AI customer support is becoming part of how small businesses answer questions, provide service and stay available beyond the hours their teams can cover.
But the terminology can make the decision harder than it needs to be. Words such as automation, live data, knowledge base, capabilities and escalation are often used as though they mean the same thing. They do not.
For a small business, the important question is not simply whether an AI system can produce a response. It is whether the system can provide useful support within boundaries the business understands and controls.
This glossary explains 20 common AI customer support terms in practical language, so founders, operators and support leads can evaluate tools with clearer eyes.
1. AI customer support
AI customer support is the use of artificial intelligence to help answer or handle customer questions.
The phrase can describe different types of systems. Some follow fixed scripts. Others generate responses from business-approved information. Some can take actions through defined capabilities, while others only provide answers.
The useful distinction is not whether a product uses AI. It is what the AI is allowed to do, what information it can use and when a person becomes responsible for the conversation.
Coffey is an AI customer support assistant designed for small businesses and lean teams that need dependable support coverage around the clock.
2. AI customer support assistant
An AI customer support assistant is a system that helps with customer conversations without necessarily replacing the people responsible for support.
An assistant may respond to routine questions, provide information or help a customer take an appropriate next step. The business still defines the information the assistant should use, the capabilities it can access and the situations that require human involvement.
This is different from claiming that an AI system can replace an entire support team. A support assistant is better understood as a support layer that extends the team's availability.
3. Always-on support
Always-on support means customer assistance remains available outside the times when a human team is actively working.
For a small business, this matters when questions arrive across different time zones or outside the team's working day. Always-on does not mean every issue should be handled automatically. It means the first layer of support does not have to disappear whenever the team steps away.
A reliable always-on model should also make room for human escalation when a question needs context, judgement or responsibility.
4. Support coverage
Support coverage describes when and where customers can receive assistance.
A business may have strong human support during normal working hours but limited coverage in the evening, overnight or across other time zones. AI customer support can extend coverage by helping with suitable routine questions when the team is unavailable.
Coverage is not the same as resolution. A system may keep assistance available while identifying matters that should be reviewed by a person.
5. Routine customer question
A routine customer question is a question that can be answered using information the business has already approved and that does not require unusual judgement.
Examples may include questions about information already covered in the company's approved support material. The exact boundaries depend on the business.
Routine does not mean unimportant. A straightforward question can still affect a customer's confidence. The answer should be clear, relevant and grounded in information the business considers reliable.
6. Approved knowledge
Approved knowledge is the information a business has chosen to make available for customer support responses.
It can include the company's accepted explanations, policies, service information and other material the business wants support to rely on. The important point is approval: the business decides what counts as an appropriate source.
Approved knowledge gives a support assistant a defined basis for answering. It also gives the team something concrete to review when an answer needs improvement.
Coffey answers using the approved knowledge provided by the business.
7. Knowledge base
A knowledge base is an organised collection of information used to answer questions.
The terms knowledge base and approved knowledge are related but not identical. A knowledge base is the collection itself. Approved knowledge emphasises the business decision about which information the support assistant may rely on.
A useful knowledge base should be clear enough for customers and maintained well enough for the team to trust it. It should not be treated as a static upload that never needs review.
For a deeper look, see How to Build an Approved Knowledge Base for AI Support.
8. Grounded response
A grounded response is an answer based on a defined source rather than an unsupported guess.
In customer support, grounding matters because a fluent answer is not automatically a correct answer. A response can sound confident while still being unsuitable for the business.
Grounding does not mean every answer will be perfect. It means the business has established what information should guide the response and can assess whether the answer stayed within that source.
9. Hallucination
An AI hallucination is an answer that presents incorrect, unsupported or invented information as though it were true.
The risk is especially important in customer support because customers may rely on what they are told. Businesses should therefore avoid judging an AI support assistant only by how natural its writing sounds.
Approved knowledge, defined capabilities and human escalation are practical ways to keep support more controlled. They do not remove the need for review, but they give the business clearer boundaries.
10. Live data
Live data is current business information that can change over time and may be relevant to a customer's question.
The phrase is broad. It does not mean an AI system should have unrestricted access to every business system or every piece of information. The business needs to decide what current information is appropriate and how the assistant may use it.
Coffey can act on live data through capabilities the business controls. That distinction matters: current information can be useful while access and action remain defined by the business.
11. Capability
A capability is an action an AI customer support assistant is permitted to perform.
