Plain definitions of the 43 terms that come up most when businesses buy AI phone agents, automation and dashboards. Each one links to the page where we explain how we build it.
Software that answers a business phone line in a natural voice. It handles common questions, books appointments, takes messages and transfers callers to staff with a short summary. Unlike voicemail, it talks with the caller. Unlike a phone menu, callers just say what they need. Self-serve plans are priced monthly; custom builds are a setup fee plus a monthly fee.
An AI system that holds spoken conversations over the phone. The term covers inbound answering, outbound calls such as appointment confirmations or missed-call callbacks, and agents that open tickets or work orders as they talk. An AI receptionist is one kind of voice agent: the one at the front desk.
Another name for a voice agent that runs on a business phone line. In a multi-location business, a phone agent usually needs each location’s own hours, prices and transfer rules, so a caller to one store hears that store’s details. Trilix uses the term for the agent it builds and audits for each location a client runs.
See: AI phone agent
AI that understands and replies to people in natural language, by voice or text, instead of through fixed menus or forms. It combines speech recognition or text input, a language model that decides what to say, and, for voice, speech synthesis. Phone agents, chat assistants and messaging agents are all conversational AI.
Answering calls when the business is closed: nights, weekends and holidays. The options are voicemail, a live answering service billed per minute or per call, or an AI receptionist that answers every call, books where it can and routes true emergencies to the right person. After-hours calls are often the ones that go to a competitor.
See: AI receptionist
An automatic text message sent to a caller whose call went unanswered, usually with a link to book or a promise of a callback. It recovers some missed calls but relies on the caller replying. Answering the call in the first place recovers more. To see what unanswered calls cost a business each month, use a missed call calculator.
Handling a call without a staff member picking up, by answering the question automatically or moving the caller to a text, link or booking page. Good deflection takes the repeat questions (hours, prices, wait times) off the team. Bad deflection traps callers who need a person, so every deflection flow should end in an easy handoff.
See: AI phone agent
The point where an AI agent passes a conversation to a person. A warm handoff transfers the caller with a short summary, so nobody has to start over. Handoff rules decide when it happens: the caller asks for someone, the request falls outside approved rules, or the topic is sensitive. Trilix agents transfer callers who ask for a person.
See: AI phone agent
The classic phone menu: “press 1 for service, press 2 for billing.” IVR routes calls with keypad presses or simple voice commands but can’t hold a conversation or answer open questions. An AI receptionist replaces the menu with a conversation, so callers say what they need in their own words and get an answer or the right person.
Technology that turns spoken words into text, also called speech recognition or transcription. In a voice agent it’s the first step: the caller speaks, speech-to-text writes it down, and a language model reads it. Accuracy drops with background noise, accents and poor phone lines, so the choice of speech model matters for busy shops and service lanes.
Technology that turns written text into a spoken voice. In a voice agent it’s the last step: the language model writes a reply and text-to-speech says it. Modern voices sound natural and can match a brand’s tone. Callers should still be told they are speaking with an automated agent.
The delay between a caller finishing a sentence and the agent starting to reply. It adds up across speech-to-text, the language model and text-to-speech. High latency causes awkward pauses, and callers start talking over the agent. Builders trade off speed and capability: faster models for simple calls, stronger ones where the rules are strict.
The legal rules on recording phone calls. In the US, some states, including California, Florida, Pennsylvania and Washington, require consent from everyone on the call, while others allow one party to consent. Canadian privacy law expects notice and consent when a business collects personal information. A short disclosure at the start of each call covers most cases.
See: Data and security
A scheduled review of an AI agent’s calls to catch mistakes before customers do. It can be done by people or by a second AI that checks the first. At a multi-store quick lube in Texas, a Trilix audit agent reviews calls every 2 hours during shop hours and emails the owner only when something looks off.
An AI model trained on large amounts of text to read and write language. LLMs power chat assistants, voice agents and document tools: they understand a request, decide what to say and draft replies. They are good at language and judgment calls, but they can be wrong, so business systems pair them with rules, sources and human checks.
Software that uses an AI model to carry out tasks, not just answer questions: looking something up, filling in a booking, opening a ticket or sending a follow-up. Agents act within rules set by the business. Sensitive actions, like refunds or postings to the books, can be set to wait for a person’s approval.
A way to make an AI answer from a company’s own documents. When a question comes in, the system first retrieves the most relevant passages from manuals, procedures or past jobs, then gives them to the language model to write the answer. Because the sources are known, the answer can show where it came from, which makes it easier to check.
See: Knowledge assistants
An AI assistant that answers staff questions from a business’s own manuals, procedures and records, usually built with RAG. A good one shows the source for each answer and says so when it has no verified answer. Typical uses are shop-floor procedures, onboarding new hires and finding past jobs.
When an AI model states something false with confidence, such as a price, policy or fact that isn’t in its sources. It happens because language models predict likely text. The usual defences are answering only from approved sources, showing citations, saying “I don’t know” when nothing matches, and audits that catch mistakes.
The rules and checks that keep an AI system inside approved behaviour. Examples: a written script of what an agent may say, book or refuse; topics it must hand to a person; approval steps before money moves; and logs of every action. Guardrails are set before launch and tested on simulated cases.
See: Data and security
A design where a person reviews or approves certain AI actions before they take effect. In back-office automation, that often means a review queue: the system flags exceptions and a bookkeeper approves or fixes them before anything posts. It keeps people in charge of sensitive steps while the system does the checking.
