{"author":{"@type":"Person","name":"Alexander Bock","url":"https://abxk.ai/en/autor/alexander-bock/"},"categories":["AI Work"],"content":{"markdown":"The phone is still where many customers turn when something is urgent or complicated. A missed call can mean a lost order, a lost booking or a patient who goes elsewhere. AI phone agents promise to answer every call, day and night. They can handle a good share of those calls, within limits that you set before the first call.\n\nThis is part 2 of Level 2 in our AI Work learning path. In Level 1, you set up [an AI chatbot for customer service](/en/ai-work/learn/ai-chatbots-for-customer-service/). A phone agent follows the same principles, but it adds three challenges: speed, speech recognition and legal rules for calls. You will learn how a phone agent works, which calls it should take, how to design a call flow with a handover, what you must tell callers and how to measure the results.\n\n## How an AI phone agent works\n\nAn **AI phone agent**, also called a voice bot or voice agent, answers or makes phone calls and talks with the caller in natural language. Unlike the old phone menus (\"press 1 for sales\"), it understands free speech. Most systems chain three technologies:\n\n1. **Speech recognition**, also called speech-to-text, turns the caller's words into text.\n2. A **large language model** reads the text, decides what to do and writes the answer. It can look things up in a knowledge base and use connected tools, such as your calendar or CRM.\n3. **Speech synthesis**, also called text-to-speech, turns the answer into a voice.\n\nSome newer systems skip the text step and process audio directly. The principle stays the same: listen, decide, speak.\n\n![Diagram: the caller's speech passes through speech recognition, a language model with a knowledge base and tools, and speech synthesis; the answer goes back to the caller](/images/learn/en/ai-phone-agents-pipeline.webp \"The voice chain behind an AI phone agent. It runs again for every sentence the caller says.\")\n\nTwo problems weigh much more on the phone than in a chat. The first is **latency**: the delay between the end of the caller's sentence and the start of the answer. Every step in the chain adds time, and people are very sensitive to pauses. A study of conversations in ten languages found that speakers in all of them avoid talking at the same time and keep the silence between turns short (Source: Stivers et al., PNAS, 2009). If a bot pauses much longer than a person would, callers repeat themselves, talk over it or think the line is dead. A good system also handles **barge-in**: when the caller interrupts, the agent stops talking and listens.\n\nThe second problem is speech recognition itself. Background noise, poor lines, accents, dialects and unusual names cause errors, and an error here travels through the whole chain. If the system hears the wrong date, the model books the wrong date. A 2020 Stanford study tested the speech recognition of five large technology companies. On average, the systems made almost twice as many errors for Black speakers as for white speakers (Source: Koenecke et al., PNAS, 2020). The systems have improved since then, but the lesson remains: test with the real voices of your callers, not only with your own team.\n\nThe voices themselves often come from text-to-speech providers. Our AI Productivity comparison lists them in the area [Text-to-Voice](/en/ai-productivity/#area-text-to-voice).\n\n## Which calls an AI agent should take\n\n**Inbound** calls come in from customers. **Outbound** calls go out from your company. The table shows typical uses and what to watch in each case:\n\n| Use case | Direction | Example | What to watch |\n|---|---|---|---|\n| Answer common questions | Inbound | Opening hours, directions, order status | Answers only from approved, current content |\n| Route calls | Inbound | \"I have a question about my invoice\" goes to accounting | Callers must reach a person quickly |\n| Book and move appointments | Inbound | A patient moves a check-up to Friday | Read back date and time; sync with the calendar |\n| Take messages after hours | Inbound | Name, number and request for a callback | Promise only callbacks you can keep |\n| Reminders and confirmations | Outbound | \"Your appointment is tomorrow at 10 a.m.\" | Consent rules; an easy way to cancel or reach a person |\n| Sales calls to new contacts | Outbound | Cold calls with an AI voice | Strict consent rules and a high risk for your reputation |\n\nStart with inbound calls that come often and carry little risk. Outbound calls with an AI voice need more care, because the person on the other end did not choose to call you. The figure sorts common calls into three groups:\n\n![Three columns: good first uses such as opening hours, routing, booking and messages after hours; uses that need safeguards such as reminders, identity checks, payments and complaints; and calls to leave to people such as advice, emergencies, cold sales calls and decisions on refunds](/images/learn/en/ai-phone-agents-use-cases.webp \"Start with frequent, low-risk calls. Some calls should stay with people.