Glossary
Plain-language definitions.
The AI, CX, and operations vocabulary an SME buyer needs in 2026, written without buzzwords. 107 terms, growing as the field does.
A
16 terms- AI receptionistAn AI agent that answers your phone, identifies the caller, books or routes, and writes the result into your system of record, without staffing a desk for it.
- AI phone answering serviceA managed service where AI agents answer business calls in your brand voice, take action in your tools, and escalate to a human only when needed.
- AI voice agentAn AI system that conducts real-time spoken conversations, calls tools, and writes results back to your systems, the same way a trained human agent would.
- After-hours answeringPicking up calls outside of business hours, evenings, weekends, holidays, without paying a human team to be on standby.
- Agentic AIAn AI system that not only answers but takes action, calling tools, writing to systems, and following a multi-step plan to resolve a request.
- AI appointment bookingAn AI agent that takes a customer through scheduling, pulling real availability, applying constraints, confirming the slot, and writing the booking back to your calendar.
- AI for clinicsAI agents purpose-built for independent practices, dental, urgent care, and specialty clinics, booking, rescheduling, intake, and after-hours triage that writes into your EHR.
- AI for auto shopsAI agents that handle service-booking calls for independent shops and dealer service departments, pulling the VIN, checking bay availability, and lining up parts before the writer takes the call.
- AI for home servicesAI agents for HVAC, plumbing, electrical, and roofing crews, triaging after-hours emergencies, quoting trip charges, and dispatching the nearest tech without waking a dispatcher.
- AI tool useThe pattern where an AI model calls structured functions, to read or write external systems, rather than only producing text. The basis of all agentic AI.
- AI guardrailsRules and runtime checks that constrain what an AI agent is allowed to do, what it can say, which tools it can call, and what it must escalate to a human.
- AHT (average handle time)The average duration of a customer interaction, from connect to wrap-up. The classic call-center efficiency metric, and a misleading one if it is the only number you track.
- Abandonment rateThe percentage of inbound calls or chats where the customer hung up before the agent could connect or resolve. The single biggest hidden cost in any phone-driven business.
- Average speed of answer (ASA)The average time from when a customer's call connects to when an agent (human or AI) starts handling it. The metric that most directly reflects whether the customer is being made to wait.
- ASR confidence scoreThe probability the speech-to-text layer assigns to its own transcription. A useful signal when the agent should pause and confirm, ignored entirely by most vendors.
- AI Act conformity assessmentThe audit-like process EU AI Act requires for high-risk AI systems before they can be deployed in the EU market. Where the documentation paperwork the regulation invents actually lands.
B
2 terms- Barge-inWhen the caller starts talking while the agent is still speaking. A useful voice agent stops mid-sentence and listens. A bad one keeps talking.
- Business associate agreement (BAA)The HIPAA-required contract between a healthcare provider and any vendor that handles PHI on their behalf. The single document that turns 'HIPAA-aligned' marketing into actual HIPAA coverage.
C
13 terms- Call deflection rateThe percentage of inbound calls an AI agent fully resolves without escalating to a human. A common, and frequently misused, vendor metric.
- Conversational AIThe broader category of AI systems that conduct natural-language conversations, covering chat, voice, and email, usually built on large language models in 2026.
- CRM-as-source-of-truthThe architectural choice to keep the customer record in your CRM, not in a vendor's data layer, and have every agent action write into that CRM natively.
- CSAT (customer satisfaction score)A short post-interaction survey, usually a single rating from 1–5, that measures whether the customer felt the interaction went well.
- Containment rateThe percentage of interactions that stay within the AI agent and never escalate to a human. A more conservative cousin of deflection rate.
- Call routingSending an inbound call to the right destination, a team, a queue, an after-hours line, based on who is calling and why.
- CCPAThe California Consumer Privacy Act. The American privacy framework with the broadest reach, applies to most US businesses handling California-resident data.
- Customer effort score (CES)A short survey asking how easy it was to get an issue resolved. Better correlated with retention than CSAT in many studies, because low effort means low friction, and friction is what drives churn.
- Chatbot vs AI agentChatbots answer. AI agents act. The distinction looks subtle on a marketing page and matters enormously in production.
