White-Label Voice AI Glossary: Every Term Explained
The voice AI industry uses terminology borrowed from telephony, SaaS, compliance law, and machine learning, often without explaining any of it. Agencies evaluating white-label platforms encounter terms like "SIP trunk," "native platform," "TCPA," and "crews" in sales calls and documentation with the assumption that everyone already knows what they mean. This glossary defines 30 terms in plain language, organized alphabetically, with a featured section comparing AI receptionists, AI agents, and AI chatbots since those three terms cause the most confusion.
Every definition below is written for agency owners and marketers, not engineers. Where a term has different meanings in different contexts, the definition reflects how it applies to voice AI specifically. The glossary starts with a featured section on the three most confused terms (AI receptionist, AI chatbot, AI agent), then covers 25 alphabetical entries from ACMA to wrapper, each with a practical note on why the term matters for agency decisions. For the full reseller playbook these terms describe, see the white-label voice AI platform guide for agencies.
AI Receptionist vs AI Chatbot vs AI Agent
These three terms describe different types of AI automation, but vendors use them interchangeably, which creates confusion when agencies try to position their services to clients.
AI Receptionist
An AI receptionist is a voice AI system that answers inbound phone calls for a business. It picks up when the business owner cannot answer, handles the conversation using a knowledge base built from the business's website and reviews, and can book appointments, qualify leads, take messages, and send SMS follow-ups. The caller experience resembles speaking with a human receptionist. AI receptionists are inbound-only and focused on phone calls, not text channels. Trillet's white-label voice AI platform uses AI receptionist technology as the core offering agencies resell under their own brand.
AI Chatbot
An AI chatbot is a text-based system that handles conversations through typed messages on websites, SMS, WhatsApp, or Facebook Messenger. Chatbots do not handle phone calls. They respond to written input with written output. Some platforms, like Stammer AI, started as chatbot builders and later added voice capabilities as a secondary feature. When an agency sells "AI chatbot" services, clients often expect text-based automation, not phone answering.
AI Agent
An AI agent is a broader category that includes any AI system capable of taking autonomous actions across multiple steps. An AI agent might answer a phone call, look up the caller's appointment history in a CRM, check calendar availability, book a slot, and send a confirmation SMS, all within a single interaction. Both AI receptionists and AI chatbots can be AI agents if they execute multi-step workflows autonomously. The term is architecture-neutral: it describes what the system does (takes actions), not how the user interacts with it (voice vs text).
Which Term Should Agencies Use?
For phone-based services, use "AI receptionist" or "voice agent." Clients understand "receptionist" immediately. For text-based services across web chat, SMS, or WhatsApp, use "AI chatbot." For marketing materials targeting business owners who may not know the category, "AI receptionist" converts better because it describes a role the business already has (or needs).
Alphabetical Glossary
Each term below gets a plain-language definition in 2-3 sentences, written for agency owners evaluating or reselling white-label voice AI platforms.
ACMA (Australian Communications and Media Authority)
ACMA is the Australian regulator that governs telecommunications, including AI-powered phone calls. Compliance with ACMA rules means the platform follows Australian laws around call recording consent, spam prevention, and the Do Not Call Register. Agencies serving Australian clients need a platform that includes ACMA compliance, not one that treats it as an afterthought.
Call Forwarding
Call forwarding is the mechanism that routes unanswered calls from a business's existing phone number to the AI receptionist. The business keeps its current number. When the owner does not pick up (busy, missed, or declined), the call forwards to the AI. Setup takes about 30 seconds on most carriers and requires no hardware, porting, or third-party telephony accounts.
Churn Rate
Churn rate is the percentage of clients who cancel their service in a given period, usually measured monthly. If an agency has 20 clients and loses 2 in a month, churn is 10%. As an illustrative pattern we see with agencies on our platform, monthly churn often runs higher in the first year and falls once onboarding and reporting processes mature; your own numbers will vary by niche and service quality. High churn destroys recurring revenue faster than new sales can replace it, which is why disciplined client onboarding in the first 7 days matters so much.