\")\n\n## Design the call flow and the handover\n\nA **call flow** is the planned path of a call, from the greeting to the goodbye. Design it on paper before you configure any tool. Most good flows have four steps: greet, understand the request, handle it and confirm.\n\nThe most important part is the exit. A **handover** passes the call to a person. Good phone agents hand over early and with context. In a **warm transfer**, the person who takes over receives a short summary, so the caller does not have to repeat everything. In a **cold transfer**, the call is simply forwarded, and the caller starts again from the beginning.\n\n![Call flow: greet, understand the request, handle it, confirm; below, the reasons for a handover lead to a warm transfer or a callback outside opening hours; a note on emergencies at the bottom](/images/learn/en/ai-phone-agents-call-flow.webp \"A call flow with a clear exit: the agent handles simple requests and hands everything else over with a summary.\")\n\nFive rules keep callers from getting stuck:\n\n- **Always offer a way to a person**, and say how: \"Say 'team' at any time.\"\n- **Limit the attempts.** If the agent has not understood the request after two tries, it hands over.\n- **Read back critical details**: names, dates, times, amounts and phone numbers.\n- **Pass on the context** with every transfer or callback ticket.\n- **Log every outcome**: resolved, transferred, callback or hung up.\n\nLevel 3 goes deeper into [quality control and escalation paths](/en/ai-work/learn/human-in-the-loop-at-work/) for all kinds of AI work.\n\n{{\u003c callout type=\"example\" title=\"A greeting that does three jobs\" \u003e}}\n\"Hello, you've reached [practice name]. I'm the practice's AI assistant, and I can book or move appointments. This call is recorded to process your request. If you'd like to talk to our team, just say 'team' at any time.\"\n\nThe greeting says that an AI is speaking, mentions the recording and shows the way to a person. Have your legal team check the wording for your country before you use it.\n{{\u003c /callout \u003e}}\n\n## Tell callers it's an AI and ask before you record\n\nBeing open about AI is not only polite. In many places, it is the law.\n\n- **AI disclosure.** Since August 2, 2026, the EU AI Act requires that AI systems which talk with people directly are designed so that people learn they are dealing with an AI, unless this is obvious (Article 50).\n- **Outbound calls.** In the US, the Federal Communications Commission ruled in February 2024 that AI-generated voices count as \"artificial\" voices under the Telephone Consumer Protection Act. Calls with them fall under its rules on consent, identification and opt-out (Source: FCC, Declaratory Ruling, 2024, fcc.gov). In Germany, advertising calls to consumers require their prior express consent under the Act against Unfair Competition (UWG).\n- **Recording.** In Germany, recording someone's private conversation without consent can be a criminal offense (Section 201 of the Criminal Code). In the US, federal law requires the consent of one party to the call, but some states, such as California, require the consent of everyone on the call.\n- **Transcripts are personal data.** Under the GDPR, recordings and transcripts need a purpose, a legal basis, a retention period and clear information for callers.\n\nThis overview is not legal advice. Have your legal team or data protection officer check your setup before the first real call.\n\n## Sensitive industries: the example of healthcare\n\nMedical, dental and veterinary practices get many calls about appointments, so phone agents are common there. Our comparison lists tools for this in the area [Customer Service \u0026 Telephony](/en/ai-work/#area-customer-service). According to the vendors, Weave is a communications platform for practices with phones, texting and scheduling, and Confido Health offers voice AI agents that answer calls, schedule visits and update the practice software. Vida builds AI agents for calls, texts, email and chat in general, while Calilio and KrispCall are cloud phone systems with AI features:\n\n{{\u003c tools \"vida calilio krispcall weave confido-health\" \u003e}}\n\nHealthcare shows why sensitive industries need extra rules:\n\n- **Health data is specially protected.** Under the GDPR, it belongs to the special categories of personal data (Article 9). In the US, a practice that lets a vendor handle protected health information needs a business associate agreement under HIPAA.\n- **No medical advice.** The agent books, moves and reminds. Questions about symptoms, medication or test results go to qualified staff.\n- **Emergencies come first.** If a caller says words like \"chest pain\" or \"can't breathe\", the agent points them to the emergency number and connects them to staff at once.\n- **Collect only what the task needs.