AI receptionist software answers phone calls, books appointments, qualifies leads, and routes urgent requests without a human picking up the phone. Off-the-shelf options (Smith.ai, Ruby, Dialzara, Goodcall) cost $200 to $1,000 per month and handle basic call answering, message taking, and appointment scheduling through pre-built integrations with common calendaring and CRM tools. Custom AI receptionist systems cost $40,000 to $100,000 to build but handle the complex scheduling logic, multi-provider routing, industry-specific intake, and deep system integrations that off-the-shelf tools cannot.
The build-vs-buy decision depends on three factors: call volume (under 300 calls per month favors SaaS, over 500 favors custom), scheduling complexity (single-provider, single-service businesses work with SaaS; multi-provider, multi-service businesses with insurance verification or emergency routing need custom), and integration depth (if the AI receptionist needs to read from and write to your practice management system, EHR, or field service dispatch software in real time, off-the-shelf integrations rarely cover it).
What does off-the-shelf AI receptionist software actually do?
SaaS AI receptionist platforms handle the standard call flow: answer the phone with a greeting, determine the caller's intent, and take the appropriate action (book an appointment, take a message, transfer to a specific person, or provide basic information). They use speech-to-text to understand the caller, a language model to determine intent and generate responses, and text-to-speech to speak back. The voice quality has improved significantly since 2024; most callers cannot tell they are speaking with an AI system within the first 30 seconds.
Smith.ai is the most established player, offering a hybrid model with AI handling initial call screening and human receptionists available for complex calls. Pricing starts at $97.50 per month for 20 calls. Ruby offers a similar hybrid model starting at $235 per month for 50 calls. Dialzara is fully AI-powered (no human backup) starting at $29 per month. Goodcall targets small businesses with a simpler feature set starting at $59 per month.
These platforms integrate with common scheduling tools (Calendly, Acuity, Google Calendar), CRMs (HubSpot, Salesforce, Zoho), and communication platforms (Slack, email). They can answer FAQs from a knowledge base you configure, transfer calls to specific team members based on caller intent, and send post-call summaries via email or SMS.
Where does off-the-shelf AI receptionist software break?
Multi-provider scheduling is the first breaking point. A dental practice with 3 dentists and 2 hygienists needs the AI to check availability for the right provider type based on the procedure requested, account for procedure-specific time blocks (a crown prep takes 90 minutes, a cleaning takes 45, an emergency exam takes 30), verify whether the patient is new or returning (new patients need longer appointment slots), and check insurance eligibility before confirming. Off-the-shelf platforms book into a single calendar. They cannot query a practice management system to determine which provider is appropriate for which procedure at which time.
Emergency classification and routing is the second breaking point. An HVAC company needs the AI receptionist to distinguish between a routine service request (schedule for next available), an urgent repair (schedule same-day or next-day and notify the dispatcher), and an emergency (gas leak, no heat in winter, no AC in a heatwave with elderly residents) that requires immediate human escalation with the caller's address, unit type, and symptom description. The classification requires understanding the caller's description, asking follow-up questions based on the symptoms described, and making a judgment call about urgency. SaaS platforms can route based on keywords ("emergency" goes to the emergency line) but cannot perform the diagnostic conversation.
Legal intake is the third breaking point. A law firm's receptionist does not just book appointments. They perform a preliminary intake: what type of legal matter (personal injury, family law, criminal defense, estate planning), when did the incident occur (statute of limitations check), is there a conflict of interest (check against the firm's client list), and is this the type of case the firm handles? The intake information determines which attorney the consultation is booked with, what the consultation fee is (if any), and what documents the caller needs to bring. This is a multi-step classification and data collection conversation that requires domain-specific knowledge, not a simple scheduling call.
Deep system integration is the fourth breaking point. A medical practice needs the AI receptionist to check patient records in the EHR before confirming an appointment (is the patient due for a follow-up? Do they have outstanding lab results to review?). An HVAC company needs the AI to check the dispatch system for technician availability by service zone. A legal firm needs the AI to check the case management system for conflicts. These integrations require reading from and writing to proprietary systems that SaaS platforms do not support.
