Key takeaways
- An AI receptionist answers the calls that come in. An AI voice agent is the broader technology that can also place calls. The distinction matters because the cost, the consent rules and the risk all change with the direction of the call.
- Quo analyzed 16.7 million missed business calls over three months and found 69% received no callback within 48 hours. Most missed inbound calls simply disappear, which is the strongest case for an AI receptionist.
- Invoca analyzed over 70 million phone calls and found 42% of phone leads convert on the call itself, a conversion rate that only counts if somebody picks up.
- At the time of writing, published pricing ranges from about $0.11 per minute on a build-your-own voice platform, to $79 to $249 a month for a packaged receptionist, to roughly $11 for a three-minute call on a live answering service. A full-time receptionist covers 40 of the 168 hours in a week, roughly 24%.
- Outbound AI calls need prior express consent in the US, since the FCC ruled on February 8, 2024 that AI voices count as artificial under the TCPA. EU rules require AI disclosure from August 2, 2026. Start with inbound, add outbound only for people who asked to hear from you.
An AI receptionist answers the calls that come in. An AI voice agent is the wider technology behind it, and it can also place calls, which is the part that changes the economics, the law and the risk.
Search results blur the two, because vendors use both names for overlapping products, and that confusion costs people money. A professional services firm that wants its phone answered buys a sales-dialer platform it never needed. A sales team that wants fast follow-up on web leads buys an answering product that cannot dial out. This article separates the two jobs, shows what the data says about each, puts real prices side by side, and ends with a five-question way to choose.
The inbound and outbound split, and why it decides everything
Think of it as a question of who picks up the phone first.
In an inbound call, the other person dialed you. They have a reason to talk, they chose the moment, and the only way you can fail is by not answering well. That is the receptionist job.
In an outbound call, you dialed them. You chose the moment, they may not want the call, and the rules about who you may call and how become the main constraint. That is where a voice agent that places calls does its work.
AI receptionist versus AI voice agent, side by side
The names overlap in vendor marketing. This comparison uses the job each one does.
| AI receptionist (inbound) | AI voice agent (outbound) | |
|---|---|---|
| Who starts the call | The caller dials your number. | You, or the software on your behalf, dial the prospect. |
| Typical trigger | A ringing phone, any hour of the day. | A form submission, an opened proposal, a missed appointment, a pipeline stage change. |
| What success looks like | The call is answered, the inquiry is captured or booked, nothing is lost. | The right person is reached quickly and a meeting or next step is agreed. |
| Main legal question (US) | Call recording consent and telling callers it is AI. | Prior express consent for AI-voiced calls to mobile and residential numbers under the TCPA. |
| Common pricing model | Monthly plan, often by unique caller or call volume. | Per minute of conversation, or a managed service. |
| Biggest risk | A wrong answer, or no easy route to a human. | Calling people who never consented, and wasting dials on cold lists. |
Deligatr analysis, based on the sources listed at the end of this article.
One more distinction that trips buyers up: some tools are a finished product, and some are a toolkit. Packaged receptionist products such as Goodcall are ready to configure and priced simply. Platforms such as Vapi and Retell give you the building blocks to assemble a voice agent yourself, and you pay for each component by the minute. A managed service builds, runs and tunes it for you. All three can do inbound, and the platforms and managed services can also do outbound. Which one you need depends on who will maintain it, not on the label.
What an AI receptionist does, and the missed-call problem it solves
An AI receptionist picks up every inbound call, answers from the knowledge you gave it, takes details, books appointments into your calendar, and hands the call to a person when it should. Its value is entirely a function of how many calls you currently miss and what they were worth.
Start with an honest note on the most famous statistic in this space. The figure that only about 38% of small business calls are answered by a live person comes from a 2016 study by 411 Locals, a 30-day look at 85 businesses across 58 industries. It is widely repeated as if it were recent. It is a small, old sample, so treat it as a sign of how common the problem is, not as a current benchmark.
Newer evidence points the same way. Quo analyzed 16.7 million missed business calls from its customers over three months and defined a callback as any outbound call to the caller's number within 48 hours.