The term is intentionally narrower than "anything the system can do." A capability describes a defined area of action that the business has chosen to make available.
For example, a business may decide that an assistant can help with a particular routine support action but should not make decisions that require human judgement. The specific capabilities will depend on the business and its operating model.
12. Permission
A permission is the business's authorisation for a system to access information or perform an action.
Capabilities describe what an assistant may do. Permissions describe the authority granted to it. Both matter when a small team wants automation without losing control.
A trustworthy support model should make it possible for the team to understand what has been permitted and what remains outside the assistant's role.
13. Controlled automation
Controlled automation is automation that operates within boundaries defined by the business.
Those boundaries may include approved knowledge, permitted capabilities and a clear route to human escalation. The goal is not to make automation appear unrestricted. The goal is to make its role understandable and useful.
For lean teams, controlled automation can provide support coverage while keeping responsibility with the people who own the customer experience.
14. Human escalation
Human escalation is the point at which a customer conversation is brought to a person.
Escalation is useful when a question requires context, judgement, empathy or a decision that should not be handled automatically. It is not evidence that the AI system has failed. In a well-designed support model, knowing when to involve a person is part of the system's value.
Coffey escalates to a human when a person is needed.
Read more in When Should AI Support Escalate to a Human? A Clear Guide.
15. Human judgement
Human judgement is the ability to interpret context and make a responsible decision when a standard answer or routine action is not enough.
Customer support often involves more than retrieving information. A customer may describe an unusual situation, combine several issues or need an answer that depends on circumstances.
AI customer support can help keep routine assistance available. Human judgement remains important where the situation needs interpretation, accountability or care.
16. Handoff
A handoff is the movement of a customer conversation from an automated support assistant to a human team member.
A useful handoff should make the next responsibility clear. The customer should not be left uncertain about whether anyone is reviewing the matter, and the team should know when a conversation needs attention.
The exact handoff process will vary by business. The core principle is simple: automation should have a clear boundary at which human responsibility begins.
17. Support layer
A support layer is one part of a broader customer-support operation.
Describing AI as a support layer helps set realistic expectations. It can extend availability, assist with routine questions and operate within defined capabilities, while the wider customer experience still belongs to the business.
This framing avoids the idea that a single tool must replace every part of a human support function.
18. Lean support team
A lean support team is a small team handling customer questions with limited staff capacity.
Lean does not mean the team is less committed to customers. It means every interruption and repeated question competes with other important work.
AI customer support can be relevant for a lean team when it helps keep routine assistance available without requiring continuous human coverage. The team still decides what should remain automated and what deserves direct attention.
19. Customer support automation
Customer support automation is the use of technology to assist with parts of the support process.
It can include answering routine questions, using approved information or taking defined actions. The term covers a wide range of approaches, so it should not be treated as proof of a particular capability.
When evaluating a product, ask specific questions: What information does it use? What can it do? What can it not do? Who controls those boundaries? When does a human become involved?
These questions are more useful than relying on broad claims about automation.
20. Trustworthy AI customer support
Trustworthy AI customer support is support that customers and teams can understand, evaluate and rely on within a defined scope.
Trust is not created by confident wording or by promising that automation can handle everything. It is built through practical choices:
- Use information the business approves.
- Define the capabilities and permissions.
- Keep customer support available where it is useful.
- Make human escalation clear.
- Review the quality of answers and actions.
- Keep people responsible for matters requiring judgement.
This is why a small business should evaluate AI customer support as an operating model, not only as a software feature.
How these terms fit together
The terms in this glossary describe different parts of one support model.
An AI customer support assistant can provide always-on support for a lean team. It can use approved knowledge to answer routine questions and, where the business permits it, act on live data through controlled capabilities.
When a matter requires context or human judgement, the assistant should escalate to a person. The result is not fully automated support. It is a support layer that combines continuous availability with defined responsibility.
That distinction is important. The strongest model is not necessarily the one that automates the most. It is the one that makes the boundaries clear enough for the business to trust and manage.
What to ask before choosing an AI customer support assistant
Before adopting a system, a small business should be able to answer five practical questions:
- What customer questions should the assistant handle? Start with the questions that are suitable for routine assistance.
- What information should it use? Identify the knowledge the business approves and expects the assistant to rely on.
- What capabilities should it have? Define the actions that are useful and appropriate for automated support.
- Where should human judgement begin? Identify the situations
