Knowledge management glossary for contact centres and service desks

Terms A to C

Agent assist

Software that suggests articles, answers or next steps to an agent during a live conversation. It can only suggest what the knowledge base contains, so its quality depends on the knowledge behind it.

Article lifecycle

The stages an article passes through: draft, review, published, updated and finally archived or retired. A healthy lifecycle includes review dates, so articles don't stay published long after they stop being true.

Average handle time (AHT)

The average time an agent spends on a contact, including talk or chat time, hold and after-contact work. Knowledge gaps raise it through searching, holds and asking colleagues, which is most of the cost of a knowledge gap.

Baseline

The earlier period a metric is compared against to decide whether it changed, such as the same weekday last week. A poor baseline produces false alerts.

Brand alignment

How closely an article follows the organisation's voice, terminology and style rules. One of the standard content quality checks.

Calibration

How well a predicted probability matches reality. A well-calibrated model is right about 70% of the time when it says 70%. A score can rank items sensibly without being calibrated.

Complexity

How hard an article is to follow, judged by steps, conditions, branches and jargon. Complex articles are slow to use on a live call and easy to apply wrongly.

Confidence score

A number showing how strongly the evidence supports a classification, match or recommendation. Useful for deciding what a person should review first.

Content health

The overall quality of a knowledge base across checks such as readability, validity, brand alignment, complexity, duplicates and contradictions. See our guide to knowledge base content health.

Contradiction

Two or more articles that give different answers to the same question. Customers then get a different answer depending on which article the agent opened. See duplicate and contradicting articles.

Coverage rate

The share of customer intents, taken from conversations, that a knowledge article answers. Tracked by topic and over time, it shows how much real demand your knowledge base can meet.

Terms D to G

Deflection

A contact that self-service resolved before it reached an agent. It is estimated rather than observed, because the contact that didn't happen leaves no record.

Disposition (wrap-up code)

The reason code an agent selects at the end of a contact. Quick to report on, but often chosen in a hurry and too coarse to show what the customer actually needed.

Drift

Gradual change in what customers ask, how they phrase it or how conversations end, too slow to trigger an alert in any single period.

Duplicate article

An article that substantially repeats another. Duplicates split maintenance effort and drift apart over time until they contradict each other.

Embedding

A numeric representation of text that places similar meanings close together. Embeddings power semantic search and the matching of customer questions to articles.

Emerging topic

A topic that is new this period or growing quickly from a small base. Often the earliest sign of a knowledge gap. See emerging topics and drift detection.

Findability

How easily the right article can be found by the people who need it, through search or browsing. Poor findability makes a correct article useless.

First contact resolution (FCR)

The share of issues resolved in the customer's first contact, with no need to come back. Missing or wrong knowledge is one of its most common causes of failure.

Gap probability

A score between 0 and 1 estimating how likely it is that no article answers a given customer question. Useful for ranking likely gaps; only as reliable as its calibration.

Grounding

Basing an AI answer on supplied sources, such as approved articles, instead of the model's general training. Grounded answers can cite where each statement came from.

Terms H to K

Hallucination

A fluent, confident AI statement that is not supported by its sources or is simply false. No supplier can rule it out entirely, which is why sources and checks matter. See can AI answers be hallucination-free?

Holiday-aware baseline

A baseline that recognises public holidays and the days around them, so the calendar's effect on volume isn't reported as a spike or drop.

Intent

A single thing a customer wanted from a contact, written in plain language, such as "find out why the direct debit failed". One conversation can contain several.

KCS (Knowledge-Centered Service)

A methodology, developed by the Consortium for Service Innovation, in which knowledge is captured, structured, reused and improved as part of solving each issue, instead of written separately afterwards. See closing the KCS loop in a contact centre.

Knowledge area

A separate scope within a knowledge operation, usually one team or line of business, with its own taxonomy, articles and reporting.

Knowledge base

The collection of articles, procedures and reference material agents and customers use to answer questions, and the system that stores it.

Knowledge gap

Any point where a customer needs an answer and the knowledge base can't supply a correct one quickly. The four kinds are missing, stale, inconsistent and unfindable. See knowledge gap analysis.

Terms L to R

Large language model (LLM)

An AI model trained on large amounts of text that can read, summarise and write language. Used to extract intents, draft articles and answer questions.

Provenance

The record of where a piece of content or a number came from: which conversations, articles or data produced it. Provenance lets a reviewer check an AI draft before approving it.

Retrieval-augmented generation (RAG)

A technique where relevant passages are retrieved from a knowledge source and given to a language model with the question, so the answer draws on that material.

Readability

How easy an article's text is to read and understand, judged by sentence length, word choice and structure. Agents reading during a live call need plain language.

Repeat contact

A customer coming back about the same issue, measured either as a return within a set window or as conversations assessed as needing follow-up. A key sign that an answer didn't work, and one of the measures in measuring knowledge article impact.

Resolution rate

The share of contacts in which the customer's issue was resolved. Read alongside repeat contact, because a contact marked resolved can still produce a callback.

Terms S to Z

Self-service

Customers finding answers themselves through help centres, chatbots or portals. It can only answer what the knowledge base contains.

Semantic search

Search that matches meaning instead of exact keywords, usually using embeddings. It improves findability but cannot show what is missing from the content.

Sentiment

A measure of how positive or negative a customer was during a conversation. Falling sentiment on one topic often points to a missing or unwelcome answer.

Stale content

An article that was once correct but now describes an old price, form, process or policy. It is found and followed, which makes it more damaging than a missing article.

Taxonomy

The structured set of categories that conversations and articles are both filed against. See building a knowledge taxonomy from conversations.

Topic spike

A sudden rise in one topic's volume well above its baseline, often caused by an outage, a campaign or a change customers didn't expect.

Transfer rate

The share of contacts passed to another agent or team. Frequent transfers on a topic often mean front-line agents lack the knowledge to answer it.

Validity

Whether an article is still true: its facts, links, dates and steps match current products and policies. Checked against authoritative sources.

Zero-result search

A knowledge base search that returns nothing. Good for spotting findability problems, but it only shows gaps someone searched for. See failed searches vs knowledge gaps.