AI voice agent statistics: real results from 19 deployments
Real AI voice agent statistics from 19 live deployments: answer rates, appointment costs and conversion numbers, case by case.
## AI voice agent statistics from live deployments These **AI voice agent statistics** come from real DeepAgent deployments on inbound calls and outbound follow-up. An Italian multi-utility increased its customer service success rate from 57.1% to 70.7% within three months of live tuning. In a comparison run by Bludental, a dental network with 70+ clinics, the team achieved a 28% lead conversion rate with an AI agent compared to 22% with human staff. For automotive group Paci Auto, the operating cost per booked workshop appointment fell from EUR 1.29 in month one to EUR 0.66 in month three. Every figure cited on this page comes from an individual customer deployment during a specific operational period. ## How to read these voice AI benchmarks Comparing phone campaigns requires clear terminology. Teams measure progress against distinct milestones, as in the Unipol Cerignola case study: * Attempt: A call triggered by the dialer, regardless of whether a carrier connects it, a voicemail picks up, or a person answers. * Answered call or conversation: A call where a person answers and talks with the agent (each case sets its own threshold). * Goal or success rate: The percentage of calls that achieve the primary objective specified for that workflow, such as completing a technical diagnostic ticket, booking a date, or completing a transfer. * Conversion on conversations vs attempts: Conversion calculated on connected calls measures dialogue effectiveness, whereas conversion calculated on total dialer attempts accounts for list reachability. * Cost per outcome: The net telephony and software cost required to deliver one confirmed business event, such as a booked visit or a signed release instruction. Operational requirements differ between inbound support desks and outbound recruitment campaigns. To understand how architectural models differ before reviewing raw figures, see our guide on what is a voicebot. ## Inbound calls: handling volumes, routing, and goal completion Inbound systems assist front-desk reception by routing calls, booking appointments in calendars, and resolving recurring account queries. These **voice AI benchmarks** show how inbound agents performed across retail, insurance, utilities, and clinics. | Customer / Organization | Period | Total Calls | Main Result | | :--- | :--- | :--- | :--- | | Agenzia Unipol Cerignola | 2 months | 725 calls | 93.2% handled correctly; 200 of 207 human transfers successful (96.6%) | | Italian multi-utility | Feb to Jun | 3,402 calls | 64.0% overall success rate; success rose from 57.1% in April to 70.7% in June | | Private medical centre | 3.5 months | 2,935 calls | 73.6% goal reach rate; failed calls dropped from 18.0% to 2.9% | | Playa Aruba | Late May to mid-Sep | 7,151 calls | 99.92% completed calls (7,145); 110 calls per day during July peak | | Gruppo Dirigì | Apr 3 to Jun 17, 2026 | 1,642 calls | 676 reservations/handoff goals reached (41%); 55.7s average duration | | SARA water treatment | 1 month | 2,260 calls | 100% immediate answer rate; yield grew from 45.5% to 76.5% in 4 weeks | | Paci Auto | 3 months | 3,094 calls | 431 confirmed appointments; 81.35% resolved without human escalation | In the Agenzia Unipol Cerignola implementation, the AI assistant Adriana managed 10 hours and 15 minutes of inbound conversation with a 99.6% transcript coverage rate across 725 calls. For Gruppo Dirigì, the voice agent handled 1,515.6 minutes of conversation across its Jesolo and Mister Pizza locations, keeping average call time to 56 seconds while capturing dinner reservations during rush periods. To discover how modern telephony stacks manage these workflows, read our breakdown of AI phone agent technology and operations. ## Outbound campaigns and where attempts go Outbound campaigns encounter gatekeepers, answering machines, and unanswered dials. In those campaigns, the **AI calling answer rate** depended on list quality, phone infrastructure, and calling hours. | Campaign / Organization | Period | Total Attempts | Main Result | | :--- | :--- | :--- | :--- | | Insurance recruiter | 1 week | 685 attempts | 40.15% conversation rate; 13.09% conversion to callback on conversations | | Aesthetic medicine centres | Aug 1 to Sep 11 | 4,872 attempts | 1,409 conversations (28.9%); 108 doctor consultations booked | | UltraSolar | 60 days | 19,362 attempts | 5,249 conversations (27.1%); 138 technical site visits booked | | Lunaga | Mar 9 to 26, 2026 | 10,133 attempts | 47.46% answer rate; 770 confirmed attendances (16.01% of answers) | | E-commerce logistics operator | 20 weeks | 18,364 attempts | 4,765 