** A booking needs a name, a way to identify the patient and a time, not the full medical history.\n\nThe same care applies to banks, insurers and law firms: identity checks, no binding advice and a quick way to a person.\n\n## Measure what callers experience\n\nPhone agents produce plenty of numbers, and some of them flatter. The most common one is the **containment rate**: the share of calls handled without a person. It counts every call that never reached your team, including callers who gave up and callers who had to call again. The example shows the gap:\n\n![Stacked bar of 100 example calls: 52 resolved by the agent, 7 called again later, 11 hung up unresolved, 30 handed to a person; the containment rate covers the first three groups, 70 percent, while only 52 percent were really resolved](/images/learn/en/ai-phone-agents-containment.webp \"Example values: a high containment rate can hide callers who gave up or called again.\")\n\nIn the example, 70 percent of calls are \"contained\", but only 52 percent were really resolved. Look at these metrics together:\n\n| Metric | What it tells you | Trap |\n|---|---|---|\n| Resolution rate | Share of calls where the request was really solved | Needs checking: listen to samples or ask one question at the end |\n| Containment rate | Share of calls without a person | Also counts callers who gave up; never use it alone |\n| Transfer rate and reasons | How often and why the agent hands over | A very low rate can mean the way to a person is hidden |\n| Hang-up rate | Callers who hang up before the end | Watch for hang-ups right after the greeting |\n| Repeat calls | Callers who call again about the same topic within a few days | The clearest sign of a \"contained\" but unsolved call |\n| Recognition errors | Wrong names, dates and numbers in transcripts | Check samples by accent and line quality |\n\n{{\u003c callout type=\"try\" title=\"Run 30 test calls before you go live\" \u003e}}\nBefore real customers call, ask colleagues to make 30 test calls from a script: common questions, a booking with a change of mind, a strong accent or dialect, background noise, an interruption in the middle of a sentence, a request for a person and an emergency word. Then read the transcripts. Did the agent understand, act correctly and hand over at the right moment? Fix the flow and repeat until the hard cases work.\n{{\u003c /callout \u003e}}","text":"The phone is still where many customers turn when something is urgent or complicated. A missed call can mean a lost order, a lost booking or a patient who goes elsewhere. AI phone agents promise to answer every call, day and night. They can handle a good share of those calls, within limits that you set before the first call.\nThis is part 2 of Level 2 in our AI Work learning path. In Level 1, you set up an AI chatbot for customer service. A phone agent follows the same principles, but it adds three challenges: speed, speech recognition and legal rules for calls. You will learn how a phone agent works, which calls it should take, how to design a call flow with a handover, what you must tell callers and how to measure the results.\nHow an AI phone agent works An AI phone agent, also called a voice bot or voice agent, answers or makes phone calls and talks with the caller in natural language. Unlike the old phone menus (\u0026ldquo;press 1 for sales\u0026rdquo;), it understands free speech. Most systems chain three technologies:\nSpeech recognition, also called speech-to-text, turns the caller\u0026rsquo;s words into text. A large language model reads the text, decides what to do and writes the answer. It can look things up in a knowledge base and use connected tools, such as your calendar or CRM. Speech synthesis, also called text-to-speech, turns the answer into a voice. Some newer systems skip the text step and process audio directly. The principle stays the same: listen, decide, speak.\nThe voice chain behind an AI phone agent. It runs again for every sentence the caller says. Two problems weigh much more on the phone than in a chat. The first is latency: the delay between the end of the caller\u0026rsquo;s sentence and the start of the answer. Every step in the chain adds time, and people are very sensitive to pauses. A study of conversations in ten languages found that speakers in all of them avoid talking at the same time and keep the silence between turns short (Source: Stivers et al., PNAS, 2009). If a bot pauses much longer than a person would, callers repeat themselves, talk over it or think the line is dead. A good system also handles barge-in: when the caller interrupts, the agent stops talking and listens.\nThe second problem is speech recognition itself. Background noise, poor lines, accents, dialects and unusual names cause errors, and an error here travels through the whole chain. If the system hears the wrong date, the model books the wrong date. A 2020 Stanford study tested the speech recognition of five large technology companies. On average, the systems made almost twice as many errors for Black speakers as for white speakers (Source: Koenecke et al., PNAS, 2020). The systems have improved since then, but the lesson remains: test with the real voices of your callers, not only with your own team.