What happens to a missed business call
Share of 16.7 million missed calls, by whether the caller got a callback within 48 hours.
69% No callback within 48 hours
31% Called back within 48 hours
Nearly all of these happened within 24 hours.
Quo, Small Business Callback Statistics (16.7 million missed calls, three months)
By industry, the callback rate was highest in healthcare at 41%, then professional services at 33%, consumer goods at 22%, and non-profits at 21%. Quo does not publish a sample size or margin of error for each industry, so the ordering is more reliable than the exact percentages.
Why do missed calls vanish? Because callers rarely wait. In a CallRail survey of 1,000 US consumers, 82% said they would call a competitor after an unanswered call, 78% said they had abandoned a business over one, and only 42% said they leave a voicemail. Note that these are stated behaviors, not observed ones, and that voicemail habits vary a lot by country: Quo cites a YouGov finding that only 6% of UK consumers are likely to leave a voicemail.
What consumers say they do when a call goes unanswered
CallRail survey of 1,000 US consumers, 2025. Stated behavior, not observed behavior.
CallRail survey of 1,000 US consumers, as reported by OnCrew's missed-call statistics review
The reason this matters is what a live conversation is worth. Invoca analyzed more than 70 million phone calls and 600 million minutes of conversation, and found that 42% of phone leads convert on the call, meaning they end with a completed appointment or sale.
Share of phone leads that convert on the call, by industry
Conversion means a completed appointment or sale during the call. Invoca, 70 million calls.
Invoca, phone lead conversion report (70 million+ calls, 600 million minutes)
Put those together and the inbound case is plain. A call that reaches a person converts at a high rate, and a call that does not reach a person usually does not come back. If you run a firm where one new client is worth thousands, an unanswered phone is a leak you can measure, and an AI receptionist exists to plug it. A missed call from someone ready to buy rarely gets a second attempt, which is why the checklist below starts with answering every call.
What an AI voice agent does that a receptionist does not
An AI voice agent that places calls works on the other side of the problem. Its best-supported job is speed.
The classic evidence is a Harvard Business Review study by Oldroyd, McElheran and Elkington, published in 2011, which audited 2,241 companies and about 1.25 million leads. It reported that firms contacting a lead within an hour were nearly seven times as likely to qualify it as firms that waited even one more hour, and more than 60 times as likely as firms that waited 24 hours or longer. The study is more than a decade old and the exact multiples will differ today, but its core point has held up: most companies respond far too slowly, and the first minutes are worth the most. We found the multiples reported consistently in secondary summaries rather than in the paywalled article text, so read them as indicative.
A voice agent that calls a web lead within a minute of the form submission is the direct answer to that finding. The jobs it does well share one feature: the person has already asked to hear from you.
- Instant follow-up on a form submission or quote request. The lead is warm and the call is expected.
- Calling when a prospect opens your proposal. The interest is live and the timing is the point.
- Appointment reminders and no-show recovery. The relationship exists and the call is useful.
- Re-engaging an existing pipeline. The contact already knows who you are.
The job it should not do is dial strangers. On February 8, 2024, the FCC issued a unanimous Declaratory Ruling that AI-generated voices count as "artificial" under the Telephone Consumer Protection Act, so calls using them need prior express consent. For marketing calls that generally means written consent, and the ruling covers calls to both wireless and residential landline numbers, which includes the mobile phones many business contacts answer. TCPA violations carry statutory damages of $500 per violation, which a court can increase to as much as $1,500 for willful or knowing ones. Our cold calling analysis covers the full rules and why AI voice is the wrong tool for cold lists.
The overlap: when one tool covers both
The honest answer is that the technology underneath is the same. A speech-to-text layer, a language model, and a text-to-speech layer, connected to a phone line, can answer a call or place one. That is why the platforms charge per minute regardless of direction, and why many vendors offer both.
The overlap is useful in a few cases:
- A call-back flow. The agent takes a missed or after-hours inbound call, then places a scheduled call back the next morning to a caller who asked for one.
- A booking loop. The agent answers inbound inquiries and also calls people who requested a quote on your site, using the same calendar and the same qualification script.
