Reference
AI Receptionist Glossary
Call automation carries vocabulary from three different industries: telephony, customer service and machine learning. This glossary explains the terms you will meet during an evaluation, in plain language.
How to use this glossary
Suppliers use these words inconsistently, which is part of why comparing quotes is difficult. The definitions below are the ones we use, and where the industry disagrees we say so rather than pretending a term has one settled meaning.
When a supplier uses a term you do not recognise, ask them to describe what the caller experiences instead. A definition that cannot be expressed as caller behaviour is usually marketing rather than capability.
Call handling terms
| Term | What it means |
|---|---|
| Call flow | The set of rules describing what happens on a call: what is asked, what outcomes exist, and when a person takes over |
| IVR | Interactive voice response. The traditional press-one-for-sales menu, distinct from a system the caller can simply speak to |
| Overflow | Routing calls to another destination only when your own line is busy or unanswered |
| Warm transfer | Passing context to the colleague before connecting the caller, so nothing has to be repeated |
| Cold transfer | Connecting the caller without context, which usually means they start their explanation again |
| Escalation | A deliberate handover to a person, triggered by rules you set rather than by system failure |
| Fallback | What happens when the primary outcome is unavailable, such as a callback when nobody answers a transfer |
| Containment | The proportion of calls resolved without a person. A high figure is only good if the right calls are escalating |
| First contact resolution | The caller’s need met on the first call, without a follow-up being required |
| Abandonment | Calls where the caller hangs up before reaching an outcome, often while waiting |
| After-hours coverage | Handling of calls outside your operating hours, including weekends and public holidays |
| Qualification | Establishing what the caller needs and whether it fits your services, before an outcome is chosen |
Technology terms
You do not need these to run a deployment, but they appear in supplier documentation and in conversations about why something behaved the way it did.
| Term | What it means |
|---|---|
| Speech recognition | Converting spoken audio into text. Also called ASR or speech to text. Accuracy varies with accent, noise and line quality |
| Text to speech | Generating the spoken voice the caller hears. Sometimes called synthesis or TTS |
| Language model | The component that interprets what the caller said and decides what to say or do next |
| Intent | What the caller is trying to achieve, such as booking, rescheduling or asking about opening hours |
| Entity or slot | A specific detail extracted from the conversation, such as a date, a service or a mobile number |
| Turn | One exchange in a conversation. Multi-turn means the system tracks context across several exchanges |
| Latency | The delay before the system responds. Short pauses feel natural; long ones make callers repeat themselves |
| Barge-in | Whether the caller can interrupt and be understood, rather than having to wait for a prompt to finish |
| Hallucination | A confident but incorrect answer generated by a language model. The reason unverified answers should never be promised to callers |
| Guardrails | Constraints preventing the system from answering outside its permitted scope, including fixed escalation rules |
| Prompt or instruction set | The written instructions defining behaviour, tone and limits. What most people mean when they say script |
| Transcript | The written record of what was said, distinct from the audio recording and from the structured summary |
Integration and commercial terms
| Term | What it means |
|---|---|
| Native integration | A prebuilt connection to a specific platform, supported by the supplier |
| API | A documented interface allowing two systems to exchange data directly |
| Middleware | An automation platform such as Zapier or Make sitting between systems, adding flexibility and an extra failure point |
| Webhook | A message sent automatically from one system to another when an event occurs, such as a completed call |
| System of record | The single system treated as authoritative for a type of data, such as your booking calendar |
| Field mapping | The agreed correspondence between what the call captures and where each value is stored |
| Data intermediary | An organisation processing personal data on behalf of another, a concept used in Singapore data protection law |
| Sub-processor | A further supplier used by your supplier, which forms part of your data path |
| Concurrency | How many calls can be handled at the same time without queueing |
| Usage-based pricing | Cost driven by volume, such as minutes or calls handled, rather than a flat headcount charge |
| Configuration scope | The number and complexity of call flows built for you, which typically affects setup effort and cost |
| Porting | Moving a telephone number from one provider to another, distinct from forwarding it |
How these translate into an actual quotation varies by supplier. Our own approach is described on the pricing page, and the evaluation questions are in choosing an AI receptionist.
Terms to treat carefully
A few words are used so loosely that they carry almost no information on their own. When you hear them, ask for the underlying behaviour.
- Human-like: ask what happens when a caller interrupts, mumbles or changes their mind
- Fully automated: ask which calls escalate, and confirm the exclusions cannot be switched off
- Unlimited: ask what happens at peak volume and how that appears on an invoice
- Instant setup: ask what is actually configured on day one, and what is not
- Integrates with everything: ask specifically about your calendar and CRM, by name
- Enterprise grade: ask what it means here, in terms you can verify
- Compliant: ask what the supplier is claiming about their service, and what remains your responsibility
One term we use precisely. Hard escalation means a category that always reaches a person: safeguarding matters involving children, clinical triage and symptoms, security incidents, allergen confirmation, callers in distress, grievances and personal HR matters, and regulated legal, medical, financial or immigration advice. These are fixed in every deployment and are not configurable.
Frequently asked questions
Is an AI receptionist the same as an IVR?
No. An IVR asks the caller to choose from a menu; a voice AI system lets them say what they need in their own words and acts on it. Some deployments combine both.
What is the difference between a transcript and a summary?
A transcript is everything that was said. A summary is the structured outcome your team actually needs, and it usually contains less personal data, which matters for retention.
Why does latency matter so much?
Because callers interpret silence as a dropped line. A delay that looks trivial on a specification sheet changes how the whole call feels.
What does containment rate really tell us?
Only how many calls ended without a person, not whether that was right. Read it alongside the outcomes described in measuring call handling performance.
Is a hallucination a bug?
It is a known characteristic of language models, which is why answers on price, availability and policy should come from your systems and rules rather than being generated freely.
Do we need to understand any of this to buy well?
Not really. Ask what the caller experiences, ask what happens when things go wrong, and ask to hear a real call. The vocabulary is useful mainly for reading supplier documentation.
Related reading
The rest of the reference material is in the AI receptionist resources library.
Skip the vocabulary and hear a call
The fastest way to understand any of this is to listen to one enquiry handled end to end, including the point where it reaches a person.