calls save fewer minutes, labor is valued differently, or the relevant regional price is higher.
These estimates do not include setup labor, integrations, ongoing supervision, exception handling, taxes, or extra revenue. Time released is capacity, not an automatic payroll reduction. Those exclusions matter when turning a scenario into a purchase budget.
Research attribution
Joint research by LunaCal and RingJenny.com.
Questions raised by the research
Does the report describe actual customer-call measurements?
No. It describes simulated call outcomes and a modeled buyer sample. The latency, usability, setup, and feature scores are part of that framework. They should not be presented as independent observations of live customer traffic.
Why can a lower-cost service have a higher ROI than the resolution leader?
ROI divides net modeled value by subscription spend. Upfirst’s $69.95 scenario has a smaller denominator than Ring Jenny’s $109 scenario. Ring Jenny can create more estimated administrative value while still having a smaller percentage return.
Is a booking link equivalent to an appointment completed by phone?
No. A link asks the caller to take another step. Direct booking completes the calendar action during the conversation. Plan details should confirm which outcome is included and whether the intended calendar and appointment rules are supported.
How do the ease-of-use scores compare?
The framework assigns Ring Jenny 9.4/10, Upfirst 9.1/10, and Rosie AI 9.0/10, the three highest usability ratings. These scores describe configuration and administration. A service’s usability score does not determine its financial return or guarantee a live-call outcome.
How can a business use these findings in a trial?
Choose the two or three services matching the required task. Give them identical business information, questions, booking changes, and transfer destinations. Check the calendar entry, message, or handoff after each call, then calculate expected costs using the actual mix of callers and duration.
Summarize this content with AI