Concurrent Calls
Concurrent calls means the number of phone calls a service can handle at the same time. A single human receptionist ordinarily handles one live call at a time, while voice AI can support several simultaneous conversations. Capacity still depends on the provider, account, carrier, campaign, and contracted limits, so agencies should confirm and load-test the expected peak rather than advertise unlimited concurrency.
Crews / Orchestration
Crews (sometimes called orchestration or multi-agent orchestration) is a feature where multiple AI agents work together within a single call, handing off to each other based on context. For example, a front-desk agent qualifies the caller, then hands off to a scheduling agent that books the appointment, then hands off to a billing agent that collects payment information. Each agent maintains its own context boundaries, preventing data contamination between steps. Trillet's Crews feature enables these clean handoffs with context isolation.
CSAT (Customer Satisfaction Score)
CSAT measures how satisfied callers are with their experience, typically on a 1-5 scale collected through post-call surveys. Lorikeet's 2026 contact-center benchmarks put a strong CSAT score at 85% or higher (Lorikeet, 2026 contact center benchmarks); well-tuned AI receptionists can reach that level on routine inbound calls, though results depend heavily on knowledge base quality. Agencies track CSAT to demonstrate value to clients and identify agents that need knowledge base improvements.
DNCR (Do Not Call Register)
The DNCR is a government-maintained list of phone numbers whose owners have opted out of receiving unsolicited calls. In Australia, the DNCR is managed by ACMA. In the US, the equivalent is the National Do Not Call Registry managed by the FTC. Platforms with outbound calling must automatically filter campaigns against these registries before dialing.
HIPAA (Health Insurance Portability and Accountability Act)
HIPAA is a US federal law governing protected health information (PHI). A badge is not enough: the covered workflow needs appropriate safeguards and contracts, including a Business Associate Agreement where required. ChatDash lists a $200/month HIPAA add-on, while VoiceAIWrapper publicly claims HIPAA support; verify whether an executed BAA is included. For Trillet, BAAs are available on Agency and Enterprise after its standard process and at no incremental BAA fee, with the covered workflow identified in an applicable Order Form. Studio is not authorized for PHI processing.
Honeypot Detection
Honeypot detection is a screening process that identifies trap phone numbers planted by regulators or litigants to catch companies making non-compliant outbound calls. If an outbound AI campaign dials a honeypot number, the agency faces TCPA fines of $500-$1,500 per call. Trillet is one of the few voice AI platforms offering honeypot detection as a built-in feature.
Inbound
Inbound refers to calls that come into a business from customers or prospects. An inbound AI receptionist answers these calls when the business owner cannot. Most voice AI deployments start with inbound because it solves the immediate problem of missed calls without requiring compliance-heavy outbound infrastructure.
Knowledge Base
A knowledge base is the information an AI agent uses to answer questions during calls. It typically includes business hours, services, pricing, FAQs, and policies. Modern platforms can draft a knowledge base automatically from the client's website and online reviews. Manual entry is also an option for specialized information the AI cannot find online.
Latency
Latency is the time between when a caller finishes speaking and when the AI begins its response. In voice AI, sub-1-second end-to-end latency (including telephony overhead) is the threshold for natural conversation. Longer delays create awkward pauses that make callers uncomfortable or cause them to hang up. Trillet responds in about 400ms, well under one second.
LTV (Lifetime Value)
LTV is the total revenue an agency earns from a single client over the entire duration of the relationship. If a client pays $400/month and stays for 14 months, LTV is $5,600. Agencies use LTV to determine how much they can afford to spend acquiring each new client. A healthy voice AI agency targets an LTV-to-CAC (customer acquisition cost) ratio of at least 3:1.
