Glossary
AI, technology and maintenance, from A to Z, explained with field service examples.
The words AI brings into the office of a maintenance, HVAC or installation company, from hallucination to RAG and from IVR to voice agent, and the trade’s own, from service request to MTTR. Each term in one or two sentences, with a field service example and, where there is one, the guide that explains it in depth.
A
AI agent
A program you hand a task to: it has a goal, tools and rules, decides the steps, takes them and checks the result. What sets it apart from a chatbot is that it acts: it looks things up and writes in the company’s systems. More in what is an AI agent.
AI receptionist
A voice agent that takes the company’s incoming calls, the way the person at the front desk would. Also called a virtual receptionist.
AI voice agent
An AI agent that answers the phone: it understands what callers say in their own words, asks for what’s missing and leaves the job logged. Also called a voicebot or AI receptionist. Explained in AI voice agent.
API
The door through which one program talks to another. If your ERP or field service software has an API, the agent can find the customer, check their contract and open the work order without anyone typing it; if it doesn’t, another way in is needed, like an export or email.
Autonomous agent
An agent that takes actions without asking for permission at each step. In field service it makes sense for what can be undone, like replying or asking for a photo; what can’t be undone, like sending the on-call technician or cancelling a visit, is better approved by a person: that’s human in the loop.
B
Barge-in
When the caller starts talking while the voice agent is still speaking. A good agent stops and listens, as a person would; a bad one carries on with its sentence.
C
Callbot
A voice bot built for scripted calls, almost always outbound: confirming an appointment, reminding a customer of a maintenance visit or running a survey.
Chatbot
A program that replies in writing, on a website or on WhatsApp. The old ones followed a tree of fixed answers; today’s understand normal language. If it also does things in the company’s systems, it’s an agent.
Cloud phone system
The company’s phone system, but in the cloud: numbers, extensions, call forwarding and recordings live online and are used from a computer or a mobile phone. Also called a virtual PBX.
CMMS
Computerized maintenance management system: the software that organizes the maintenance of equipment, with its history, the preventive maintenance plan, work orders, spare parts and KPIs. More in what is a CMMS.
Context window
How much text the model can take into account at once: the instructions, the conversation and the documents it is given. What doesn’t fit, it doesn’t see.
Corrective maintenance
Maintenance that repairs equipment once it has already failed. It’s what comes in as a service request.
D
Data processor
Under the GDPR, whoever handles personal data on behalf of another company. The provider of an AI agent is one: it handles your customers’ names, phone numbers and addresses, so a data processing agreement (Article 28 of the GDPR) is needed, saying what it does with them and how long it keeps them.
Dispatch
Deciding which technician goes to each job and when, based on urgency, area, who is on call and who knows the site.
E
Embeddings
A way of turning text into numbers that keep its meaning. It lets the agent find “leak in the circuit” in the manual when the customer said “the boiler is losing water”.
ERP
The software that runs the whole company: purchasing, sales, inventory and accounting. Many have a field service module with service requests, work orders and service reports.
EU AI Act
The European Union’s regulation on artificial intelligence. Its Article 50 requires that people know when they are talking to an AI; on a call, that’s said at the start. It applies from 2 August 2026.
Evaluation
Testing the agent with the company’s real cases before switching it on, and measuring how many it handles well: whether it asks for what’s needed, recognizes the customer and sets the right urgency. It’s repeated with every change, so fixing one thing doesn’t break another.
F
Field service management (FSM)
The software a field service company uses to run each job from start to finish: the service request, the work order, technicians’ schedules, the visit, the signed service report and the invoice. What it does and how to choose one in field service management software.
Fine-tuning
Training an existing model further with your own examples so it learns a task or a way of answering. It changes the model, but it isn’t how it learns facts that change, like rates or on-call rotas: that’s what RAG is for.
First-time fix rate
The share of jobs with a visit that are closed without going back: fixed on the first visit divided by jobs with a visit. In Aquant’s 2025 benchmark of 157 companies the median was 75%, and the top 20% reached 86%.
G
GDPR
The European Union’s General Data Protection Regulation. An agent that takes service requests handles personal data, so callers must be told if the call is recorded and why, and a contract must be signed with the provider, who is a data processor.
Guardrails
The rules that limit what the agent can say or do, whatever happens in the conversation: no safety instructions of its own, no price or time the office hasn’t confirmed, and a handoff to a person if someone mentions a gas smell.
H
Hallucination
When the model gives a fact that sounds right but is false: a part that doesn’t exist, made-up opening hours or a price nobody approved. It’s avoided by making it check the ERP and the company’s documents before stating anything, and say “I don’t know” when it can’t find it.
Human handoff
When the AI agent passes the conversation to someone on the team: because the customer asks, because it can’t solve it or because there’s a risk. If the caller mentions a gas smell, sparks or water in an electrical panel, the right thing is to put the call through to the person on call straight away.
Human in the loop
A person reviews or approves what the agent proposes before it happens. In field service, the usual setup is that the agent prepares the service request or work order and the office approves it with one click.
I
Instructions
The text that tells the model who it is, how it should speak, what it must ask and what it must never do. Also called the system prompt. For a field service company they’re written with the office: how a request is taken, what is urgent and who gets told.
IVR
Interactive voice response: the “press 1 for repairs, 2 for billing” menu. It routes calls between departments, but it doesn’t solve anything: at the end of every branch there’s a person or a voicemail box.