- Contact volumeThe total count of customer interactions over a period, calls, chats, emails. The base number every CX team plans against, and the one AI agents most directly change the shape of.
- Conversation memoryWhat the agent remembers about a caller across calls, preferences, prior issues, household members, vehicle history. The thing that turns a transactional bot into a relationship.
- Change data capture (CDC)The integration pattern where downstream systems subscribe to a stream of changes from the source database, instead of polling for them. The right way to keep a CX-AI agent's view of the world fresh without hammering APIs.
- CPT codeCurrent Procedural Terminology, the code set that describes the procedures, services, and visits a clinic bills for. The pairing of CPT (what was done) and ICD-10 (why) is how clinical billing works.
D
5 terms- Deterministic tool useA pattern where an AI agent can only call tools that are pre-compiled, typed, and replayable, not invented at runtime.
- Dispatch softwareThe system home-services companies use to schedule and route field technicians. ServiceTitan, Housecall Pro, and Jobber are the canonical examples.
- Data residencyThe contractual guarantee that customer data is stored and processed in a specified geographic region. The hard requirement for serving regulated customers across borders.
- Dialogue stateThe running snapshot of what the agent knows so far, who the caller is, what they want, which slots are filled, what has been confirmed. The conversational equivalent of a working memory.
- Data processor vs controllerThe GDPR distinction between the company that decides why and how data is processed (controller) and the company that processes it on their behalf (processor). A CX-AI vendor is almost always the processor.
E
5 terms- EHR integrationA connection between the AI agent and a clinical record system (Epic, Athena, NexHealth, eClinicalWorks) that allows the agent to read patient history and write back into the practice schedule.
- EndpointingThe decision of when the caller is finished speaking. The single biggest determinant of whether a voice agent feels natural, and one of the hardest problems in voice.
- EU AI ActThe European Union's regulation on AI systems. High-risk classifications carry substantial compliance overhead, and the rules apply to any AI vendor serving EU customers, regardless of where the vendor is based.
- Escalation rateThe percentage of AI-handled interactions that get transferred to a human. Read alongside resolution rate, not on its own, or you will reward the wrong behavior.
- Entity extractionPulling structured values out of a caller's sentence, names, dates, phone numbers, vehicle VINs, dollar amounts, so the agent can act on them.
F
2 terms- Function callingThe protocol by which an AI model invokes typed tools, checking availability, sending a refund, looking up a record, instead of only producing text. The foundation of every agentic system.
- First contact resolution (FCR)The percentage of customer issues resolved in the first interaction, without a follow-up. The single most predictive metric for customer satisfaction in support operations.
G
1 termH
4 terms- HIPAA complianceThe U.S. healthcare data law that governs how protected health information (PHI) is collected, stored, and shared. Any AI vendor handling clinic calls must be HIPAA-aligned and have a signed BAA.
- Human handoffThe moment an AI agent transfers a conversation to a human teammate, ideally with context, recommended next steps, and a running SLA clock.
- Hallucination (AI)When the AI confidently produces information that is wrong, invented dates, made-up policies, fabricated phone numbers. The single largest reliability problem in production AI.
- Hold timeThe duration the caller spends on hold during an interaction. The most-hated single number in customer service, and the one AI agents collapse to near zero.
I
7 terms- IVR replacementReplacing a press-1-for-billing interactive voice response menu with an AI agent that understands plain spoken language and resolves the call directly.
- ISO 27001The international standard for information security management systems. Common globally, especially in Europe, increasingly expected alongside SOC 2 for cross-border CX-AI deployments.
- ISO 42001The AI-specific extension of ISO 27001, a management system standard for organizations that build or operate AI systems. The compliance signal that separates serious AI vendors in 2026.
- iPaaSIntegration platform as a service. Zapier, Make, Workato, Mulesoft. A useful middle layer for low-volume sync work, and a layer to keep off your AI agent's critical path.
- Intent classificationIdentifying what the caller is trying to do, book, cancel, complain, ask a price, so the agent can route or resolve. The first reasoning step on every call.
- Idempotency keyA unique token attached to a write request so the receiver can safely deduplicate retries. The cheap mechanism that prevents an AI agent from accidentally booking the same appointment twice.
