Every sales call for an AI phone or chat tool arrives with the same stack of terms: NLU, latency, intent, resolution rate, LLM, omnichannel. Most of it is borrowed straight from an engineering team's internal vocabulary, and it can make a simple buying decision feel like a technical exam.
Some of those words matter to whether the product will work for your business. Most don't, once you know what they mean. The problem is that vendors rarely tell you which is which, and an owner who doesn't want to look uninformed tends to nod along and judge the product on how confident the salesperson sounded.
This conversational AI glossary is a plain-language pass through the terms that come up most often when small businesses evaluate an AI receptionist, voice agent, or chat assistant. Each term gets a short definition, why it matters (or doesn't), and where relevant a quick test you can run yourself.
If you're short on time, skip to the table of which terms deserve your attention. We build one of these products, so where a term touches Aiva Chat, we'll say plainly what it does and doesn't do.
What is conversational AI?
Conversational AI is software that holds a back-and-forth conversation in ordinary language, by voice or text, rather than making people choose from fixed menus or buttons. For a small business, it usually shows up as an AI receptionist on the phone, a chat assistant on the website, or a drafting tool for text messages.
Under the hood, a voice conversation typically involves turning speech into text, understanding what the person wants, deciding on a reply or an action, and turning the reply back into speech. Some newer systems handle audio directly; OpenAI's Realtime API documentation describes building a speech-to-speech voice agent that works with audio and keeps conversation state.
You don't need to know which approach a vendor uses. What matters is whether the conversation feels natural, gets the facts right, takes the right action, and hands off to a person when it should.
Terms about what the AI hears and says
ASR (automatic speech recognition)
The part that turns a caller's spoken words into text the system can act on. If this is weak, everything downstream answers the wrong question. Test it with a real accent and some background noise, not a quiet demo room.
TTS (text-to-speech)
The part that turns the reply back into a voice. Judge it mid-conversation, not on a single sentence. The tell is usually in pauses and interruptions, not the first "hello."
Latency
The gap between a caller finishing a sentence and the reply starting. It's one of the few terms worth testing live, because a second or two of silence makes a system feel stuck even when the answer is right. We've written about why response time matters more than most owners expect.
Barge-in (interruption handling)
The ability to keep up when a caller talks over the system: finishing its sentence, cutting in with a correction, or answering before the question ends. Without it, the system talks over people or loses the thread. Interrupt it mid-answer and see what happens.
Multilingual and code-switching
Handling more than one language, including within a sentence, such as a caller opening in Hindi and finishing in English. Aiva Chat works in 13 languages and picks up which one the caller is using; its bilingual mode can greet in one language and follow the caller into another. See the full language list.
Terms about what the AI understands
NLU and NLP
Natural language understanding and natural language processing: the general idea of making sense of ordinary sentences rather than matching keywords. What it buys you is that customers can phrase things however they naturally would.
Intent
What the customer is actually trying to do. "Anything free tomorrow?" and "can I still get in this week?" are different sentences with the same booking intent. A system that only matches sentences treats them as unrelated.
LLM (large language model)
The layer doing the reasoning, and the reason a system can handle phrasing it hasn't seen before. It's not a spec-sheet item to compare. Judge the behavior on an unusual question, not the model's name.
Bot vs. agent
Often used interchangeably in marketing, but worth separating. A bot matches keywords or menu choices to scripted replies. An agent understands open-ended phrasing and takes action, like checking a calendar or completing a booking. Our breakdown of AI agent vs. bot covers how to test it.
Jargon describes how something is built. It doesn't tell you whether it works for your business; only trying it does.
Terms about how well it's working
Resolution rate
The share of conversations where the customer's need was met without a person stepping in, not just "the system said something." The definition varies by vendor, so ask how it's calculated. Our post on what resolution rate means explains the nuance.
Escalation
Handing a conversation to a person. Done well, it's a design choice: a well-run system escalates on purpose for upset customers, sensitive topics, or explicit requests. We've written about setting escalation rules in more detail.
Repeat contact
A customer reaching out again soon after a conversation that looked resolved, often about the same issue. It's one of the better signs that "resolved" didn't really answer the question.
Analytics
Where the measurements live. In Aiva Chat, analytics shows call volume, bookings and lead counts, automatically tagged topics, and lets you ask plain-language questions across your conversations. Ask any vendor to show a real topic breakdown rather than define "intent."
Terms about what you might be replacing
IVR (interactive voice response)
The "press 1 for appointments, press 2 for billing" phone tree. It routes by button press but doesn't understand anything said. Ask it a question and it can only send you back to the menu. See our comparison of AI receptionists and IVR.
Auto-attendant
A simpler cousin of IVR that routes by extension or department, with no language understanding at all. If a vendor's "AI receptionist" is an auto-attendant with a friendlier recording, catch that before signing.
Answering service and call center outsourcing
Human agents, usually shared across many client businesses, answering on your behalf. You get a person's judgment on every call, but each agent handles one call at a time, so busy periods mean queues. Compare the models in our post on AI receptionist vs. answering service.