completed calls; 1,549 parcel release instructions secured | | HVAC brand | Oct 2025 to Mar 2026 | 2,928 attempts | ~50% useful answer rate vs ~10% digital forms; 278 NPS ratings collected | Examining call drops shows where outbound dials land. In the aesthetic medicine campaign, 71.1% of attempts ended without dialogue: 26.8% reached voicemail, 16.3% went unanswered, 14.6% were busy, and 12.1% were unreachable. In the B2B insurance recruiting campaign, 31.24% of dials went unanswered, 17.66% reached switchboards or voicemail, and 6.57% experienced technical connection failures, with 32 of the 36 qualified contacts between 10:00 and 12:00. In those campaigns, timing and follow-up rules shaped connection rates. In the logistics operation, 64.2% of contacts were reached at the first attempt, while 35.8% were reached only after the first attempt. In the UltraSolar campaign, calls lasting two minutes or longer converted at 46.94% into booked site inspections, whereas calls under two minutes converted at 0.91%. You can review structured retry strategies in our guide to lead follow-up automation cadences. ## Lead follow-up and qualification In those campaigns, teams configured fast outreach rules to follow up with digital leads shortly after submission before interest cooled down. | Deployment | Period / Scope | Volume Handled | Main Result | | :--- | :--- | :--- | :--- | | Bludental | 70+ clinic locations | 4,000 test leads/mo | 28% AI conversion vs 22% human; 65% answer rate vs 50% human | | Klio consumer credit | 2 months | 5,246 attempts | Answer rate rose from 52.4% to 60.6%; conversion rose from 5.80% to 13.55% | | VEGA inbound qualification | 3 months | 7,290 attempts | 3,030 conversations; 1,510 positive outcomes passed to sales | | Flor de Pacifico | 5.5 days | 207 calls | 37% answer rate; 18 qualified investor consultations (25% on answers) | | Centro Fitness gym chain | Sep 1 to 15, 2025 | 192 leads | Lead-to-appointment conversion reached 46.8%; no-show rate fell from 35% to 21% | | Collettivo Editoriale | Sample evaluation | 680 calls | 120 appointments booked; 54.55% conversion on completed calls | For Bludental, the AI workflow contacts leads on WhatsApp within five seconds of web submission and follows up by phone one minute later. For every 100 inbound leads processed through this sequence, 65 answered, 59 received an automated qualification score, and 28 booked clinic consultations. At Centro Fitness, average response time dropped from four to six hours under human reception to three minutes with an AI caller, shifting conversion from 30% to 46.8%. ## What moved the numbers over time In those deployments, prompt adjustments, vocabulary updates, and path corrections based on call recordings improved performance over the first month. * Multi-utility resolution: Inbound resolution started at 57.1% in April, climbing to 65.4% in May and 70.7% in June as common billing questions were refined in the knowledge base. * Agenzia Unipol Cerignola handling rate: The share of calls handled correctly (calls lasting over 30 seconds) increased from 63.4% in July to 93.2% in August, gaining 29.8 percentage points in one month. * SARA water treatment yield: The operational success rate began at 45.5% during its initial days and reached 76.5% by the end of week four. * Medical centre failed calls: Calls classified as failed dropped from 18.0% in June to 2.9% in September, a 15.1 percentage point reduction in dropped inquiries. * Klio conversion trajectory: Lead-to-appointment conversion moved from 5.80% in month one to 13.55% in month two as conversation paths were simplified. * Gym chain speed-to-lead: In the 1-15 September 2025 period, first response time dropped from 4-6 hours to 3 minutes, with no-shows falling from 35% to 21%. Reviewing audio recordings, edge-case misunderstandings, and handoff triggers is necessary to produce these results. To see how engineering teams evaluate conversations, read our guide on AI call analytics and transcripts. ## Costs per outcome actually observed Evaluating customer expenditures against completed outcomes shows true operational return. Paci Auto saw software and telephony expenses per confirmed dealership workshop appointment drop from EUR 1.29 in month one to EUR 0.66 in month three. In consumer finance, Klio reduced its direct cost per booked advisor appointment from EUR 10.97 to EUR 5.86 while operating at an infrastructure cost of EUR 0.50 per conversation minute. For Collettivo Editoriale, 680 calls generated 120 author consultations at a total campaign cost of EUR 268.90, working out to EUR 0.395 per dial attempt and EUR 2.24 per booked consultation. In high-volume parcel management, the Italian logistics operator spent EUR 5,401 over 20 weeks to resolve 1,549 warehouse holds. This