- ICD-10 codeThe international diagnosis-classification code system used for billing, reporting, and clinical analytics. The taxonomy a clinical AI has to map free-form patient language into accurately.
K
1 termL
1 termM
3 terms- Missed call recoveryCatching the calls a business would otherwise lose, to voicemail, to busy signals, to after-hours, and converting them into bookings or follow-ups instead of dropped revenue.
- Mindbody integrationA native connection between the AI agent and Mindbody, the dominant practice management system for salons, spas, fitness studios, and wellness clinics.
- Model Context Protocol (MCP)An open standard for connecting AI models to external tools and data sources. A useful coordination layer, but not a replacement for serious tool engineering.
N
4 terms- NPS (net promoter score)A loyalty metric that asks "how likely are you to recommend us to a friend" on a 0–10 scale. Used to track whether AI handling helps or hurts the customer relationship over time.
- Natural language understanding (NLU)The umbrella term for the parts of a system that interpret what the user means, intent, entities, sentiment, context. In 2026, mostly absorbed into the LLM itself.
- Natural language generation (NLG)The flip side of NLU, producing the agent's spoken or written reply from structured intent, slot values, and context. Now the default output of any modern LLM.
- No-show rateThe percentage of booked appointments where the patient or customer simply does not show up. The single biggest hidden cost in any appointment-driven business, and the metric outbound AI moves the most.
O
5 terms- Outbound voice AIAI agents that place calls, appointment reminders, no-show follow-ups, parts-ready notifications, win-back outreach, rather than only answering inbound traffic.
- OmnichannelThe pattern where the same customer relationship is handled coherently across phone, chat, email, SMS, and WhatsApp, with shared context, not separate silos per channel.
- Occupancy rateThe percentage of an agent's logged-in time they spend actually handling interactions, as opposed to waiting between them. The classic workforce-management efficiency metric.
- OAuth scopeThe mechanism by which an integration declares which permissions it actually needs, read appointments, write bookings, instead of asking for full access. The difference between least-privilege and a footgun.
- On-the-way notificationThe text or call that goes out when a home-services tech is heading to the customer's address. The single cheapest improvement to home-services CSAT, and a natural job for outbound AI.
P
7 terms- Practice management systemThe software a clinic uses to run the front desk, scheduling, intake, billing, and reporting. NexHealth, Mindbody, Athena, and Epic Practice Management are common examples.
- Pay per resolved caseA billing model where the vendor charges only for interactions that actually closed something in the system of record, not for seats, minutes, or "platform fees."
- Prompt injectionAn attack where a malicious caller embeds instructions in their input that try to hijack the agent's behavior. The dominant security concern for production LLM agents.
- PCI DSSThe payment card industry's data security standard. The framework that governs anything handling credit card data. The right architectural choice for a CX-AI vendor is to stay out of scope entirely.
- ProsodyThe rhythm, stress, and intonation of speech. The musical layer of a conversation that conveys urgency, sarcasm, doubt, warmth, the things the words alone do not say.
- Protected health information (PHI)Patient data covered by HIPAA, name, date of birth, medical reason, anything that ties a person to a health event. The category every clinic AI vendor has to handle correctly or not handle at all.
- Patient intakeThe process of collecting a new or returning patient's information before their visit, demographics, insurance, reason for visit, medication list, consent forms. The front-desk work an AI receptionist handles end-to-end.
Q
1 termR
6 terms- Resolution rateThe percentage of interactions where the customer's actual problem was solved, not just answered. The most honest single number for measuring AI agent quality.
- Retrieval-augmented generation (RAG)The technique of pulling relevant documents into the model's context at runtime, so the agent answers from a real source instead of model priors. The standard way to ground knowledge.
- Right to be forgottenThe data subject right, under GDPR, CCPA, and similar laws, to have their personal data deleted from a vendor and its sub-processors on request. The capability most CX-AI vendors implement halfway.
- Rate limitThe cap on how often an integration can call an external API in a window. The invisible ceiling that bottlenecks every agent integration when traffic spikes.
- Retry policyThe rules that govern how an agent reacts to failed external calls, what to retry, how often, with what backoff. The difference between a graceful integration and a thundering herd.
- Repair order (RO)The document that captures the work a vehicle is in the shop for, customer concerns, services to perform, parts, labor, total. The auto-shop equivalent of a clinical encounter note.
