{"course":{"slug":"ai-for-small-business","title":"AI for a Small Business","url":"https://addaly.com/learn/ai-for-small-business","level":"beginner","track":"work"},"lesson":{"slug":"phone-and-voice","title":"The phone: what a voice assistant can and cannot do yet","url":"https://addaly.com/learn/ai-for-small-business/phone-and-voice","position":36,"of":88,"minutes":9,"answer":"Voice adds latency, interruption and irrecoverable mis-hearing to every text problem, so it suits narrow bounded jobs with no money in them.","keyIdea":"Voice adds latency, interruption and irrecoverable mis-hearing to every text problem, so it suits narrow bounded jobs with no money in them, and for most small businesses transcribed voicemail with a summary solves the real problem, which is failing to call back."},"body":{"format":"markdown","text":"## Why voice is harder than text\n\nEvery problem in this course is present on the phone, plus three that are specific to speech.\n\n**Latency.** In text, a two-second pause is nothing. On a call it is a silence, and people start talking over it. A voice system needs to transcribe, think and speak in well under a second to feel natural, and the engineering that achieves this is the main reason voice costs more than text.\n\n**Interruption.** Real callers cut in. Handling that (stopping mid-sentence and listening) is called barge-in, and systems that lack it feel robotic in a way that no amount of good writing fixes.\n\n**Accent, noise and names.** The transcription lesson applies with more force, because a caller cannot see and correct what was heard. Names and addresses are exactly what a booking call is made of and exactly what recognition handles worst.\n\n## Where it works today\n\nNarrow, bounded jobs with short answers and no money in them.\n\n- Opening hours, address, directions, whether you are open on a holiday.\n- Taking a message and a callback number.\n- Confirming or cancelling an existing appointment.\n- Triage: understanding what the call is about, then routing it.\n\nWhere it fails: anything requiring an address to be captured accurately, anything where the caller is upset, anything commercially significant, and anything in an accent or language the system handles poorly, which is a great deal of the world.\n\n## Cost, and what it replaces\n\nHosted voice agents typically run somewhere between USD 0.05 and 0.30 a minute all in. A hundred five-minute calls a month is roughly USD 25 to 150.\n\nCompare that against what it displaces. If calls currently go to voicemail and half never call back, almost anything is an improvement. If a person answers now, a voice agent is a downgrade for most callers and you should be very sure before installing one.\n\nThere is a middle option that is much better than its reputation: **voicemail with automatic transcription and a summary**, sent to your phone. It costs a few dollars a month or is free in several phone systems, has none of the latency problems, cannot mishandle an angry caller, and solves the real problem, which for most small businesses is not answering the phone but forgetting to call back.\n\nStart there. It fixes more than it should for what it costs.\n\n## Disclosure\n\nTell callers they are speaking to an automated system, in the first sentence. Some jurisdictions require it (several US states have bot-disclosure laws, and the EU AI Act adds transparency duties as it phases in) and the reason to do it anyway is that people work out something is wrong within about fifteen seconds and feel deceived rather than merely inconvenienced.\n\nOne line: *This is an automated assistant. I can help with opening hours and appointments, or put you through to someone.*\n\nAnd note the recording question from the transcription lesson applies here too, with more force. A voice system that keeps recordings is recording every caller.\n\n## Numbers, always confirmed\n\nThe single rule that prevents most of the damage: any number, name, address or date captured by voice is read back and confirmed, and appears in a text message afterwards.\n\n\"That's a booking for Thursday the 14th at 3pm, and I'll text you a confirmation now.\" Ten seconds, and it converts the highest-risk part of the call into something both parties can check.\n\n## An honest position\n\nVoice is improving quickly and it is the area where the gap between a demonstration and daily reliability is widest. A demo is a quiet room, one clear speaker, a scripted request. Your callers are in a car park with a toddler, saying a street name the system has never heard.\n\nWait until you have a specific, narrow, bounded job, and until the free option of transcribed voicemail has demonstrably run out of road.\n\n## What to do with the calls you already miss\n\nBefore any of this, count. For one week, note how many calls go unanswered and how many of those people call back.\n\nMost small businesses have never measured it, and the number is usually larger than expected. It also decides everything else: if 40% of calls go unanswered and half of those never return, the opportunity is enormous and the cheapest fix (a transcribed voicemail that reaches your phone, and a rule that every message is returned the same day) captures most of it.\n\nOnly if that is already in place does a voice agent have a job to do.\n\n## Route rather than resolve\n\nIf you do install one, the safest design is the same containment idea as the website bot: let it understand and route, not decide.\n\n\"Is this about a new booking, an existing booking, or something else?