No-code vs developer platform vs custom build: how to choose an AI appointment booking agent

Most comparison articles assume you either click together a no-code agent or you have engineers who will live in Vapi. A lot of service businesses have neither. One location, a calendar, and an owner who answers the phone. Or several locations, a CRM, and no one whose job is prompts and function calls. The choice below is for that buyer.

The five ways to get an AI booking agent

No-code builders

Best for a single location booking into one calendar. HighLevel Voice AI lets you describe the call and connect a calendar without writing a function. CloudTalk sells voice AI on top of its phone system. HighLevel publishes a Voice Engine rate of $0.045 per minute plus the voice. AI Employee Growth is $50 a month per location and includes 100 Voice AI minutes, then pay as you go. Unlimited is $97 a month per location under a fair-use policy. Phone charges are separate. Synthflow moved upmarket in 2026, so for a single location HighLevel is now the realistic self-serve option. Synthflow retired its self-serve monthly plans, and its pricing page now shows enterprise contracts from $30,000 a year.

The limitation is the builder. Provider-specific hours, a deposit before the slot is held, or an insurance check that changes the duration are where the visual flow stops. You then either simplify the business or leave the tool.

Developer platforms

Best for a company with an engineer on staff. Vapi and Retell are the two KeenCraft ships on when the buyer has someone to own the project. Vapi hosting is $0.05 per minute. Speech-to-text, the model, the voice, and the carrier are billed by those providers, or at no Vapi charge if you bring your own keys. Retell voice infrastructure is $0.055 per minute, with voices and models on its own menu. A standard all-in minute on either one lands around $0.11 to $0.15, which is broken down in what an AI voice agent actually costs in 2026.

The limitation is maintenance. Someone has to own the prompt, the function that checks the calendar, and the calls that fail after launch. A platform account with no owner becomes a demo that answers the phone and books nothing.

Enterprise conversational AI

Best for a contact center already handling thousands of calls. Sierra and PolyAI sell that deployment: conversation design, integrations, and a contract scoped with a sales team. Neither publishes a self-serve per-minute rate.

The limitation is the shape of the deal. The timeline is months, and the price is a contract, not a subscription you can cancel after a slow week. A dental office or a trades company is not the buyer they staff for.

Hybrid AI plus a person

Best for legal intake and other calls where a person should stay in the loop. Smith.ai answers with trained receptionists. The published human plans start at $300 a month for 30 calls, $810 for 90, and $2,100 for 300. Booking onto a calendar is a separate $1.50 per call. The page also offers an AI-first track beside the human one.

The limitation is the meter. Included calls on the published human plans land around $7 to $10 each, and booking onto the calendar is another $1.50. That is the right price when a missed conflict check costs more than the call. It is the wrong price for a high volume of routine bookings.

Custom build by a studio

Best for a business that outgrew no-code and has no engineers of its own. KeenCraft builds that on Vapi, Retell, or LiveKit. One location is typically $3,000 to $12,000, and a multi-location build with CRM write-back is typically $12,000 to $30,000, as a fixed price. The full ranges are in what an AI voice agent actually costs. The production bar, latency and a real calendar write, is in how to build an AI voice agent that actually books appointments.

The limitation is time and the upfront number. This is a project with a written scope, not a same-day setup. VoiceCake is the live version of the outcome: appointments on the schedule a dental, healthcare, fitness, or mortgage office already opens in the morning.

How the five approaches compare

Prices checked September 2026. HighLevel's Voice Engine is $0.045 per minute from 20 May 2026, plus text-to-speech from $0.015, on its AI product pricing page. Synthflow's pricing page shows a single enterprise track, with contracts from $30,000 a year. Vapi hosting is $0.05 per minute and Retell voice infrastructure is $0.055 per minute, with the model, voice, and phone line on top. Smith.ai's human plans and the $1.50 booking add-on are on smith.ai/pricing. Sierra and PolyAI do not publish a self-serve rate.

When HighLevel or another self-serve no-code tool is genuinely enough

No-code is the right call when all of these are true:

• One location, and one calendar.
• Under a few thousand call minutes a month.
• The questions are standard: service, name, phone, and a time the calendar already offers.
• You are still testing whether callers will stay on the line with an AI agent.
• Nobody on the team would maintain a custom system after launch.