Terms about setup and control
Configuration vs. training
What you do is configuration: adding hours, prices, FAQs, booking rules, and handoff instructions. The language skills come built in, so your job is describing your business. The difference matters because you can change configuration yourself.
Omnichannel
The same system and the same answers across voice, web chat, and SMS. Ask the same question on two channels and see whether the answers match.
Data residency
Where conversation processing happens geographically, and whether you can choose. Aiva Chat is hosted on AWS and doesn't offer a choice of region today. Ask any vendor where processing happens, not just where their office is.
Usage-based pricing
Paying per minute, conversation, or message rather than a flat fee. Aiva Chat splits the two: a flat plan with every feature, plus usage you pay providers directly.
| Part | Who bills it |
|---|---|
| Aiva Chat plan | Aiva Chat, from $49/month (Starter, Growth, Pro) |
| AI model usage | OpenAI, on your own API key |
| Numbers, call minutes, SMS | Twilio, on your own account |
Aiva Chat adds no markup on usage. Compare the plans against other vendors at your real volume, since resold minutes can be cheaper or dearer depending on how much you use.
Uptime and fallback
Uptime is how often a system is available, usually a percentage. Fallback is what happens when something genuinely breaks. A good fallback still gets the caller to a voicemail or another number rather than a dead line.
Terms about what you connect it to
Webhook
A way for one system to notify another the moment something happens, like a new booking or a finished call, without anyone checking a dashboard. Aiva Chat sends signed webhooks for events such as a booking made, moved, cancelled or paid, which can feed your own systems or any tool that accepts a webhook URL.
Integration and CRM
An integration connects the AI to a tool you already use, which stays the system of record. Aiva Chat connects directly to Google Calendar, HubSpot, Twilio, and your own Stripe account (payment links during calls, plus deposits and passes on booking pages); other CRMs can be reached via signed webhooks.
External-tool API keys and custom actions
Credentials that let the assistant call out to a system you already run, such as looking up an order during a call. That's different from a public developer API, which Aiva Chat doesn't offer; we've explained why.
Embeddable widget
A chat box added to your existing website, usually with a snippet of code, rather than a separate page. Aiva Chat's widget installs with one script tag and supports text and voice.
Which terms deserve your attention
| Term | Worth testing yourself? | Why |
|---|---|---|
| Latency | Yes | Silence on a call is felt immediately |
| Barge-in | Yes | Real callers interrupt constantly |
| Escalation | Yes | Decides what happens to upset or complex calls |
| Integrations | Yes | Determines whether bookings and leads land where you work |
| Data retention | Yes, ask directly | Affects privacy obligations |
| NLU, LLM, intent | No | Describe how it's built, not whether it works |
| Uptime percentage | Ask, but don't over-weight | Hard to verify before you're live |
The terms in the top half describe what your customers will experience or where your data ends up, and you can test or ask about each in minutes. The bottom half matters to engineers but tells you little in a sales call. A vendor using an impressive model can still have slow replies, poor handoffs, or no calendar connection.
Uptime deserves a question, but a quoted percentage is hard to verify in advance. A free trial on real traffic tells you more.
Common mistakes owners make with AI jargon
Judging fluency instead of results
A salesperson who explains NLU well isn't evidence the product handles your calls well. Judge the product on a live call.
Accepting metrics without definitions
"Resolution rate" can be calculated many ways. Ask how it's measured and whether escalations count as resolved.
Assuming "AI receptionist" means one thing
Products with the same label range from auto-attendants to full agents that book and transfer. Test what it actually does.
Ignoring the boring terms
Integrations, retention, and fallback sound dull, but they decide whether the tool fits your business day to day.
Best practices for cutting through vendor jargon
- Ask for a live call, not a slide. Interrupt it, ask two things at once, and request a person.
- Use your own FAQs. A demo built on the vendor's script hides gaps.
- Ask how every metric is calculated. Get definitions in plain words.
- Check integrations by name. Confirm the exact calendar and CRM connections you need.
- Ask about data in writing. Hosting, retention, scrubbing, and deletion.
- Compare total cost at your real volume. Plan price plus usage, not the headline number.
Conclusion
If you remember one thing from this glossary, make it the dividing line in the table above: some words describe what your callers will feel, and the rest describe the engine. Spend your sales-call time on the first group, latency, barge-in, escalation, integrations, and retention, and politely skip past the second.
The trap to watch for is a vendor whose vocabulary is better than its product. A metric with no definition, or a demo you can't steer, should count against them however fluent the pitch. If you're comparing vendors now, our guide on how to choose an AI receptionist turns these terms into a practical checklist.
Ready to test it on your own questions?
Put the glossary down and pick up the phone. Call +1 (970) 972-0974, interrupt Aiva Chat mid-answer, ask two things at once, then ask for a person, and see which terms from this page you can actually hear.
To check it against your own FAQs and hours, a trial account runs on your own OpenAI and Twilio keys, so you'll see real usage costs from day one.
Start your 14-day free trial with no card required, or book a demo and bring the terms you've been hearing from other vendors.
Posts from the people building Aiva Chat in Rajkot — how the product works, what we have learned building it, and guides for getting the most out of an AI receptionist.