S
13 terms- Service writerThe auto-shop role that takes the customer-facing call, scopes the work, and books the vehicle into the shop. The biggest single bottleneck the AI agent removes.
- ServiceTitan integrationA native connection between the AI agent and ServiceTitan, the dominant dispatch and operations platform for home-services businesses.
- Speech-to-text (STT)The component of a voice-AI pipeline that converts audio into text the model can reason over. The first link in the chain, and one of the cheaper ones to optimize.
- Speech-to-speech modelsEnd-to-end models that take audio in and emit audio out, skipping the STT-LLM-TTS pipeline. The promise of the next generation, and not quite ready for production yet.
- SOC 2 Type IIA security audit that examines whether a vendor maintains the security controls they claim, sustained over a period of months, not at a single point in time. The baseline trust signal for any SaaS in 2026.
- Sentiment analysisReal-time classification of the customer's emotional state during a conversation, positive, neutral, negative, escalating. A useful signal when wired into the right action, useless when it is just a number on a dashboard.
- Self-serviceLetting the customer accomplish things, booking, status check, account change, without talking to a human. The right design pattern for the routine 80%. The wrong pattern for the rest.
- Service level agreement (SLA)A contractual promise about service quality, response time, uptime, latency, with a penalty if the promise is missed. The right place to nail a vendor down on the things that matter.
- ShrinkageThe percentage of scheduled agent time that is unavailable for handling customers, breaks, training, meetings, sick days. The fudge factor every workforce-management plan budgets for.
- Slot fillingThe pattern of collecting the structured values a task needs, name, date, service type, vehicle, until the agent has enough information to act. The bookkeeping of a useful conversation.
- Speaker diarizationIdentifying who said what when multiple people are on the line, the spouse picks up an extension, the patient and the parent are both talking. Solves a class of bugs most voice agents do not handle.
- System of recordThe application that is the authoritative source for a particular kind of data, the CRM for customer records, the EHR for clinical data, the dispatch tool for technician schedules. The system every other view defers to.
- SOAP noteThe standard clinical documentation format, Subjective, Objective, Assessment, Plan, that providers use to record patient encounters. The structure any clinical AI scribe has to write into.
T
4 terms- Text-to-speech (TTS)The component that turns the model's text reply into spoken audio. The voice your customer hears, and the layer where vendors quietly differ most on brand experience.
- Turn-takingThe mechanics of who speaks when in a voice conversation. The thing humans do unconsciously and voice AI has to get right for the call to feel natural.
- TCPAThe Telephone Consumer Protection Act. The US law that governs outbound calling and texting. The compliance regime that punishes vendors for sending the wrong message to the wrong number at the wrong hour.
- Technician utilizationThe percentage of a tech's billable hours they spend on revenue-generating work, on a job, not driving between jobs, not waiting for dispatch. The single biggest profitability lever for a home-services business.
V
4 terms- Voice agent latencyThe time between when a caller stops speaking and when the AI agent starts responding. The single biggest determinant of whether a voice call feels natural.
- Voice authenticationConfirming the identity of the caller during a voice conversation, either by matching their voice biometrics or by stepping them through a knowledge check.
- Voice AI for small businessAI voice agents priced and shaped for SMEs, single-location and multi-location independents, rather than enterprise contact centers.
- Voice activity detection (VAD)The algorithm that decides when the caller has started or stopped speaking. The wrong VAD turns a smart agent into an interrupting one.
W
3 terms- WebhookA real-time HTTP callback from one system to another. The cheapest integration mechanism. Useful for one-way notifications, insufficient for anything that needs a two-way write into a system of record.
- Wrap-up timeThe minutes a human agent spends finishing notes and updating systems after a call ends. The hidden cost of every conversation, and one of the biggest wins from a CRM-native AI agent.
- Work orderThe home-services equivalent of a repair order, the document that captures a job's scope, address, customer, parts, technician, and status. The unit of work a dispatch tool revolves around.
Term we missed? Tell us.
We add new terms when SME buyers ask about them. Email [email protected] and we'll write it up.
The next call doesn’t have to go to voicemail.
Book a thirty-minute demo. We point Vorel at one of your real numbers on the same call.