In that situation, pay the subscription. A custom build is money you do not need to spend yet. KeenCraft will say so on a strategy call. The AI voice agents work starts when the list above stops being true.

Signs you have outgrown no-code

• Booking rules the builder cannot express: provider-specific availability, insurance checks, deposits, or service durations that change the slot.
• CRM write-back past copying a name and a phone number into HubSpot or Salesforce. Duplicates and pipeline stages are the usual break.
• More than one location, and those locations must not see each other's calendars.
• Callers noticing the pause before the agent speaks.
• Volume past roughly 10,000 minutes a month, where platform fees start to dwarf a smaller build.
• Healthcare calls that need HIPAA-conscious handling.
• A requirement to keep the call audio and the transcripts on your own systems.

Latency is a measured number, not a feeling. On Rawk.ai, KeenCraft cut end-to-end voice response time from about 900ms to about 320ms by owning the pipeline. Multi-location isolation is the same kind of requirement. Dynaris runs tools per workspace so one site cannot book another site's calendar. The CRM half of either problem is automation that writes back to HubSpot, Salesforce, and Zoho.

Vapi vs Retell for booking, from production

KeenCraft has put booking agents on both Vapi and Retell, and on LiveKit when the stack had to be owned outright. The difference that shows up after launch is who owns the pieces. A demo on either platform can sound fine. The appointment is the write.

Retell is one product. Voice infrastructure is a published $0.055 per minute, the voice and the model are choices on the same bill, and a standard inbound flow can sit on a phone number in days when the calendar is one they already connect. Twenty concurrent calls are included. An engineer still owns the prompt, the function that reads open slots, and the turn where the caller talks over the agent. If that person leaves, the agent does not maintain itself.

Vapi is the thinner layer, at $0.05 per minute of hosting. Speech-to-text, the model, the voice, and Twilio are billed by those providers, or at no Vapi charge if you bring your own keys. Bringing your own keys helps when those providers are already chosen. Ten concurrent lines are included, then about $10 a month for each extra line. For a clinic, HIPAA is a published add-on at $2,000 a month on top of usage. For a trades scheduler, that add-on is irrelevant.

On both platforms the failure KeenCraft sees is the same. The call sounds finished, and the calendar or the CRM has no record, or it has a duplicate. Retell is the default when one engineer and a standard calendar describe the team. Vapi is the default when the providers are already picked and that engineer wants to swap them. LiveKit is the step after either one, when the requirement is the latency number on Rawk.ai or the location boundary on Dynaris. That step is a fixed build, not a larger subscription.

A four-question decision guide

1. How many locations, and how many calendars, does the agent book into?
2. Does it need to create and update CRM records, or only add a calendar event?
3. How many call minutes a month do you expect?
4. Who fixes it when a caller says something nobody scripted?

One location, one calendar, a few thousand minutes, and you will fix it yourself: no-code. An engineer on staff who will own the functions: Vapi or Retell. A contact center and a procurement process: Sierra or PolyAI. Legal intake or any call a person must still touch: Smith.ai. Several locations, a real CRM write, or no engineer and a builder you have already outgrown: a studio build on KeenCraft's voice-agent work.

You can start on no-code and move later

Yes, and often you should. A month of real calls teaches the booking rules cheaper than a workshop. What carries over is the script, the list of questions, the exceptions callers actually hit, and the hours the office will honor. What gets rebuilt is the integration: the calendar read, the CRM write, the transfer, and the flow that lived inside the no-code tool.

Treat the no-code month as the requirements. When you outgrow it, the cost of the build is scoped against those calls, not against a guess. Book a call when you can point at the rule the builder cannot express. A written proposal typically lands in 2 to 5 business days.

Sources

• HighLevel AI product pricing: Voice Engine $0.045 per minute from 20 May 2026, text-to-speech from $0.015, AI Employee Growth $50 a month for 100 minutes, Unlimited $97 a month under fair use, phone charges separate.
• Synthflow pricing: enterprise contracts from $30,000 a year.
• Vapi pricing: hosting $0.05 per minute. Model providers are billed separately. HIPAA add-on $2,000 a month.
• Retell AI pricing: voice infrastructure $0.055 per minute, most voices $0.015, ElevenLabs $0.040.
• Smith.ai pricing: human plans from $300 for 30 calls, $810 for 90, $2,100 for 300. Booking an appointment is $1.50 per call.