\" then transfer, take a message, or send a booking link by text. It never quotes a price, never confirms availability from memory, and never captures an address by voice. That is a narrow job and it is one voice systems do adequately today."},"check":{"question":"A garage is considering a voice agent to take bookings, including customer names, vehicle registrations and addresses. What is the strongest technical objection?","options":[{"text":"Voice agents cannot yet handle interruptions naturally, so callers who cut in will find the experience frustrating and abandon the call.","correct":false,"tempting":"Names a real limitation that degrades the experience, while the booking data itself would still be captured wrongly on calls that ran smoothly to the end."},{"text":"Registrations and addresses are exactly the content speech recognition handles worst, and a caller cannot see and correct what was heard as they can with text.","correct":true},{"text":"Per-minute costs make booking calls uneconomic compared with a booking link, particularly for a garage where each call may run to several minutes of back-and-forth about availability.","correct":false,"tempting":"Argues on price, when at a few tens of dollars a month the cost is not what makes this unsuitable and a cheaper agent would be equally wrong."},{"text":"Disclosure requirements mean the garage must announce the system is automated, which will cause some callers to hang up before making a booking.","correct":false,"tempting":"Treats a one-line disclosure as the obstacle, when it costs seconds and the substantive problem is the accuracy of what gets written down."}]},"prev":{"slug":"reviews-and-reputation","title":"Reviews: asking, answering, and the line you must not cross","url":"https://addaly.com/learn/ai-for-small-business/reviews-and-reputation"},"next":{"slug":"working-across-languages","title":"Serving customers in a language you do not read","url":"https://addaly.com/learn/ai-for-small-business/working-across-languages"},"authorship":"Founder and series editor: Dr Mustafa Mun. Written by AI models to a written standard. Exam keys checked blind by a second model: 3,665 of 3,666 agreed. Academic review invited.","review":{"status":"Draft, open for academic review"},"cite":"Addaly (2026) 'The phone: what a voice assistant can and cannot do yet', in AI for a Small Business. Edited by M. Mun. Addaly. Available at: https://addaly.com/learn/ai-for-small-business/phone-and-voice (Accessed: 13 October 2026).","licence":{"name":"CC BY-NC-SA 4.0","url":"https://creativecommons.org/licenses/by-nc-sa/4.0/"},"reader":{"schema":1,"title":"The phone: what a voice assistant can and cannot do yet","courseTitle":"AI for a Small Business","module":{"index":4,"slug":"customers-and-the-front-desk","title":"Customers, from first message to resolved complaint"},"answer":null,"keyIdea":"Voice adds latency, interruption and irrecoverable mis-hearing to every text problem, so it suits narrow bounded jobs with no money in them, and for most small businesses transcribed voicemail with a summary solves the real problem, which is failing to call back.","hero":{"kind":"plate","image":null,"alt":null,"caption":null},"blocks":[{"t":"h2","id":"why-voice-is-harder-than-text","runs":[{"text":"Why voice is harder than text"}]},{"t":"p","runs":[{"text":"Every problem in this course is present on the phone, plus three that are specific to speech."}]},{"t":"p","runs":[{"text":"Latency.","b":true},{"text":" In text, a two-second pause is nothing. On a call it is a silence, and people start talking over it. A voice system needs to transcribe, think and speak in well under a second to feel natural, and the engineering that achieves this is the main reason voice costs more than text."}]},{"t":"p","runs":[{"text":"Interruption.","b":true},{"text":" Real callers cut in. Handling that (stopping mid-sentence and listening) is called barge-in, and systems that lack it feel robotic in a way that no amount of good writing fixes."}]},{"t":"p","runs":[{"text":"Accent, noise and names.","b":true},{"text":" The transcription lesson applies with more force, because a caller cannot see and correct what was heard. Names and addresses are exactly what a booking call is made of and exactly what recognition handles worst."}]},{"t":"h2","id":"where-it-works-today","runs":[{"text":"Where it works today"}]},{"t":"p","runs":[{"text":"Narrow, bounded jobs with short answers and no money in them."}]},{"t":"ul","items":[[{"text":"Opening hours, address, directions, whether you are open on a holiday."}],[{"text":"Taking a message and a callback number."}],[{"text":"Confirming or cancelling an existing appointment."}],[{"text":"Triage: understanding what the call is about, then routing it."}]]},{"t":"p","runs":[{"text":"Where it fails: anything requiring an address to be captured accurately, anything where the caller is upset, anything commercially significant, and anything in an accent or language the system handles poorly, which is a great deal of the world."}]},{"t":"h2","id":"cost-and-what-it-replaces","runs":[{"text":"Cost, and what it replaces"}]},{"t":"p","runs":[{"text":"Hosted voice agents typically run somewhere between USD 0.05 and 0.30 a minute all in. A hundred five-minute calls a month is roughly USD 25 to 150."}]},{"t":"p","runs":[{"text":"Compare that against what it displaces. If calls currently go to voicemail and half never call back, almost anything is an improvement. If a person answers now, a voice agent is a downgrade for most callers and you should be very sure before installing one."