The AI voice agent glossary.

AI calling agent

An AI voice agent that places outbound calls at scale — reminders, collections, cart recovery, confirmations and win-back — in the customer's language.

AI receptionist

An AI voice agent that answers inbound calls 24/7 — greeting callers, booking appointments, answering questions, and routing to the right person.

AI voice agent

Software that makes and answers phone calls with a natural voice — understanding the caller, taking action, and handing off to a human when needed.

Chatbot

Software that holds a text conversation, usually on a website or app. A chatbot is one form of conversational AI; a voice agent is its phone counterpart.

Conversational AI

Technology that lets software understand and respond in natural language across voice and chat, holding a real back-and-forth conversation.

SCIMSCIM

protocol for automated user provisioning. Enterprise SSO feature.

Virtual receptionist

A remote or automated service that answers calls on a business's behalf. Modern virtual receptionists are AI voice agents rather than human call centers.

Voice AI

Artificial intelligence applied to spoken language — recognizing speech, understanding intent, and responding with a synthetic voice in real time.

Voicebot

A bot that interacts by voice rather than text. Modern voicebots use conversational AI to hold natural phone conversations, not scripted menus.

Webhook

HTTP POST Finn sends to your URL on events (outbound), or HTTP POST your endpoint receives from Finn mid-call (inbound). See Integrations → Webhooks.

Automatic speech recognitionASR

The formal name for the technology that converts speech into text. ASR and speech-to-text (STT) refer to the same thing.

Barge-in

The ability for a caller to interrupt the agent mid-sentence and have it stop and listen — essential for natural, non-frustrating conversation.

Custom vocabulary

list of words/phrases the speech recognizer should expect. Improves transcription accuracy for brand names, drug names, jargon. See Creating a Finn.

Endpointing

Detecting when the caller has finished speaking so the agent knows it's its turn to respond. Bad endpointing causes awkward pauses or interruptions.

Speech-to-textSTT

Technology that transcribes spoken audio into text in real time. STT is how a voice agent understands what the caller is saying.

Text-to-speechTTS

Technology that converts written text into natural spoken audio. TTS is what gives an AI voice agent its voice.

Voice cloning

Creating a synthetic text-to-speech voice that mimics a specific person's voice from a short sample. Used to give agents a consistent brand voice.

Voice latency

The delay between a caller finishing speaking and the agent responding. Low latency (sub-second) is what makes an AI voice conversation feel natural.

Dialogue management

Keeping track of context across a multi-turn conversation — what's been said, what's still needed — so the agent stays coherent.

Entity extraction

Pulling the specific details out of what a caller says — dates, names, amounts, phone numbers — that the agent needs to complete a task.

Intent recognition

Identifying the goal behind what a caller says — e.g. 'book appointment', 'check balance', 'cancel' — so the agent can act on it.

Knowledge base

searchable corpus of reference material attached to a Finn. Files, pasted text, website imports. See Knowledge Base.

Large language modelLLM

An AI model trained on vast text that generates and understands language. LLMs are the reasoning engine inside modern conversational and voice AI.

Natural language processingNLP

The broad field of AI concerned with computers understanding, interpreting and generating human language — the foundation under conversational AI.

Natural language understandingNLU

The part of AI that works out what a person means — their intent and the key details — from natural language, not just the literal words.

Retrieval-augmented generationRAG

A technique that grounds an LLM's answers in your own documents — retrieving relevant content at query time — so a voice agent stays accurate to your data.

System prompt

The instructions that define an AI agent's role, tone, rules and boundaries. The system prompt shapes how a voice agent behaves on every call.

Attestation

a cryptographic signature attached to outbound calls in some regions (notably US/Canada STIR/SHAKEN) that lets receiving carriers verify the caller-ID…

Call routing

Directing an incoming call to the right destination — person, team, or queue — based on rules, caller input, or what the AI agent understands.

CNAMCNAM

Caller Name. The text label that shows on the recipient's caller-ID display in the US. Requires registration to avoid spam labeling.

Conversational IVR

An IVR upgraded with conversational AI — callers say what they want in natural language instead of navigating a menu tree.

DIDDID

Direct Inward Dial. Synonym for "phone number" in telecom contexts.