MBG (Money-Back Guarantee)
An MBG is a refund policy for a defined period; a free trial delays the first plan charge under stated limits. The two should not be used interchangeably. Trillet's 28-day money-back guarantee belongs to its direct-to-business AI Receptionist plan. Trillet Studio and Agency instead use a 7-day free trial with a card required, usage credit ($12 on Studio, $24 on Agency), stated trial limits, and billing on day 8 unless cancelled.
MRR (Monthly Recurring Revenue)
MRR is the predictable revenue an agency earns each month from active subscriptions. If an agency has 15 clients paying $400/month each, MRR is $6,000. MRR is the primary financial metric for voice AI agencies because the business model is subscription-based. Growth means increasing MRR through new clients while minimizing churn.
Native Platform
A native application platform operates its own application, orchestration, workflow, and customer-management layer rather than requiring the agency to connect a separate upstream voice-platform account. It can still use contracted model, speech, telephony, cloud, and other subprocessors; using those suppliers does not make it a wrapper. On Trillet's standard hosted route, the agency has one core commercial and support path with Trillet, while bring-your-own telephony and custom deployments are explicit exceptions. Architecture changes account, billing, credential, and escalation boundaries; it does not guarantee support quality or eliminate dependencies. See the wrapper versus native platform guide.
Outbound
Outbound refers to calls the AI initiates to contacts, rather than answering incoming calls. Common outbound voice AI use cases include lead follow-up callbacks, appointment reminders, re-engagement campaigns, and post-service surveys. Outbound calling carries stricter compliance requirements (TCPA, ACMA, DNCR) than inbound and is not available on all platforms.
SIP Trunk
A SIP trunk is the digital equivalent of a phone line. It connects a voice AI platform to the telephone network so it can make and receive calls. Agencies do not need to understand SIP trunk architecture in detail. What matters is whether the platform handles SIP connectivity internally (simpler for the agency) or requires the agency to provision its own SIP trunks through a provider like Twilio or Telnyx (more complex, more potential failure points).
SOC 2 (Type II)
SOC 2 Type II is an auditing standard that verifies a company's controls for data security, availability, processing integrity, confidentiality, and privacy over a sustained period (typically 6-12 months). It is more rigorous than SOC 2 Type I, which only evaluates controls at a single point in time. Agencies serving clients in healthcare, finance, or legal should require SOC 2 Type II certification from their platform provider.
STT (Speech-to-Text)
STT is the technology that converts spoken words into written text during a phone call. The AI needs to "hear" what the caller says before it can process and respond. STT accuracy directly affects conversation quality: if the system mishears "Thursday" as "Saturday," it books the wrong appointment. Leading STT engines report very low word error rates on clean speech; OpenAI's Whisper large-v3, for instance, posts around 2.7% WER on the LibriSpeech test-clean set (roughly 97% accuracy), though real-world phone audio with background noise and accents is consistently harder (Hugging Face Open ASR Leaderboard).
Sub-Account
A sub-account, called a workspace in Trillet, is a separate client environment within a white-label platform. It can isolate the client's agents, phone numbers, knowledge, calls, and access. Trillet Studio has a hard cap of three workspaces; the retired $15 additional-workspace option must not be used. Trillet Agency includes unlimited workspaces at $299/month.
TCPA (Telephone Consumer Protection Act)
TCPA is a US federal law that restricts certain telemarketing calls, auto-dialed or prerecorded calls, and unsolicited text messages. Potential damages can reach $500 per violation and up to $1,500 for a willful or knowing violation, subject to the law and facts. Platform controls such as Do Not Call filtering can help, but they do not automate legal compliance: the agency and client remain responsible for consent, notices, calling windows, list provenance, configuration, and supervision.
TTS (Text-to-Speech)
TTS is the technology that converts the AI's text responses into spoken audio during a phone call. TTS quality determines how natural the AI sounds to callers. Low-quality TTS produces robotic, monotone speech. High-quality TTS produces natural intonation, appropriate pacing, and realistic pronunciation. The quality of TTS has improved dramatically since 2024, with modern engines producing speech that most callers cannot distinguish from a human in short interactions.