K
Knowledge base
What the company gives the agent to look things up: rates, service areas, hours, contracts, equipment manuals and what to do in each emergency. It’s what makes it answer like your office and not like a generic assistant.
L
Large language model (LLM)
The piece of AI that understands text and decides what to answer and what to do. It’s the part that can make up a fact, which is why a serious agent checks the company’s systems before stating anything.
Latency
The time the agent takes to start answering after you stop talking. On a call it’s what people notice most: if it’s slow, they repeat themselves or think the line dropped.
M
MCP
Model Context Protocol: an open standard for connecting AI agents to tools and data, so one connection works for several assistants.
MTBF
Mean time between failures: how long equipment runs without breaking down. Operating time divided by the number of failures. Higher is better. How to calculate it in maintenance KPIs.
MTTR
Mean time to repair: how long it takes to fix a breakdown. Total repair time divided by the number of repairs. Lower is better.
Multimodal
A model that understands more than text: photos, audio and documents. In field service it reads the photo of an equipment nameplate or the error code on a display that a customer sends on WhatsApp.
O
OCR
Optical character recognition: reading the text in an image or a scanned PDF, like a delivery note, an invoice or an equipment nameplate.
On-call
The technician or person who handles emergencies outside office hours: nights, weekends and holidays.
P
PM compliance
The share of scheduled preventive maintenance done on time: visits done on time divided by visits scheduled. Higher is better.
Predictive maintenance
Maintenance that measures the condition of the equipment (temperatures, pressures, vibration, power use) and acts when the data calls for it, just before it fails.
Preventive maintenance
Maintenance that services equipment on fixed dates, by calendar or by hours of use, so it doesn’t fail. Often shortened to PM.
Process automation
Having software do a repetitive office task. The traditional kind (RPA) follows fixed steps, like copying a field from one screen to another, and breaks when something changes; an AI agent understands the case and decides the steps, so it copes with an email written differently or a request with a detail missing.
Prompt
The text a model receives before it answers: the instructions, the context, the examples and the question. With a vague prompt, the model fills what it doesn’t know with what’s most likely, like a price nobody gave it.
R
RAG
Retrieval-augmented generation: before answering, the agent searches the knowledge base for the passages related to the question and answers from them, instead of from memory. It’s what stops it making up a rate or opening hours.
Repeat visit
Going back because the job wasn’t solved the first time: a part was missing, the technician didn’t know which unit it was or couldn’t get in. It’s the most hidden cost in field service.
Response time
How long the technician takes to arrive from when the request comes in. Lower is better, and in contracts with an SLA it’s usually in writing.
S
Service report
What the technician fills in at the end of the visit: what they found, what they did, what parts they used, how long it took and who signed it off. The work order opens the job and the service report closes it.
Service request
The problem the customer reports: “the boiler won’t fire”, “the cold room is beeping”. It’s where all the work of a field service company starts, and how complete it arrives decides whether the technician fixes it the first time.
SIP
The protocol calls travel over on the internet. A SIP trunk brings calls to your existing number into the phone system or the voice agent, without changing number.
SLA
Service level agreement: the response time agreed in the contract, for example four hours for an emergency and twenty-four for everything else.
Speech recognition
The piece that turns what the caller says into text. When it fails, it mixes up brands, models, street names or error codes, especially with background noise.
T
Text-to-speech
The piece that turns the agent’s answer into a voice. When it fails, it sounds robotic or reads numbers and addresses wrong.
Token
The chunk of text a model works with: roughly a piece of a word. Models are billed by tokens read and written, and the context window is measured in tokens.
Tools
The actions the agent can ask for: look up a customer in the ERP, check the schedule, open a service request, send a WhatsApp message. The model decides which one to use and with what data, and the company’s system runs it. Also called function calling.
Turn detection
The piece of a voice agent that notices when you’ve finished speaking. When it fails, the agent cuts in mid-sentence or goes quiet; it’s one of the things that most decides whether a call sounds natural.
V
Vector database
Where the embeddings of documents are stored, so they can be searched by meaning and not only by exact words.
Voicebot
The general name for a bot that talks on the phone. If it also looks things up and writes in the company’s systems, it’s an AI voice agent.
W
Webhook
An automatic message one program sends another when something happens: an urgent request comes in and a message lands in the office’s Teams or Slack channel, without anyone sending it by hand.
WhatsApp 24-hour window
On WhatsApp, the 24 hours after the customer’s last message. Inside it, the company can reply with free text; outside it, only with approved templates.
WhatsApp Business API
The version of WhatsApp for software: the one that lets an AI agent and the whole team reply from one number, connected to the ERP. Unlike the app, Meta charges for it per message delivered.
WhatsApp template
A message approved by Meta, the only kind a company can send to message a customer first or outside the 24-hour window: “your appointment is tomorrow at 10”, “the technician is on the way”.
Work order
The document the office uses to have a job done: which customer, which equipment, what’s wrong, with what priority, which technician goes and when. It’s opened by a service request or a preventive visit and closed with the service report. Guide and template in work order.
Where Delegic fits
Delegic puts AI agents in the office of maintenance and installation companies: they take the service request by phone, WhatsApp, email or web chat, ask what the technician needs to know and leave the work order in your ERP or field service software, for the office to approve. More in what is Delegic.