Glossary
The AI calling glossary, in plain language.
Every term you will hit while evaluating AI voice agents, dialers and outbound compliance, defined without the vendor spin. Where a term deserves more than a paragraph, there is a full guide behind it.
Dialing and outbound
Dialing and outbound
The four dialer types differ only in how many lines they place per available rep. That single number drives cost, talk time and compliance risk.
Predictive dialer
Places more calls than you have free reps and uses live pacing data to guess when the next rep will be available. Highest talk time per rep, highest abandonment risk, and it needs real volume before the prediction means anything.
Power dialer
Dials a fixed number of lines per available rep, a ratio you set rather than one an algorithm decides. Simpler than predictive and far easier to keep compliant, because at one to one there is no abandonment risk at all.
Progressive dialer
Dials automatically as soon as a rep becomes free, strictly one call per rep. No prediction and no overdialing, so no abandoned calls. Many vendors use this term and power dialer interchangeably.
Preview dialer
The rep sees the record and decides when to dial. Slowest by a wide margin and the right choice when each conversation is worth a lot or the context is complicated.
Local presence dialing
Displays a caller ID sharing the recipient’s area code, which lifts answer rates because unfamiliar area codes get screened. Lawful when you own the number and it rings back to you.
Speed to lead
How long it takes to contact a lead after they raise their hand. Measured in minutes rather than hours, because contact rates fall sharply with delay and the first caller usually wins the deal.
Warm transfer
Passing a live caller to a colleague along with the context already gathered, so the customer does not repeat themselves. A cold transfer sends them through with none of it.
Connect rate
The share of dials that reach a live person. It drives every pacing decision a dialer makes, and it moves through the day, which is why pacing tuned on an average overdials the quiet hours.
Voice AI technology
Voice AI technology
The pieces that sit between a ringing phone and a natural conversation. Latency is usually what separates a good agent from an obviously robotic one.
AI voice agent
Software that holds a real two way phone conversation, works to a goal such as qualifying a caller or booking an appointment, and hands off to a person when it should. Judged on resolved calls rather than time on the line.
Conversational AI
The broad category covering any system that converses in natural language, across voice and text. A useful umbrella term and a poor buying criterion, since it says nothing about whether the thing completes a task.
IVR
Interactive voice response, the press one for sales menu. It routes by matching a keypress or keyword to a branch, which is why it fails the moment a caller says something the tree does not anticipate.
Text to speech
Turning written words into spoken audio, often shortened to TTS. Modern neural voices carry intonation and pacing, which is most of what makes an agent sound human rather than announced.
Speech to text
Transcribing spoken audio into words, also called ASR or automatic speech recognition. Accuracy drops with accents, background noise and industry jargon, which is why domain tuning matters more than headline accuracy claims.
NLU
Natural language understanding, the step that works out what a caller actually wants rather than only what they said. It is the difference between hearing the words and knowing that cancel my appointment and I need to reschedule are different requests.
Latency
The delay between a caller finishing a sentence and the agent starting to answer. Humans notice well under a second, so latency is usually what makes an agent feel robotic even when the voice itself is excellent.
Backchanneling
The small sounds people make while listening, the mm hm and right that signal attention. Agents that skip it feel like they are waiting rather than listening.
Metrics
Metrics
Most of these break when they become targets rather than diagnostics. Handle time is the clearest example.
Average handle time
Talk time plus hold time plus after call work, divided by contacts handled. Leaving out after call work is the most common way the figure gets flattered, and targeting it tends to move work rather than remove it.
First contact resolution
The share of contacts fully resolved without the customer needing to come back. The natural counterweight to handle time, because it catches the callbacks that a shorter call creates.
Abandonment rate
The share of answered calls where nobody was there to speak, or where the caller hung up waiting. It is also a regulated number for outbound campaigns, which caps how aggressively any dialer may pace.
Contact rate
The share of attempted contacts that reached the right person, as opposed to a voicemail or a wrong number. Distinct from connect rate, which counts any human answering.
Occupancy
The share of an agent’s logged in time spent on contacts rather than waiting. Push it too high and quality drops, because there is no recovery time between difficult conversations.
Cost per resolved contact
Total cost divided by contacts actually resolved. The metric that survives automation, because deflecting simple calls raises handle time while lowering this, and only one of those tells you the truth.
Compliance
Compliance
Consent decides whether you may call. Disclosure decides whether you say a machine is calling. They are separate problems and conflating them causes most of the confusion.
The federal rules behind these terms sit in 47 CFR 64.1200 and 16 CFR 310.4.
TCPA
The Telephone Consumer Protection Act, the US law governing automated calls and texts. It sets the consent, calling window and opt out obligations that apply regardless of which dialer or AI platform you use.
Prior express written consent
The higher standard of permission required for certain marketing calls and texts. It has to be specific, provable later, and cover the kind of contact you are actually making.
Telemarketing Sales Rule
The FTC rule covering telemarketing practice, including the abandoned call safe harbor that caps abandonment at three percent of calls answered by a person and requires a live rep within two seconds of the greeting.
Do Not Call
Registries of people who have opted out of telemarketing, at national, state and company level. Screening against all three before dialing is the obligation, not screening against the national list alone.
Quiet hours
The times of day when calling is prohibited, measured in the recipient’s own time zone rather than yours. A rule that trips up any team dialing across states from one location.
AI disclosure
Telling the person that AI is involved in the conversation. Largely a trust decision rather than a settled legal one, and the wording matters, because defensive phrasing creates the suspicion it was meant to avoid.
Deliverability and messaging
Deliverability and messaging
Whether your call connects and whether your text arrives are now separate technical problems, and both come down to how carriers see your numbers.
STIR SHAKEN
The call authentication framework that lets carriers verify a caller ID has not been spoofed. Numbers you can properly attest to as your own get treated better, which is why churning through numbers now performs worse than keeping a small pool.
Number health
How carriers currently score your outbound numbers. It degrades with volume from unfamiliar numbers and with complaint rates, and it is the usual reason a campaign that worked last quarter stops working.
10DLC
Ten digit long code, the carrier approved route for automated texting from a normal US business number. Requires registering the brand and each use case, and unregistered traffic is filtered silently rather than rejected.
A2P
Application to person, meaning any message sent by software rather than typed by a human. The distinction matters because carriers treat A2P traffic on long codes differently from person to person texting.
Trust score
The rating assigned to a registered messaging brand, which sets how many message segments you may send per minute. Two fully registered businesses on the same platform can have very different real capacity.
Spam likely
The label carriers place on a number they suspect of unwanted calling. It appears before the phone rings, which means the call is judged before your agent says anything at all.
Know the terms.
Now hear it work.
See an AI voice agent answer a real call on your script, in your industry, with the compliance checks running before it dials.