White-Label
White-label means the agency resells the platform's technology under the agency's own brand. The end client never sees the underlying platform provider's name, logo, or domain. Full white-labeling includes custom domains, branded dashboards, custom emails, and branded client portals. The depth of white-labeling varies by platform and plan tier. Some platforms offer only logo swaps on a shared domain. Others, like Trillet's Agency plan, provide complete brand separation with custom domains and branded email communications.
Wrapper
A wrapper is an agency interface that requires a separately supplied upstream voice-platform relationship. Voicerr, Vapify, VoiceAIWrapper, and ChatDash document Vapi, Retell, ElevenLabs, or other upstream providers. Wrappers can offer low entry prices and provider choice, but the standard setup may leave the agency coordinating another account, credentials, usage bill, and escalation path. That is an operating-boundary distinction, not proof of worse uptime. Ordinary use of LLM, speech, telephony, or cloud subprocessors does not by itself make a product a wrapper.
Frequently Asked Questions
What is the difference between an AI receptionist and an AI agent?
An AI receptionist is a specific type of AI agent designed to answer inbound phone calls for a business. An AI agent is a broader category covering any AI system that takes autonomous multi-step actions. All AI receptionists are AI agents, but not all AI agents are AI receptionists. Some AI agents handle outbound calling, text chat, or internal workflows instead of inbound phone answering.
What does "native platform" mean in voice AI?
A native application platform operates the product and orchestration layer that runs the agents without requiring the agency to maintain a separate upstream voice-platform account. It may still use model, speech, telephony, and cloud subprocessors. On a standard hosted route this can give the agency one core vendor and support entry point, but support effectiveness and uptime still need independent evaluation.
Why does HIPAA compliance matter for voice AI agencies?
HIPAA-regulated workflows require appropriate safeguards and contracts. If an agency will process PHI, it should verify the vendor's BAA eligibility, subprocessors, data flow, configuration, and applicable order rather than rely on a pricing-page badge. ChatDash lists a $200/month HIPAA add-on, while VoiceAIWrapper claims HIPAA support; confirm the signed BAA terms. Trillet BAAs are available on Agency and Enterprise after its standard process, not Studio.
What is the difference between a wrapper and a native platform?
A wrapper adds a dashboard and agency features on top of another company's voice AI infrastructure (usually Vapi or Retell). A native platform owns and controls the entire stack. The practical difference for agencies: wrappers have multiple failure points and no control over outages, pricing changes, or feature development from the underlying provider. Native platforms handle everything in-house.
How is MRR different from total revenue for a voice AI agency?
MRR counts only the predictable, recurring subscription revenue from active clients. It excludes one-time setup fees, consulting charges, or project-based work. MRR is the metric investors and acquirers value most because it represents stable, repeatable income. A voice AI agency with $8,000 MRR and low churn is more valuable than one with $12,000 in mixed project and subscription revenue.
Understanding these terms is the first step; the higher-value question for most agencies is which of them translates into billable services. Our overview of voice AI use cases for agencies maps the vocabulary in this glossary onto concrete deployments you can sell, and the white-label voice AI guide walks through how to package and price them under your own brand.
Ready to resell a first-party voice AI application platform under your own brand, with one core hosted-platform relationship and compliance co-delivery available for eligible workflows? Explore Trillet's white-label platform, compare white-label agency pricing, and read the full white-label voice AI guide.
Updated for September 2026: Corrected Studio's hard workspace cap, separated the D2C 28-day guarantee from the white-label 7-day trial, qualified BAA and TCPA responsibilities, and rewrote native-platform and wrapper definitions around verified account and support boundaries rather than ownership of every dependency. Also corrected the white-label trial credit to $12 (Studio) / $24 (Agency), and corrected knowledge-base sources to the website and online reviews.
