Skip to main content
    Knowledge Centre

    AI & Automation Glossary

    32 terms, each defined in one self-contained paragraph followed by the practical implication for a business deploying it. Definitions are linked automatically wherever the term appears across the site.

    A

    • Agentic workflow

      An agentic workflow is a business process in which one or more AI agents handle the decision points and system actions, with humans supervising exceptions rather than performing each step.

    • AI agent

      An AI agent is software that receives a goal, plans the steps needed to reach it, executes those steps using external tools or APIs, and reports the outcome. What separates an agent from a chatbot is tool use and multi-step execution, not conversation.

    • AI governance

      AI governance is the set of policies, approvals, records and controls that determine which AI systems a business may deploy, on what data, with what oversight, and who is accountable for outcomes.

    B

    • Build vs buy

      Build vs buy is the decision between developing an AI capability in-house and licensing an existing product. The deciding factors are whether the capability is a competitive differentiator, how specific your data is, and whether you can staff maintenance.

    C

    • Chatbot

      A chatbot is a conversational interface that answers questions in natural language. Modern business chatbots are usually retrieval-grounded, drawing answers from the company's own documentation rather than generating freely.

    • Context window

      The context window is the maximum amount of text — measured in tokens — a model can consider in a single request, covering the instructions, supplied documents, conversation history and the response.

    • CRM automation

      CRM automation removes manual data entry and follow-up from the sales process: enriching records, logging activity, routing leads, drafting follow-ups and keeping pipeline data accurate without rep effort.

    D

    • Deflection rate

      Deflection rate is the share of support contacts fully resolved by automation without a human agent. It is the primary efficiency metric for customer service AI and should always be read alongside CSAT.

    E

    • Embedding

      An embedding is a list of numbers representing the meaning of a piece of text, produced by a model so that semantically similar texts sit close together in that numeric space.

    • Evaluation

      Evaluation is the systematic scoring of AI output against a fixed set of test cases with known good answers. It is what turns 'the demo felt good' into a measurable quality baseline you can defend changes against.

    F

    • Fine-tuning

      Fine-tuning further trains an existing model on your own examples so it adopts a specific format, tone or classification behaviour. It changes how a model responds, not what facts it knows.

    G

    • Generative engine optimisation

      Generative engine optimisation is the practice of structuring content so AI answer engines — ChatGPT, Gemini, Perplexity, Google AI Mode — can extract, trust and cite it. It favours direct answers, clear structure, verifiable sources and machine-readable schema.

    • Guardrails

      Guardrails are the technical constraints that stop an AI system from acting outside its permitted scope — allowed tools, spending limits, output filters, escalation triggers and refusal rules.

    H

    • Hallucination

      A hallucination is a confident, fluent model output that is factually wrong or fabricated. It is a structural property of next-token prediction, not a bug that a better prompt fully removes.

    • Human in the loop

      Human in the loop means a person reviews or approves AI output before it takes effect. It is the standard control for automated decisions with financial, legal, safety or reputational consequences.

    I

    • Inference

      Inference is the act of running a trained model to produce an output. In business terms it is the recurring per-request cost of an AI feature, as distinct from the one-off cost of building it.

    L

    • Large language model

      A large language model is a neural network trained on very large text corpora to predict the next token in a sequence. That single capability, at scale, produces summarisation, translation, classification, code generation and reasoning-like behaviour.

    • Lead scoring

      Lead scoring ranks inbound prospects by likelihood to buy, using firmographic fit and behavioural signals. AI improves it by reading unstructured signals — email replies, call notes, site behaviour — that rule-based scores miss.

    O

    • Orchestration

      Orchestration is the coordination layer that decides which model, tool or sub-process handles each step of a task, manages retries and failures, and keeps state across the sequence.

    P

    • Process mapping

      Process mapping documents each step, decision, system and handoff in a business process before automating it. It is the step most AI projects skip and the one that most often determines whether they succeed.

    • Prompt engineering

      Prompt engineering is the practice of structuring the instructions, context and examples given to a language model so it produces reliable, correctly formatted output for a specific task.

    • Prompt injection

      Prompt injection is an attack in which malicious instructions hidden in content the model reads — a web page, an email, a document — cause it to ignore its original task or misuse its tools.

    • Proof of concept

      A proof of concept is a time-boxed build that tests whether an AI approach can meet a defined success threshold on real data, before committing to production investment.

    R

    • Retrieval-augmented generation

      Retrieval-augmented generation is a pattern where relevant documents are fetched from a knowledge base at query time and inserted into the model's prompt, so answers are grounded in your own content rather than the model's training data.

    • Return on investment

      Return on investment for AI automation is the net benefit — labour hours recovered, revenue gained, error cost avoided — divided by total cost of ownership, including build, licences, maintenance and change management.

    S

    • Structured data

      Structured data is machine-readable markup — usually JSON-LD following schema.org — that states explicitly what a page is about: article, product, FAQ, organisation, author. Search engines and AI models use it to interpret and attribute content.

    T

    • Token

      A token is the unit of text a language model processes — roughly three-quarters of an English word. Model pricing, context limits and latency are all measured in tokens.

    • Topic cluster

      A topic cluster is a pillar page covering a broad subject plus a set of child pages covering its sub-questions, all linked to each other. It is the standard architecture for building topical authority.

    • Topical authority

      Topical authority is the degree to which a site is treated as a credible source across an entire subject area rather than for individual keywords. It comes from covering a topic comprehensively, coherently interlinked, with consistent expertise signals.

    • Total cost of ownership

      Total cost of ownership is the full lifetime cost of an AI system: implementation, model and infrastructure spend, integration upkeep, monitoring, retraining, and the internal time spent supervising it.

    V

    • Vector database

      A vector database stores text (or images) as numeric embeddings and retrieves items by semantic similarity rather than keyword match. It is the retrieval layer behind most production RAG systems.

    W

    • Workflow automation

      Workflow automation is the use of software to move work between systems and people without manual re-entry — triggers, conditions and actions connecting the tools a business already runs.