Learn glossary
AI Automation Glossary
A
Acceptable Use Policy
A short written policy telling staff which AI tools they may use for work, what information must never go into them and who to ask when they are unsure. A good one names the approved tools, lists banned data such as client records and passwords, and says what happens next. It works best alongside an approved alternative, so staff are not left choosing between the rule and getting work done.
Agentic AI
AI that can plan a sequence of steps and call tools on its own to reach a goal, rather than answering one question at a time. That is useful for multi-step work, like preparing a quote from a price list, past jobs and a customer record. The capability does not itself authorize any action, and it does not remove the need for review: what the system may touch, and who signs off, is still a business decision.
AI Agents
Software that uses a model and a set of tools to work through a task, such as looking up an order, drafting a reply or summarising a week of support tickets. The useful scope, the permissions it holds and the review it needs must be defined for each workflow. In Switchboard, staff can create read-only agents just by asking; an agent that changes records is built with you, and a person approves each change.
AI Orchestration
Coordinating the context, instructions, tools and controls an AI task needs, so the model works from the right records with the right permissions. The orchestration layer decides which systems are connected, what each request may read, what gets logged and when a person must approve an action. This is part of what EdgeKeeper provides for your business; it is not a separate product you buy.
AI Readiness Assessment
A measured review of how work gets done and which systems it touches, used to find improvements worth paying for. Edgerton runs on staff computers with consent, records where time goes, and produces written procedures, a findings report and a proposal with estimated payback. It is a business assessment that informs a decision; it is not a compliance certification or a security audit.
AI Receptionist
Software that answers phone calls or web chats, handles routine requests such as booking, rescheduling and opening hours, and passes everything else to staff. A useful one books straight into the scheduling system, hands over smoothly when a caller needs a person and keeps a record of what happened on each call. Judge one by how it handles the calls it cannot complete, not only the easy ones.
B
Business Associate Agreement (BAA)
A contract HIPAA requires between a healthcare provider and any vendor that creates, receives, stores or sends protected health information on its behalf. It sets out how the vendor must safeguard that data, report breaches and return or destroy it. Before patient data goes into any AI tool, ask whether the vendor will sign a BAA and which plans it covers. A signed BAA is necessary, but it does not make a tool safe on its own.
C
Capacity Value
The estimated value of staff time that automation gives back, usually hours saved multiplied by a loaded hourly cost. It is useful for ranking opportunities against each other. It is not a cash saving unless spending actually falls, for example through less overtime or a vacancy you no longer need to fill. More often it shows up as the same team handling more work.
D
Data Export Rights
The terms that describe how you can take your business records and accumulated knowledge with you, in a usable format, if an engagement ends. Check them before you sign any AI service agreement, because leaving is much harder once your procedures live inside someone else's system. With Automated Edge, your data is yours to export; the software and the server remain ours to operate.
Data Retention
How long a vendor keeps the information you send it, including prompts, uploaded files, outputs and logs, and what happens to that information afterwards. Retention terms often differ by plan and by feature, so the free tier and the business tier of the same tool can behave differently. Before using any AI tool for business data, find out how long data is kept, whether it trains models and whether you can request deletion.
H
Human Approval
An explicit decision by an authorized person before a proposed action runs, such as sending an email, updating a record or issuing a refund. The person sees what will change and chooses to approve or reject it. EdgeKeeper ties each approval to one specific action, so approving one change never grants blanket permission for the next one, and the decision is recorded for review.
I
Insurance Eligibility Verification (270/271)
Checking a patient's insurance coverage before an appointment: whether the plan is active, what it covers and what the patient is likely to owe. Electronically, the practice sends a standard 270 eligibility inquiry and the payer returns a 271 response. AI can run these checks in bulk ahead of the schedule and flag gaps, but staff still review unusual plans, frequency limits and anything the response leaves unclear.
M
MAISP (Managed AI Services Provider)
A company that assesses, builds, runs and supports AI systems for other businesses on an ongoing basis, much as a managed service provider does for IT. A good MAISP measures the work before proposing anything, keeps people in charge of changes, maintains what it delivered and lets clients export their data. Some MSPs partner with one rather than building that capability in-house.
Managed AI Service
An engagement that defines who builds, runs, maintains and supports an agreed AI system, rather than leaving a business with a tool and a login. It covers the server, the connections to business systems, monitoring, fixes when something breaks and changes as the business changes. The agreement should also say how you export your data if you ever decide to leave.
P
Private AI
AI set up so a business controls who can reach its data and what the system can do. In practice that means a controlled layer: access scoped to the person asking, requests logged, a person approving changes, and records the business owns and can export. Private does not automatically mean no outside model ever processes a prompt, so ask any vendor exactly where your data goes and who can see it.
Prompt Injection
An attack where instructions hidden in content an AI reads, such as an email, a web page or a shared document, try to make it ignore its rules or reveal information. A model cannot always tell the difference between data and instructions. The practical defences are limiting what an agent can access, keeping agents read-only where possible and requiring a person to approve any action with real consequences.
Protected Health Information (PHI)
Health information that can identify a patient and is held by a healthcare provider, a health plan or one of their vendors. Names, dates of birth, phone numbers, insurance IDs and appointment notes all count when they are linked to care. Pasting PHI into a chatbot with no BAA in place is one of the most common ways a practice exposes patient data without meaning to.
R
Read-Only Agent
An AI agent that can look up and summarise information from connected systems but cannot change anything in them. It can answer a question, build a report or flag an overdue invoice, but it cannot send, edit or delete. Read-only is the sensible default for most business questions. In Switchboard, staff can create read-only agents just by asking, and anything that makes changes goes through human approval.
S
Shadow AI
AI tools staff use for work without the business knowing or approving, most often personal chatbot accounts and browser extensions. It usually starts with good intentions: someone wants to write a reply faster. The risk is that customer, patient or financial data ends up in accounts the business cannot see, control or close. The fix is a clear policy plus an approved tool that is genuinely useful.
W
White Label AI
AI services delivered by one company and sold under another company's brand. For an MSP or IT provider, white label AI means offering assessments, builds and ongoing support to its own clients, with its brand on the portal, the report and the invoice, without building an AI practice from scratch. The agreement should be clear about who supports the client day to day, who owns the client relationship and who is responsible for the client's data.
Workflow
The steps, systems, decisions, handoffs and checks that turn a request into completed work. Answering a customer email, booking a job and chasing an unpaid invoice are each workflows. Looking at the whole workflow, not one task inside it, shows where time actually goes, which steps need a person's judgment and where automation could safely take over the routine parts.