- A shared record. Every call, in either direction, lands in one CRM with a transcript, so a person picking up the conversation has context.
What the overlap does not remove is the difference in rules. An agent that does both is still two things legally: an answering service for callers who chose to call, and an outbound dialer that needs consent. Configure and audit them separately, even when one product does both.
What each one costs, per call
Prices below were taken from each vendor's published pricing page at the time of writing. They change often and exclude setup, your own time, and any add-ons, so treat them as a guide to the shape of the market and check each page before you decide.
The three models charge in different units, which makes a direct comparison tricky. To compare them, we express each as a rough cost per handled call, with the assumptions stated.
Rough cost per handled call, by pricing model
Illustrative. Each bar uses a different billing unit, so read the caption under each one.
A 3-minute call at roughly $0.11 per minute, Retell's worked example with GPT-4.1. Vapi's all-in range of $82 to $129 per 1,000 minutes implies about $0.25 to $0.39 for the same call.
$249 per month for 500 unique callers, with unlimited minutes. Cost per unique caller at full use.
$79 per month for 100 unique callers. Cost per unique caller at full use, and it rises if you use less of the allowance.
$720 per month for 200 minutes of live receptionist time, which is $3.60 a minute, so about $10.80 for a 3-minute call.
Vendor pricing pages (Retell, Vapi, Goodcall, Ruby), at the time of writing. Costs per call are our arithmetic from published prices.
Two readings follow from that chart. First, the gap between software and people is large: about thirty times between a three-minute call on a voice platform and the same call on a live answering service. Second, software pricing is cheap enough that the decision rarely turns on the subscription itself. It turns on setup quality, whether the agent gets answers right, and what happens to the calls it cannot handle.
For a sense of what you are replacing, the US Bureau of Labor Statistics puts the median receptionist wage at $18.27 an hour, or $38,010 a year, before benefits, payroll costs, or cover for holidays and sickness. A full-time receptionist works about 40 hours of the 168 in a week, so even a well-staffed front desk leaves around three-quarters of the week with nobody answering. The BLS also projects receptionist employment to decline 2% from 2025 to 2035, though it expects about 105,100 openings a year from turnover. The comparison that matters for most firms is not a full-time receptionist against software. It is today's reality (missed calls, voicemail, slow call-backs) against an agent that answers every call.
Deligatr sells DeliVoice, which does both inbound and outbound, so weigh this section accordingly. Pricing is on the DeliVoice page. If a vendor will not tell you how it charges before a sales call, that tells you something too.
What to look for when evaluating an AI voice agent
Whatever you buy, these criteria separate effective products from impressive demos:
- Voice quality and latency. Test with a real call before committing. Stilted pauses lose callers.
- Knowledge training. Can you train it on your services, pricing, FAQs and team? A generic agent gives generic answers.
- Calendar integration. Booking during the call is the key feature. If the caller has to be sent a link afterwards, you have lost the moment.
- CRM logging. Every call should create or update a record with a transcript and summary. Deligatr syncs with any existing CRM at no extra cost, and DeliHub is free for two seats.
- Escalation rules. What happens when a caller needs a human: transfer, take a message, or schedule a callback?
- Outbound capability, and consent. Can the agent also place calls? If so, use it only for people who have asked to hear from you, such as a form submission, a quote request or an opened proposal. Opening a document is not consent to be phoned by an AI voice. Get prior express consent first, because the TCPA covers AI-voiced calls to mobile and residential numbers. This is general information, not legal advice.
Setup reality: what the first week actually involves
The technology is the easy part. What decides results is the preparation, and most of it has nothing to do with software. Here is what the first week should include, whichever product you pick.
- Decide what the agent may and may not do. Write down what counts as a qualified inquiry, what it can book on its own, and which calls must go straight to a person: complaints, existing key clients, anything urgent.
- Build the knowledge base from real calls. Pull the 20 questions your team answers most often, with the answers you actually give, including pricing and what you do not do. A thin knowledge base is the most common reason an agent sounds unhelpful.
- Connect the calendar and the CRM. Booking and logging are where most of the value sits. An agent that cannot see your real availability can only take messages.