See: Workflow automation
AI that interprets images and video: detecting vehicles, people or objects, reading their position and tracking them over time. In operations it turns camera footage into numbers, such as how many cars came in, how long each spent in a bay, or whether a shelf is empty. It usually runs on cameras a business already has.
Reports built from security or operations cameras using computer vision. For auto service and car washes, that means car counts by bay or lane, time in the bay, peak hours and checks that a service on the ticket happened. Accuracy depends on camera angles and lighting, so counts should be tested against POS tickets during setup.
Software that records and grades conversations between staff and customers, on the phone or at the counter, so managers can coach from specifics. It can score whether an offer was made and how a service was explained. Raw recordings can be limited to the employee in them, with managers seeing graded summaries.
Software that does the repeat steps between systems: pulling data from one tool, checking it and putting it in another, sending the follow-up or building the report. Many steps are plain rules, which are cheap and never guess. AI handles the parts that need judgment, like reading documents or summarising calls.
See: Workflow automation
An automation that takes each location’s daily sales, payments and fees from the point-of-sale system and posts them into QuickBooks, mapped to the right accounts and locations. The common approach is one summarized entry per site per day, with card deposits reconciled against POS totals and exceptions flagged for a person.
Matching the card deposits that land in the bank against the card sales the POS says were made, by day and location. Gaps point to processor fees, chargebacks, batching errors or missing deposits. Done by hand it is slow; automated, it runs daily and leaves a short list of mismatches for a person to check.
An application programming interface: the official way one piece of software lets another read or write its data. A POS, scheduler or accounting tool with an API can be connected directly. Without one, the fallback is usually a scheduled export. Whether a system has a usable API is one of the first questions in any integration.
See: Integrations
A message one system sends to another the moment something happens, such as “a booking was made” or “a payment failed.” Webhooks let automations react right away instead of checking on a schedule. Not every system offers them, so many integrations combine webhooks, API calls and scheduled exports.
See: Integrations
A file a system produces on a timetable, such as a nightly sales report from the POS, that another system then reads. It’s the usual route when software has no API. Exports are reliable and simple, but the data is only as fresh as the last file, so they suit daily reports better than live answers.
See: Integrations
A connection that lets a system see data in another tool without being able to change anything there. It’s the safest starting point for any integration. For example, an assistant can read project boards and team chat to know what’s happening, while all changes stay with the people who use those tools.
See: Data and security
Software that holds customer records, contact history, leads, quotes and follow-ups in one place. Off-the-shelf CRMs suit standard sales processes. A custom CRM is shaped around how a business actually sells and delivers, with pipelines, scheduling and invoicing that match its jobs instead of generic stages.
One screen that shows every store, clinic or site side by side, built from the POS, phones, payroll, cameras and other systems. Good dashboards compare locations on rates and averages, not raw totals, flag exceptions as data arrives and send each manager the view for their own location on a schedule.
See: Executive dashboard
A record of every action a system takes: calls answered, bookings made, alerts sent and automated changes, with the time and what triggered them. It lets anyone trace a result back to its cause and is often required in security reviews. With AI systems it also shows whether a person approved a step.
See: Data and security
Setting what each person can see and do by their role. An owner sees the whole group, a store manager sees their store, and an employee sees only their own records or recordings. It limits who can reach sensitive data, like payroll, without building separate systems for each team.
See: Data and security
Where data is physically stored and processed, usually stated as a country or region. Contracts, regulators or customers sometimes require that data stays in a particular place, such as Canada or the EU. Trilix systems can be hosted in Canada, the US or Europe, or inside a cloud account the client owns.
The Personal Information Protection and Electronic Documents Act, Canada’s federal privacy law for private-sector businesses. It requires meaningful consent to collect personal information, use only for the purposes identified, reasonable safeguards and access for individuals to their own data. Some provinces, such as Quebec, Alberta and British Columbia, have their own similar laws.
See: Data and security
The General Data Protection Regulation, the European Union’s data protection law, in force since May 2018. It applies to organisations that handle personal data of people in the EU, wherever the organisation is based. It requires a lawful basis for processing, gives people rights over their data and allows large fines. The UK has its own version, the UK GDPR.
See: Data and security
An audit framework from the American Institute of CPAs. An independent auditor reports on a company’s controls for security and, optionally, availability, processing integrity, confidentiality and privacy. A Type I report checks the design of controls at one point in time; a Type II report tests them over a period. Ask any vendor for its report and scope.
The Telephone Consumer Protection Act, the US law on automated calls and texts. In 2024 the FCC ruled that AI-generated voices count as “artificial” voices under the TCPA, so robocalls that use them need the called person’s prior express consent. Outbound AI calls should only go to people who have agreed to hear from the business.
The average amount a customer spends per visit, called average repair order (ARO) in repair shops. It rises when advisors offer the right add-ons, such as filters, wipers or fluids, on every visit. Tracking it by store and by advisor shows whether offers stopped or customers started saying no.
See: Upsell AI
How long a vehicle spends in a service bay, from pull-in to pull-out. It is one of the main capacity measures at quick lubes and repair shops: long bay times mean fewer cars per hour and longer waits. Camera analytics can measure it for every car, by bay, hour and shift.
The number of vehicles a store services or a car wash washes, by day, hour, bay or lane. It is the basic traffic number for auto service and wash operators. Comparing camera car counts with POS tickets shows whether every car that came in was rung up.
See: Car wash analytics
On a 30-minute call we show you working systems for businesses like yours, in plain language, and sketch what we’d build first.