What does a custom AI receptionist system include?
The telephony layer handles call routing, hold, transfer, and conference. Twilio, Vonage, and Telnyx provide the infrastructure. The system needs to handle simultaneous calls (a busy practice receives 5 to 10 calls during peak hours), warm transfers (providing the receiving person with a summary of the conversation so far), and fallback routing (if the AI cannot resolve the call, transfer to a human with full context).
The speech-to-text layer converts caller speech to text in real time. Deepgram and AssemblyAI are the leading options for production systems, offering sub-300-millisecond latency and medical/legal vocabulary support. The total round-trip latency (caller speaks, STT transcribes, LLM generates response, TTS speaks it back) needs to stay under 1 second for the conversation to feel natural. Anything over 1.5 seconds and callers start saying "hello? Are you there?"
The conversation engine is the core: a language model with a structured prompt that defines the business's call handling rules, a knowledge base containing FAQs and service information, and tool-calling capability to check calendars, query systems, and book appointments during the conversation. The engine manages conversation state (remembering what the caller said 3 turns ago), handles interruptions (the caller changes their mind mid-sentence), and knows when to escalate to a human.
The text-to-speech layer converts the AI's response to spoken audio. ElevenLabs and PlayHT produce the most natural-sounding voices. Custom voice cloning allows the system to use a voice that matches the business's brand (a warm, professional tone for a medical practice; a friendly, energetic tone for a home services company). The voice quality matters: a robotic-sounding AI receptionist undermines the professionalism the business is trying to project.
The integration layer connects the conversation engine to the business's systems. For a dental practice: the practice management system (Dentrix, Eaglesoft, Open Dental) for patient records and scheduling. For HVAC: the field service management system (ServiceTitan, Housecall Pro, FieldEdge) for technician dispatch and job scheduling. For legal: the case management system (Clio, MyCase, PracticePanther) for conflict checking and matter intake. Each integration is bidirectional: the AI reads from the system during the call and writes back after the call completes.
How does the build-vs-buy math work by industry?
For HVAC, plumbing, and electrical companies, the average inbound call is worth $350 to $800 in revenue. Missing 5 calls per day during peak season (a reasonable estimate when 2 technicians are in the field and the office manager is handling walk-ins) costs $1,750 to $4,000 per day in lost revenue. Over a 90-day peak season, that is $157,500 to $360,000 in missed calls. An AI receptionist that captures even 50% of those missed calls pays for a custom build ($60,000 to $80,000) within the first season.
For dental practices, the economics are similar but the scheduling complexity is higher. A new patient appointment is worth $300 to $500 for the initial exam and X-rays, with a lifetime value of $3,000 to $8,000. A dental practice receiving 40 calls per day with 2 front desk staff misses calls during lunch, before/after hours, and when both staff members are helping patients in the office. An AI receptionist handling after-hours and overflow calls captures 10 to 15 additional new patient appointments per month. At $400 average first-visit revenue, that is $4,000 to $6,000 per month in additional revenue from captured calls alone.
For law firms, the math is even more decisive. A personal injury intake call that goes to voicemail is almost certainly lost: the caller will immediately call the next firm on their list. The average personal injury case is worth $10,000 to $50,000 in fees. Missing 3 viable intake calls per month costs $30,000 to $150,000 in potential revenue. An AI receptionist that qualifies callers (matter type, date of incident, basic facts) and books consultations during evenings and weekends captures the callers that voicemail loses.
What does AI receptionist software cost to build custom?
A basic custom AI receptionist (call answering, FAQ responses, simple appointment booking with one calendar, message-taking with email/SMS notifications) costs $25,000 to $40,000 and takes 6 to 8 weeks. This is appropriate for a single-location service business with straightforward scheduling that just needs after-hours and overflow coverage.