\nThe voices themselves often come from text-to-speech providers. Our AI Productivity comparison lists them in the area Text-to-Voice.\nWhich calls an AI agent should take Inbound calls come in from customers. Outbound calls go out from your company. The table shows typical uses and what to watch in each case:\nAI Phone Agents and Voice Bots Use case Direction Example What to watch Answer common questions Inbound Opening hours, directions, order status Answers only from approved, current content Route calls Inbound \u0026ldquo;I have a question about my invoice\u0026rdquo; goes to accounting Callers must reach a person quickly Book and move appointments Inbound A patient moves a check-up to Friday Read back date and time; sync with the calendar Take messages after hours Inbound Name, number and request for a callback Promise only callbacks you can keep Reminders and confirmations Outbound \u0026ldquo;Your appointment is tomorrow at 10 a.m.\u0026rdquo; Consent rules; an easy way to cancel or reach a person Sales calls to new contacts Outbound Cold calls with an AI voice Strict consent rules and a high risk for your reputation Start with inbound calls that come often and carry little risk. Outbound calls with an AI voice need more care, because the person on the other end did not choose to call you. The figure sorts common calls into three groups:\nStart with frequent, low-risk calls. Some calls should stay with people. Design the call flow and the handover A call flow is the planned path of a call, from the greeting to the goodbye. Design it on paper before you configure any tool. Most good flows have four steps: greet, understand the request, handle it and confirm.\nThe most important part is the exit. A handover passes the call to a person. Good phone agents hand over early and with context. In a warm transfer, the person who takes over receives a short summary, so the caller does not have to repeat everything. In a cold transfer, the call is simply forwarded, and the caller starts again from the beginning.\nA call flow with a clear exit: the agent handles simple requests and hands everything else over with a summary. Five rules keep callers from getting stuck:\nAlways offer a way to a person, and say how: \u0026ldquo;Say \u0026rsquo;team\u0026rsquo; at any time.\u0026rdquo; Limit the attempts. If the agent has not understood the request after two tries, it hands over. Read back critical details: names, dates, times, amounts and phone numbers. Pass on the context with every transfer or callback ticket. Log every outcome: resolved, transferred, callback or hung up. Level 3 goes deeper into quality control and escalation paths for all kinds of AI work.\n“A greeting that does three jobs\n\u0026ldquo;Hello, you\u0026rsquo;ve reached [practice name]. I\u0026rsquo;m the practice\u0026rsquo;s AI assistant, and I can book or move appointments. This call is recorded to process your request. If you\u0026rsquo;d like to talk to our team, just say \u0026rsquo;team\u0026rsquo; at any time.\u0026rdquo;\nThe greeting says that an AI is speaking, mentions the recording and shows the way to a person. Have your legal team check the wording for your country before you use it.\nTell callers it\u0026rsquo;s an AI and ask before you record Being open about AI is not only polite. In many places, it is the law.\nAI disclosure. Since August 2, 2026, the EU AI Act requires that AI systems which talk with people directly are designed so that people learn they are dealing with an AI, unless this is obvious (Article 50). Outbound calls. In the US, the Federal Communications Commission ruled in February 2024 that AI-generated voices count as \u0026ldquo;artificial\u0026rdquo; voices under the Telephone Consumer Protection Act. Calls with them fall under its rules on consent, identification and opt-out (Source: FCC, Declaratory Ruling, 2024, fcc.gov). In Germany, advertising calls to consumers require their prior express consent under the Act against Unfair Competition (UWG). Recording. In Germany, recording someone\u0026rsquo;s private conversation without consent can be a criminal offense (Section 201 of the Criminal Code). In the US, federal law requires the consent of one party to the call, but some states, such as California, require the consent of everyone on the call. Transcripts are personal data. Under the GDPR, recordings and transcripts need a purpose, a legal basis, a retention period and clear information for callers. This overview is not legal advice. Have your legal team or data protection officer check your setup before the first real call.