- Set the human fallback. Choose who receives a transfer, when, and what happens at 2am. No call should end without a booked meeting or a logged callback request.
- Write the disclosure and recording line. The first sentence should say it is an AI assistant, and, where recording applies, that the call may be recorded.
- Test with real calls, then read the transcripts every day for the first week. Fix wrong answers when they appear. Most improvement comes from reading what actually happened, not from adjusting settings in advance.
Measure four things from day one: the share of calls answered, the share the agent handled without a human, the booking rate, and how many transfers were correct. Those four numbers tell you whether it is earning its place.
The rules and the trust problem
Inbound answering avoids the consent question that outbound faces, because the caller started the call. It brings two other obligations.
Recording. Under federal law, a participant in a call may record it without telling the other party, but a number of states require every party to consent, and California, Florida, Illinois, Pennsylvania and Washington are among those commonly listed. Source lists vary between nine and thirteen states depending on how hybrid rules are counted, and rules for calls crossing state lines are decided case by case. If a caller or your business is in an all-party state, announce the recording at the start and keep proof.
Telling people it is AI. In the EU, Article 50 of the AI Act, enforceable since August 2, 2026, requires the provider of an AI system that interacts directly with people to inform them they are dealing with AI, unless that is obvious from the circumstances. The Digital Omnibus left that date unchanged. Fines under the Act can reach 15 million euros or 3% of worldwide annual turnover, whichever is higher, and the obligations apply to systems whose output is used in the EU, including from non-EU companies. If you answer calls from EU residents, build the disclosure in.
Then there is the human factor. In a SurveyMonkey study of 2,017 US adults in December 2025, most said they would rather talk to a person.
79%
of Americans strongly prefer interacting with a human over an AI agent
SurveyMonkey, 2,017 US adults, December 10-11, 2025
The same survey found 89% believe companies should always offer a human option. Comfort depends heavily on the task: 65% said they would use AI to order food and drinks and 59% to return an item, while 69% were uncomfortable taking medical advice from AI and 68% uncomfortable taking investment advice from it. The lesson for a receptionist deployment is practical rather than philosophical. Use the agent for the routine layer (hours, directions, booking, intake), keep a fast human route for everything else, and say clearly that it is AI. A voice that sounds human matters less than a call that does not get stuck.
Choosing: five questions
Answer these in order. Each one narrows the decision.
1. Who starts most of your important calls? If prospects and clients mostly ring you, you need an inbound receptionist. If your revenue depends on you reaching people first, you need outbound capability. Most small and mid-sized firms are in the first group.
2. Where is the revenue leaking today? If calls go to voicemail while you are in meetings, the Quo and CallRail data describe your problem, and inbound answering comes first. If web leads sit untouched for hours, the HBR speed-to-lead finding describes it, and outbound follow-up comes first.
3. Have the people you would call actually agreed to AI-voiced calls? If the answer is no, do not run outbound AI calls to them. Consent is the line. Inbound has no equivalent question.
4. How many calls will it handle a month? At a few dozen calls, a packaged plan or a managed service is simple. At hundreds, per-minute and per-caller pricing diverge and the unit economics matter. Run your own number, not a vendor's example.
5. What must always stay human? List the calls a person must take: complaints, high-value existing clients, anything emotionally charged or legally sensitive. Then check that the product can transfer them quickly and cleanly. If it cannot, it is not ready for your phone.
If you answered "inbound" to the first two questions, an AI receptionist is the right starting point. If your answers point to outbound, start with consented follow-up, never cold lists. If most of your inquiries arrive through your website rather than the phone, a chat agent such as DeliChat may fit better than voice.
Book a consultation and we will map your inbound and outbound call flows, and tell you plainly which one to start with.
Book a ConsultationWhere DeliVoice fits
DeliVoice answers inbound calls 24/7, qualifies callers, books meetings into your calendar, logs every call with a transcript, and can place outbound follow-up calls triggered by events such as a form completion or an opened proposal. It connects to the CRM you already use. It is part of The Deli Suite, alongside the other Deligatr tools.