resulted in an average cost of EUR 1.13 per completed call and EUR 3.49 per positive delivery release, compared to an estimated EUR 8,200 to EUR 10,250 for equivalent human phone shifts. These unit costs stem from individual operational setups, telephony routes, and call durations. To project your own economics, consult our breakdown on estimating voice agent pricing and ROI alongside our analysis of measuring voice AI return on investment. ## How to build your own benchmark in a pilot Running a controlled pilot delivers clear before-and-after data while keeping scope manageable. Use these six steps to establish baseline metrics: 1. Select a single use case: Choose one repetitive inbound workflow, such as clinic appointment booking, or one outbound trigger, such as web form qualification. 2. Define one unambiguous primary outcome: Establish a binary pass-or-fail goal for each call, such as writing a confirmed booking into Google Calendar or updating a CRM field. 3. Document your pre-pilot baseline: Record your current answer rate, staff hours spent per 100 leads, response latency, and cost per qualified outcome before routing calls to AI. 4. Maintain strict tracking definitions: Track conversion against total dials and completed conversations separately to evaluate both reach and script performance. 5. Review transcripts weekly: Audit the first 100 call transcripts to identify where callers hesitate, where speech recognition fails on regional terms, and where prompts can be shortened. 6. Calculate unit cost per delivered milestone: Compare the direct cost per completed outcome against your historical baseline to verify financial viability. For a structured testing methodology, review our walkthrough on how to test an AI voice agent. To see how these calls flow into pipeline forecasting, see our article on RevOps pipeline measurement for AI calls. When evaluating infrastructure suppliers, refer to our criteria on how to choose an AI voice agent provider. ## DeepAgent platform plans DeepAgent provides monthly plans for businesses deploying conversational phone agents, with a 17% discount on annual billing. All prices exclude VAT: * Starter: EUR 99 per month * Pro: EUR 299 per month * Growth: EUR 500 per month * Scale: EUR 799 per month * Enterprise: Custom quote Teams can evaluate the platform with 30 minutes free, no card, by visiting the DeepAgent pricing page. ## Frequently asked questions ### What answer rate can an AI voice agent get? In those campaigns, contact rates depended on the audience and call direction. On inbound deployments such as Playa Aruba and SARA water treatment, answer rates reached 99.92% and 100%. On outbound campaigns, results varied by workflow in those deployments: a 27.1% completed call rate for UltraSolar, 40.15% completed conversations for the insurance recruiter, a 47.46% answer rate for Lunaga, and a 65% answer rate for Bludental. ### Do AI voice agents convert better than humans? In the head-to-head comparison run by dental chain Bludental across 70+ clinics, the AI agent achieved a 28% end-to-end conversion rate compared to 22% for human operators. In that test, the AI agent also secured a 65% answer rate compared to 50% for human agents, reaching 30% more leads in the test. ### How long until results improve after launch? Calibration on real call transcripts produces measurable gains within weeks. SARA water treatment increased operational yield from 45.5% to 76.5% across its first 4 weeks. Agenzia Unipol Cerignola raised correctly handled requests from 63.4% in July to 93.2% in August. The multi-utility improved its goal success rate each month, moving from 57.1% in April to 65.4% in May and 70.7% in June. ### How much does an appointment cost with an AI voice agent? In those campaigns, costs per outcome depended on list quality and conversation duration. Paci Auto saw its cost per appointment drop from EUR 1.29 in month 1 to EUR 0.66 in month 3. Klio reduced cost per appointment from EUR 10.97 to EUR 5.86 in consumer lending. Collettivo Editoriale booked author consultations at EUR 2.24 per appointment, while the Italian logistics operator resolved parcel holds at EUR 3.49 per positive release. ### Are these numbers guaranteed? No. These figures reflect real results recorded by specific companies during defined operating periods. Conversions and answer rates depend on lead freshness, CRM integration depth, telephony reputation, script design, and regular review of call transcripts. ## Validate voice AI metrics on your own lines Reviewing real deployment figures sets practical expectations for speed, answer rates, and booking costs. You can test your own call logic and measure initial performance across live scenarios directly on the DeepAgent platform. Talk to sales → · Create your agent →