\nSensitive industries: the example of healthcare Medical, dental and veterinary practices get many calls about appointments, so phone agents are common there. Our comparison lists tools for this in the area Customer Service \u0026amp; Telephony. According to the vendors, Weave is a communications platform for practices with phones, texting and scheduling, and Confido Health offers voice AI agents that answer calls, schedule visits and update the practice software. Vida builds AI agents for calls, texts, email and chat in general, while Calilio and KrispCall are cloud phone systems with AI features:\nVidaCustomer Service \u0026amp; Telephony · AI at the coreAI phone agents that answer calls, SMS, email and web chat around the clock and schedule appointments. CalilioCustomer Service \u0026amp; Telephony · Software with AI featuresCloud telephony and call center software with AI features and virtual numbers from more than 100 countries. KrispCallCustomer Service \u0026amp; Telephony · Software with AI featuresCloud phone system with calls, SMS, virtual numbers from more than 100 countries and an AI assistant. WeaveCustomer Service \u0026amp; Telephony · Software with AI featuresCommunications platform for medical, dental and veterinary practices: phone, texting, scheduling and payments. Confido HealthCustomer Service \u0026amp; Telephony · AI at the coreVoice AI agents for healthcare providers: answer calls, schedule visits and update the practice software. Short descriptions from our tool comparison, based on the vendors' information — not tested by ABXK.\nHealthcare shows why sensitive industries need extra rules:\nHealth data is specially protected. Under the GDPR, it belongs to the special categories of personal data (Article 9). In the US, a practice that lets a vendor handle protected health information needs a business associate agreement under HIPAA. No medical advice. The agent books, moves and reminds. Questions about symptoms, medication or test results go to qualified staff. Emergencies come first. If a caller says words like \u0026ldquo;chest pain\u0026rdquo; or \u0026ldquo;can\u0026rsquo;t breathe\u0026rdquo;, the agent points them to the emergency number and connects them to staff at once. Collect only what the task needs. A booking needs a name, a way to identify the patient and a time, not the full medical history. The same care applies to banks, insurers and law firms: identity checks, no binding advice and a quick way to a person.\nMeasure what callers experience Phone agents produce plenty of numbers, and some of them flatter. The most common one is the containment rate: the share of calls handled without a person. It counts every call that never reached your team, including callers who gave up and callers who had to call again. The example shows the gap:\nExample values: a high containment rate can hide callers who gave up or called again. In the example, 70 percent of calls are \u0026ldquo;contained\u0026rdquo;, but only 52 percent were really resolved. Look at these metrics together:\nAI Phone Agents and Voice Bots (2) Metric What it tells you Trap Resolution rate Share of calls where the request was really solved Needs checking: listen to samples or ask one question at the end Containment rate Share of calls without a person Also counts callers who gave up; never use it alone Transfer rate and reasons How often and why the agent hands over A very low rate can mean the way to a person is hidden Hang-up rate Callers who hang up before the end Watch for hang-ups right after the greeting Repeat calls Callers who call again about the same topic within a few days The clearest sign of a \u0026ldquo;contained\u0026rdquo; but unsolved call Recognition errors Wrong names, dates and numbers in transcripts Check samples by accent and line quality ▸Run 30 test calls before you go live\nBefore real customers call, ask colleagues to make 30 test calls from a script: common questions, a booking with a change of mind, a strong accent or dialect, background noise, an interruption in the middle of a sentence, a request for a person and an emergency word. Then read the transcripts. Did the agent understand, act correctly and hand over at the right moment? Fix the flow and repeat until the hard cases work.\n","words":1915},"date":"2026-10-14","description":"How AI phone agents work, which calls they should take, how to design handover, AI disclosure and recording consent, and which metrics show real results.","image":"https://abxk.ai/images/cover/ai-phone-agents.webp","json":"https://abxk.ai/en/ai-work/learn/ai-phone-agents/index.json","kind":"page","language":"en","learning_path":{"category":"ai-work","level":2,"level_name":"Intermediate"},"modified":"2026-10-14","related_tools":["Vida","Calilio","KrispCall","Weave","Confido Health"],"section":"ai-work","tags":["voice-ai","phone-agents","customer-service","call-center"],"title":"AI Phone Agents and Voice Bots","translations":[{"json":"https://abxk.ai/de/ai-work/learn/ai-phone-agents/index.json","language":"de","url":"https://abxk.ai/de/ai-work/learn/ai-phone-agents/"}],"url":"https://abxk.ai/en/ai-work/learn/ai-phone-agents/"}