A mid-complexity system (multi-provider scheduling, emergency classification and routing, practice management or field service system integration, call recording and analytics, multi-location support) costs $40,000 to $80,000 and takes 8 to 14 weeks. This covers most multi-provider medical/dental practices and multi-technician home services companies.
A full AI phone system (inbound and outbound calling, appointment reminders, lead qualification, insurance verification, intake workflows, EHR/PMS integration, multilingual support, analytics dashboard with conversion tracking) costs $80,000 to $120,000 and takes 3 to 5 months. This is for larger practices, multi-location businesses, and companies that want the AI handling the full phone operations workflow, not just answering calls.
Monthly operating costs for a custom system run $500 to $2,000. This includes telephony costs (Twilio charges per minute of call time), STT and TTS API costs (priced per minute of audio processed), LLM API costs (priced per token), hosting, and monitoring. At 1,000 calls per month averaging 3 minutes each, telephony plus AI processing costs run approximately $400 to $800 per month. Compare this to a human receptionist at $3,000 to $4,500 per month, or an answering service at $500 to $1,500 per month with per-call charges that scale with volume.
When should you buy off-the-shelf vs build custom?
Buy off-the-shelf when the business has a single location, a single provider or a small team with interchangeable scheduling, fewer than 300 calls per month, standard appointment types with uniform durations, and no need to integrate with industry-specific software. A solo dentist, a single-attorney law office, or a one-truck plumbing company fits this profile. Smith.ai or Dialzara handles the calls for $200 to $500 per month, and the ROI is immediate.
Build custom when the business has multiple providers with different specialties and availability patterns, over 500 calls per month, complex scheduling logic (procedure-based time blocks, insurance pre-verification, emergency classification), integration requirements with practice management, EHR, or dispatch systems, and multi-location operations with location-based routing. A 5-dentist practice, a multi-location HVAC company, or a law firm with 8 attorneys across 3 practice areas fits this profile.
The hybrid approach works for businesses in between: start with off-the-shelf to prove the concept and measure the value, then build custom once the call volume and scheduling complexity justify the investment. The off-the-shelf platform generates 3 to 6 months of call data (peak times, common requests, caller demographics, missed-call patterns) that makes the custom build specification more accurate.
What metrics should you track with AI receptionist software?
Call resolution rate measures the percentage of calls the AI handles without human intervention. A well-configured system resolves 60 to 80% of calls. The remaining 20 to 40% are transferred to humans because they require judgment the AI is not trained for, the caller specifically requests a person, or the situation is an emergency that needs immediate human attention. A resolution rate below 50% indicates the AI's knowledge base or call handling rules need tuning.
Appointment conversion rate measures the percentage of callers who request an appointment and successfully book one during the call. This should be 70 to 85%. A lower rate indicates scheduling friction: the AI cannot find availability that works for the caller, the booking process takes too long, or the AI does not handle rescheduling and cancellation requests properly.
After-hours capture rate measures how many calls the system handles outside business hours as a percentage of total daily call volume. For most service businesses, 25 to 40% of calls come in before 8am, after 5pm, or on weekends. Each of these calls previously went to voicemail. The after-hours capture rate directly translates to recovered revenue: calls that would have been lost are now answered, qualified, and booked.
Caller satisfaction (measured through post-call surveys or callback rates) indicates whether the AI experience meets caller expectations. The target is that fewer than 5% of callers call back to speak with a human after the AI has resolved their call. A higher callback rate means the AI is confirming appointments or providing information that callers do not trust, and they want a human to verify.
Madgeek builds custom AI receptionist and phone systems for service businesses where call handling complexity exceeds what SaaS platforms support. The BPO operations AI case study demonstrates the approach: AI-powered call quality monitoring enabled a contact center to scale from 50 to 80+ agents in 3 months by removing the QA bottleneck. The same engineering pattern (real-time voice processing, classification, and action) applies to AI receptionist systems for HVAC, dental, legal, and medical practices.
Need a team to build this for your business?