DTMFDTMF

The tones a phone keypad makes when you press a key. DTMF is how callers enter numbers (like an account ID) that a voice agent can capture.

E.164E.164

international phone number format (e.g. +15551234567). The platform normalizes to E.164 internally.

Erlang

The unit of telephone traffic. One Erlang is one hour of calls per hour — a single line occupied continuously.

Erlang C

The queuing formula that turns call volume and handle time into the number of agents a service-level target needs. The standard basis for contact-centre staffing.

Inbound calling

Calls that come in to a business. For AI voice agents, inbound means answering, understanding and resolving calls customers place to you.

Interactive voice responseIVR

The classic 'press 1 for sales' phone menu system. IVR is menu-driven and can't understand free speech — the predecessor to conversational voice AI.

Port

transfer phone-number ownership from another carrier into Finn.

SIP trunkingSIP

A method of connecting a phone system to the telephone network over the internet using the SIP protocol — how voice platforms carry calls at scale.

STIR/SHAKEN

US/Canada caller-ID attestation framework. Mandatory for outbound calls; the platform handles it automatically.

Voicemail detection

heuristic that recognizes a voicemail greeting and decides whether to end, leave a message, or proceed.

VoIPVoIP

Voice over Internet Protocol — carrying phone calls over the internet instead of traditional phone lines. The transport layer under most modern voice agents.

Warm transfer

Transferring a call to a human while passing along context — who's calling and why — so the caller doesn't have to repeat themselves.

AI BDRAI BDR

A voice agent focused on outbound prospecting — the business-development sibling of the AI SDR. In practice the same agent, pointed at cold outreach.

AI SDRAI SDR

A voice agent that does the outbound part of the SDR role — calling, qualifying and booking meetings automatically — so reps only talk to warm leads.

Concurrency

The number of phone calls a voice platform can handle at the same time. Concurrency sets how much inbound spike or outbound volume you can run.

Outbound calling

Calls a business places out to customers or leads — reminders, sales, collections, follow-ups. AI voice agents run outbound at scale, concurrently.

BAABAA

Business Associate Agreement. Required for HIPAA compliance; available on Enterprise.

Do Not CallDNC

A registry of numbers that have opted out of telemarketing calls. Outbound voice agents must scrub against DNC lists to stay compliant.

Suppression list

global do-not-call list. Numbers on it are silently dropped from all deployments.

TCPA complianceTCPA

Following the US Telephone Consumer Protection Act — the law governing automated calls: consent, calling hours, opt-outs and disclosures.

Abandon rate

The share of callers who hang up before reaching an agent. The clearest signal that a queue is under-staffed relative to demand.

Average handle timeAHT

The average time an agent spends on a contact end to end — talk time plus hold time plus after-call work. The core input to every staffing calculation.

Average speed of answerASA

The mean time callers wait in queue before an agent answers. Useful alongside service level, misleading on its own.

Call deflection

Resolving a customer's need without a human agent — the share of calls an AI agent handles end to end. A core ROI metric for voice AI.

Call summarization

Automatically generating a concise summary of a phone call — key points, outcome, next steps — from the transcript, so nothing is logged by hand.

Cost per call

Total contact-centre cost divided by contacts handled. Commonly $6–12 for a live agent — but the denominator has to be stated or the figure means nothing.

First contact resolutionFCR

The share of contacts resolved without a follow-up. Benchmarked at 70–75%, and the correct counterweight to average handle time.

Occupancy

The share of logged-in time an agent spends handling contacts. Healthy between 75% and 85%; sustained higher figures drive burnout and attrition.

Service level

The share of calls answered within a target time — commonly 80% within 20 seconds. A convention rather than a standard, despite how often it is quoted as one.

Shrinkage

The share of paid time agents are not available to take contacts — breaks, training, meetings, absence. Typically 30%, and the reason rostered headcount exceeds calculated agents.

Speech analytics

Analyzing recorded or live calls at scale to surface trends — topics, sentiment, outcomes, compliance — instead of manually reviewing a handful.

Action node

a node in a Workflow that performs a side effect (book calendar, call webhook, transfer call). See Workflows.

Active call

a call currently in progress. Visible in real time on Live Deployments.

Ask node

a Workflow node that asks the caller a question and stores the answer in a variable. See Workflows.

