Reference · updated 17 August 2026
The AI terms your board will ask you about.
Plain-English definitions of the 22 terms UK business leaders actually meet, written to be understood the first time. No vendor language, no hedging.
Shadow AI
Shadow AI is the use of personal or unapproved AI accounts for work, outside any company oversight. It typically looks like staff pasting client data into a personal ChatGPT or Claude login because no sanctioned tool exists. The risk is not the technology but the absence of a record: no audit trail, no data boundary, and no way to answer a regulator asking where the information went.
More on shadow ai →AI governance
AI governance is the set of decisions, boundaries and records that determine how an organisation is allowed to use AI. In practice it answers four questions: what AI may be used for, what it must never be used for, which data may cross into it, and who is accountable when it is wrong. Good governance speeds adoption up rather than slowing it down, because people stop guessing where the line is.
More on ai governance →AI Manifesto
An AI Manifesto is a one-page, signable statement of how an organisation works with AI, written for humans rather than auditors. It has three parts: the Shield, meaning what you will never do; the Heart, meaning the principles that guide grey-area decisions; and the Engine, meaning the pre-approved uses your people can get on with today.
More on ai manifesto →ISO 42001
ISO/IEC 42001 is the international management-system standard for artificial intelligence. It specifies how an organisation should plan, operate, monitor and improve its use of AI, in the same structural way ISO 27001 handles information security. It is a management system rather than a technical test: it certifies how you govern AI, not whether a particular model is safe.
More on iso 42001 →EU AI Act
The EU AI Act is the European Union's regulation governing artificial intelligence, which classifies AI uses by risk and attaches obligations to each tier. It reaches UK businesses that place AI systems on the EU market or whose AI output is used in the EU, so a UK-only registered office does not by itself put an organisation outside its scope.
More on eu ai act →Agentic AI
Agentic AI describes systems that pursue a goal across multiple steps, choosing their own actions and using tools along the way, rather than answering a single prompt. The governance question changes with it: instead of reviewing one output, you are deciding in advance which actions an agent may take unsupervised and where a human must approve.
More on agentic ai →AI Operating System
An AI Operating System is AI configured around how a specific business actually works, rather than a general assistant that every employee has to re-explain their job to. It holds the organisation's context, tone, processes, knowledge base and governance in one consistent environment, so the quality of output does not depend on who is typing.
More on ai operating system →AI readiness assessment
An AI readiness assessment is a structured diagnostic of whether an organisation can absorb AI before it spends on it. It examines strategy, leadership, culture, governance, capability and delivery rather than technology alone, on the basis that the constraint on most AI programmes is organisational rather than technical.
More on ai readiness assessment →Human-in-the-loop
Human-in-the-loop means a person reviews and approves an AI output before it takes effect. It is the practical control behind most AI governance commitments, and it only works when it is specific: naming which decisions require review, by whom, and what evidence is kept of the review having happened.
More on human-in-the-loop →Responsible AI
Responsible AI is the operational practice of using AI in ways an organisation can defend afterwards. It differs from an ethics statement in that it is measurable: defined permitted uses, named accountable owners, recorded human oversight, and evidence retained for the decisions AI participated in.
More on responsible ai →AI adoption
AI adoption is the point at which people use AI in their real work by default, rather than the point at which a licence is purchased. Most programmes stall in the gap between the two, because access was solved and capability, permission and habit were not.
More on ai adoption →AI transformation
AI transformation is the redesign of how work is done around AI capability, as opposed to adding AI tools to existing processes. The distinguishing test is whether any process was reimagined: if the workflow is unchanged and merely faster in places, that is automation rather than transformation.
More on ai transformation →Pilot purgatory
Pilot purgatory is the state of an AI programme that repeatedly proves value in small trials but never reaches production. It is usually caused by the absence of an owner, a governance route to approval, or a plan for the change in work that scaling would require.
More on pilot purgatory →Claude Partner Network
The Claude Partner Network is Anthropic's partner programme for organisations that build with and deliver Claude. Membership gives early visibility of capability development, support access, and a working relationship with Anthropic that informs roadmap advice to clients.
More on claude partner network →Fractional Chief AI Officer
A fractional Chief AI Officer is an experienced AI executive engaged part-time to own an organisation's AI strategy, governance and delivery. It suits mid-sized organisations that need senior AI accountability without a full-time appointment at that level.
More on fractional chief ai officer →Prompt engineering
Prompt engineering is the practice of structuring instructions, context and examples so an AI model produces reliable output. In business use its value is mostly in repeatability: turning a result that one person can get once into one the whole team can get every time.
More on prompt engineering →Retrieval-augmented generation
Retrieval-augmented generation, or RAG, is the technique of supplying an AI model with relevant documents at the moment of asking, so its answer is grounded in that material rather than in memory alone. It is how an assistant is made to answer from your policies and files instead of generalities.
More on retrieval-augmented generation →Data boundary
A data boundary is the explicit rule stating which categories of information may and may not be placed into an AI system. It is the single most useful governance artefact for staff, because it converts an abstract instruction to be careful into a list they can actually check against.
More on data boundary →Zero data retention
Zero data retention is a contractual arrangement in which a provider does not store the content sent to its models beyond the time needed to return a response. It is frequently the control that makes a regulated use case viable, and it needs confirming in the agreement rather than assumed from marketing material.
More on zero data retention →SPARK Assessment
SPARK is Ignite AI Solutions' structured AI discovery audit for UK SMEs. It combines an 18-dimension readiness diagnostic, human capability mapping and an adversarial advisory board pressure-test, and produces a findings report the client keeps regardless of what they do next.
More on spark assessment →C-H-A-N-G-E framework
C-H-A-N-G-E is Ignite AI Solutions' six-element cultural operating system for AI programmes: Communication, Human oversight, Attitude, Network, Governance and Enablement. It runs continuously alongside delivery rather than as a workstream that finishes.
More on c-h-a-n-g-e framework →The 3-5-4 Method
The 3-5-4 Method is a 45-minute process-mapping technique for finding a first AI use case: three signs of friction, five questions that map the workflow, and four decisions about what to automate. It is designed to produce a scoped candidate rather than a list of ideas.
More on the 3-5-4 method →Missing a term you have been asked about? Tell us and we will add it. If you would rather have these explained to your team in the context of your own work, that is what the training day is for.