We would give you the same advice whether or not you buy it from us: start with inbound, measure what you were missing, then add consented outbound follow-up once the first layer is working.
Sources
- Quo, Small Business Callback Statistics, 16.7 million missed calls, three months
- Invoca, Phone lead conversion report, 70 million+ calls and 600 million minutes
- OnCrew, Missed Call Statistics: Verified Numbers, Real Sources, source and year for the 411 Locals study (2016) and the CallRail survey (2025, 1,000 US consumers)
- Oldroyd, McElheran and Elkington, The Short Life of Online Sales Leads, Harvard Business Review, 2011
- US Bureau of Labor Statistics, Receptionists, Occupational Outlook Handbook
- Goodcall, Pricing, Ruby, Pricing, Retell AI, Pricing, Vapi, Pricing, accessed at the time of writing
- Cornell Legal Information Institute, 47 U.S.C. ยง 227, TCPA statutory damages
- Harvard Journal of Law & Technology, FCC Cracks Down on AI-Powered Robocalls, the February 8, 2024 ruling
- Wiley, FCC Extends Regulatory Reach Over AI, scope of the ruling to wireless and residential numbers
- Baker Botts, EU AI Act Article 50: Transparency Obligations Go Live
- SurveyMonkey, Customer Service Statistics: Humans vs AI, December 2025
- Recording Law, US Recording Laws by State, federal one-party baseline and all-party states
FAQ
What is an AI receptionist?
An AI receptionist is software that answers your business phone for you. It picks up every inbound call, handles common questions, takes details, books appointments, and passes anything it can't handle to a person. It does not usually place calls of its own. Most are sold as a ready-made product on a monthly plan, which is what separates them from the build-your-own voice platforms.
What is the difference between an AI receptionist and an AI voice agent?
Direction and scope. An AI receptionist is a product built for one job, answering inbound calls. An AI voice agent is the broader technology: any software that holds a spoken phone conversation, which can be configured to answer calls, place calls, or both. Vendors use the two terms loosely, so ask any provider directly whether it handles inbound, outbound, or both before you compare prices.
How much does an AI receptionist cost?
It depends on how the vendor charges. At the time of writing, Goodcall's published plans run from $79 to $249 per month and charge by unique caller (100 to 500 included, then $0.50 per extra caller) rather than by minute. Voice platforms that you configure yourself charge by the minute: Retell lists voice agents from $0.07 to $0.31 per minute, with a worked example at about $0.11, and Vapi charges a $0.05 per minute platform fee plus separate speech, language-model and telephony costs. Live answering services that put a person on the line cost far more per call: Ruby's published plans work out to roughly $3.45 to $5.00 per minute. Prices change, so check each vendor's pricing page.
Can an AI receptionist book appointments?
Yes, if it is connected to your calendar. The agent reads your real availability, offers open slots during the call, and confirms the booking before the caller hangs up. Without a calendar connection it can only take a message or a callback request, so confirm the integration exists for the calendar you actually use before you buy.
Do I need both inbound and outbound AI voice?
Most businesses should start with inbound only. Answering the calls you already receive needs no consent from the caller because they chose to call you, and the revenue you are missing is already measurable. Add outbound later, and only for people who have asked to hear from you: form submissions, quote requests, opened proposals. Outbound AI calls to people who have not consented bring the TCPA into play in the US.
Will callers know it is AI?
Many will realize quickly, and increasingly you are required to tell them. In the EU, Article 50 of the AI Act has required since August 2, 2026 that people be told when they are interacting with an AI system unless that is obvious from the circumstances. Good practice everywhere is to say so in the first sentence of the call. SurveyMonkey's December 2025 survey of 2,017 US adults found 79% strongly prefer a human to an AI agent, so a disclosure line and a fast route to a person matter more than a convincing voice.
Do I have to tell callers that the call is being recorded?
In many cases, yes. Federal law allows a participant to record a call without telling the other party, but a number of states require every party to consent, and California, Florida, Illinois, Pennsylvania and Washington are among those most often listed. Lists differ by source, and rules for calls that cross state lines are not settled, so the safe default is to announce recording at the start of every call. Ask your lawyer to confirm for the states you serve.