Audience

a list of contacts a Finn can call (or be expected to handle on inbound). Comes from CSV upload or CRM sync. See Audience.

Audience mapping

the mapping between CRM fields and Finn audience fields, configured per CRM connection.

Buffer

the queue of contacts the dispatcher has staged for an active deployment. A "buffer drained" warning means the deployment is about to run out of work.

BYO numberBYO number

Bring Your Own number. Use a phone number you already own with another provider via port-in or SIP-trunking.

Call duration

wall-clock time the call was connected. Billed to the second.

Call history

searchable log of every call across all deployments.

Call type

outbound (Finn dials out), inbound (Finn answers ring-ins), or both.

Calling-hour window

a per-contact local-time range (e.g. 9 AM – 6 PM) during which the platform will attempt calls. Respects each contact's timezone.

Completion rate

percentage of attempted calls that reached a live human and finished the conversation. Distinct from success rate.

Custom field

an extra column in an audience beyond name/phone/country code. Available as {{contact.field_name}} inside Finn instructions.

Data Extractor Sidebar

the per-call drill-in view, opened by clicking any call row. Shows transcript, audio, AI analysis, post-call fields.

Decision node

a Workflow node that branches based on a variable or LLM-evaluated condition.

Defined begin message

a verbatim opening script the Finn reads when a call connects. Alternative: dynamic (LLM-generated).

Deployment

the act of putting a Finn on real calls. Binds Finn + Audience + Phone Number + Schedule. See Campaigns & Deployments.

Dispatcher

internal component that pulls contacts from a deployment's buffer and triggers outbound calls.

Dynamic deployment

a deployment launched on-demand (vs. scheduled).

Finn

the AI voice agent. Configured with identity, voice, instructions, knowledge base, optional workflow.

Finn icon

optional avatar shown next to a Finn in the dashboard. Cosmetic.

Finn tasks

structured side-effects a Finn can trigger (book calendar, webhook, CRM update). Defined per Finn.

Free-form mode

a Finn that operates without a workflow, driven by identity + guardrails + knowledge base. Default mode.

From number

the phone number that appears on caller ID for outbound calls.

Guardrails

rules in a Finn's "Style & Guardrails" field that constrain what it can say or do. See Creating a Finn.

Handoff number

phone number the Finn transfers to when escalation triggers (e.g. caller asks for a human).

Identity text

the "who you are" field on a Finn. Drives tone and behavior.

Idle reminder

the line the Finn says after a period of silence to nudge the caller. Default: "Are you still there?".

Live deployment

a deployment currently dialing. Visible on the Live Deployments page.

Number pool

a group of phone numbers used by parallel deployments to spread outbound dialing across many DIDs. See Phone Numbers → Pools.

Parallel deployment

a deployment that dials thousands of contacts simultaneously. Enterprise feature.

Pasted text

a knowledge base source where you paste content directly. Alternative to file upload.

Phone number

a DID owned by your account, used as caller ID (outbound) or ring destination (inbound). See Phone Numbers.

Post-call analysis

structured fields extracted from the transcript after a call completes (e.g. appointment_booked: true). Defined per Finn.

Recording

the audio file of a completed call. Default retention 90 days.

Response guidelines

rules on how the Finn should format and structure responses. See Creating a Finn.

Retry

re-attempt of a failed call (no answer, busy, voicemail). Configured per deployment.

Scheduled deployment

a deployment with a future start time. Sits in queue until it transitions to Live.

Segment

a filtered subset of an audience based on custom-field criteria.

Standard deployment

a deployment that dials contacts sequentially or in small batches. The default.

Success rate

percentage of completed calls that achieved the goal defined by the Finn's primary success field.

Telephony account

the underlying carrier connection, either Finn-managed or BYO via SIP trunk.

Test on Call

feature that calls your phone with the Finn for pre-launch testing.

Test on Chat

text-based browser simulation of the Finn for fast iteration.

Time zone

used for scheduling and calling-hour windows. Per-contact (derived from country code, overridable via custom field).

Voice library

the catalog of voices a Finn can be assigned. Mix of provider voices and Finn-tuned voices.

Welcome message

what the Finn says when a call first connects. Defined or dynamic.

Workflow

visual node-based editor for structured conversation flows. Alternative to free-form mode. See Workflows.