}]},{"t":"p","runs":[{"text":"There is a middle option that is much better than its reputation: "},{"text":"voicemail with automatic transcription and a summary","b":true},{"text":", sent to your phone. It costs a few dollars a month or is free in several phone systems, has none of the latency problems, cannot mishandle an angry caller, and solves the real problem, which for most small businesses is not answering the phone but forgetting to call back."}]},{"t":"p","runs":[{"text":"Start there. It fixes more than it should for what it costs."}]},{"t":"h2","id":"disclosure","runs":[{"text":"Disclosure"}]},{"t":"p","runs":[{"text":"Tell callers they are speaking to an automated system, in the first sentence. Some jurisdictions require it (several US states have bot-disclosure laws, and the EU AI Act adds transparency duties as it phases in) and the reason to do it anyway is that people work out something is wrong within about fifteen seconds and feel deceived rather than merely inconvenienced."}]},{"t":"p","runs":[{"text":"One line: "},{"text":"This is an automated assistant. I can help with opening hours and appointments, or put you through to someone.","i":true}]},{"t":"p","runs":[{"text":"And note the recording question from the transcription lesson applies here too, with more force. A voice system that keeps recordings is recording every caller."}]},{"t":"h2","id":"numbers-always-confirmed","runs":[{"text":"Numbers, always confirmed"}]},{"t":"p","runs":[{"text":"The single rule that prevents most of the damage: any number, name, address or date captured by voice is read back and confirmed, and appears in a text message afterwards."}]},{"t":"p","runs":[{"text":"“That’s a booking for Thursday the 14th at 3pm, and I’ll text you a confirmation now.” Ten seconds, and it converts the highest-risk part of the call into something both parties can check."}]},{"t":"h2","id":"an-honest-position","runs":[{"text":"An honest position"}]},{"t":"p","runs":[{"text":"Voice is improving quickly and it is the area where the gap between a demonstration and daily reliability is widest. A demo is a quiet room, one clear speaker, a scripted request. Your callers are in a car park with a toddler, saying a street name the system has never heard."}]},{"t":"p","runs":[{"text":"Wait until you have a specific, narrow, bounded job, and until the free option of transcribed voicemail has demonstrably run out of road."}]},{"t":"h2","id":"what-to-do-with-the-calls-you-already-miss","runs":[{"text":"What to do with the calls you already miss"}]},{"t":"p","runs":[{"text":"Before any of this, count. For one week, note how many calls go unanswered and how many of those people call back."}]},{"t":"p","runs":[{"text":"Most small businesses have never measured it, and the number is usually larger than expected. It also decides everything else: if 40% of calls go unanswered and half of those never return, the opportunity is enormous and the cheapest fix (a transcribed voicemail that reaches your phone, and a rule that every message is returned the same day) captures most of it."}]},{"t":"p","runs":[{"text":"Only if that is already in place does a voice agent have a job to do."}]},{"t":"h2","id":"route-rather-than-resolve","runs":[{"text":"Route rather than resolve"}]},{"t":"p","runs":[{"text":"If you do install one, the safest design is the same containment idea as the website bot: let it understand and route, not decide."}]},{"t":"p","runs":[{"text":"“Is this about a new booking, an existing booking, or something else?” then transfer, take a message, or send a booking link by text. It never quotes a price, never confirms availability from memory, and never captures an address by voice. That is a narrow job and it is one voice systems do adequately today."}]}],"check":{"question":[{"text":"A garage is considering a voice agent to take bookings, including customer names, vehicle registrations and addresses. What is the strongest technical objection?"}],"options":[{"text":[{"text":"Voice agents cannot yet handle interruptions naturally, so callers who cut in will find the experience frustrating and abandon the call."}],"correct":false,"tempting":[{"text":"Names a real limitation that degrades the experience, while the booking data itself would still be captured wrongly on calls that ran smoothly to the end."}]},{"text":[{"text":"Registrations and addresses are exactly the content speech recognition handles worst, and a caller cannot see and correct what was heard as they can with text."}],"correct":true,"tempting":null},{"text":[{"text":"Per-minute costs make booking calls uneconomic compared with a booking link, particularly for a garage where each call may run to several minutes of back-and-forth about availability."}],"correct":false,"tempting":[{"text":"Argues on price, when at a few tens of dollars a month the cost is not what makes this unsuitable and a cheaper agent would be equally wrong."}]},{"text":[{"text":"Disclosure requirements mean the garage must announce the system is automated, which will cause some callers to hang up before making a booking."}],"correct":false,"tempting":[{"text":"Treats a one-line disclosure as the obstacle, when it costs seconds and the substantive problem is the accuracy of what gets written down."}]}],"order":[3,1,2,0],"why":null},"prev":{"slug":"reviews-and-reputation","title":"Reviews: asking, answering, and the line you must not cross","minutes":9},"next":{"slug":"working-across-languages","title":"Serving customers in a language you do not read","minutes":9}}}