AI strategy glossary: 30 terms in one sentence each

The short answer

Planning AI work mixes two vocabularies: the technology terms that describe what the systems do, and the governance, risk and investment terms that describe what a business has to decide. The second half is where most confusion sits, because the technology is easy to acquire and the obligations are not obvious. The thirty terms below cover planning, governance, risk, money and measurement, each in one plain sentence. When one appears in a proposal or a board paper, ask what decision it changes; the useful ones always change one, and the rest are decoration.

Planning

Term	In one sentence
AI strategy	A short written statement of where AI will and will not be used in your business, in what order, with measures and owners.
Register of AI uses	The list of every AI use in the business, with its owner, its data and its safeguards, which is the foundation of any governance.
Readiness	Whether a process, its data, its systems and its people can support automation yet, assessed before any building.
Maturity stage	Where a business sits on the path from scattered personal use to a managed, measured capability.
Roadmap	The ordered sequence of AI projects with dates and dependencies, usually twelve months ahead.
Pilot	A small, time-boxed deployment with a defined success measure, which either earns production or is stopped.
Proof of concept	A test of whether something is technically possible, which proves nothing about whether it is worth doing.
Exclusion list	The written record of what you will not do with AI, which is what makes a strategy a strategy.
Willing owner	A named person who wants a given AI use to exist and will look after it, without whom projects fail.
Baseline	The measured state before a project starts, without which no claim of improvement can be checked.

Governance and risk

Term	In one sentence
Human oversight	A person with the information, time and authority to decide differently from a system’s suggestion, and who is accountable.
High-risk system	The EU AI Act’s category for AI in sensitive uses such as employment, credit, education and essential services, bringing specific obligations.
Prohibited practice	Uses the AI Act forbids outright, such as social scoring and certain manipulation and biometric practices.
Transparency obligation	The duty to tell people when they are interacting with AI or receiving AI-generated content.
Data processing agreement	The contract under which a provider processes personal data for you, required before customer data reaches any AI service.
Data protection impact assessment	A structured assessment of privacy risk required before certain processing, including much AI use on personal data.
Risk register	The list of what could go wrong with each AI use, how likely, how bad, and what you are doing about it.
Bias	Systematic difference in outcomes between groups, arising from data, design or deference, detected by outcome testing.
Explainability	The ability to tell an affected person what was decided, on what basis, by whom and how to challenge it.
Shadow AI	Tools staff adopt without approval, which is where most small business data exposure actually happens.
Incident	Any case where an AI system caused harm, exposed data or produced a materially wrong result, which should be logged and reviewed.

Money and measurement

Term	In one sentence
Total cost of ownership	Everything an AI use costs over its life: licences, usage, build, integration, maintenance, review and the people’s time.
Cost per task	What one unit of work costs through the AI system, which is the number to compare with doing it by hand.
Return on automation	The honest comparison of hours and errors saved against total cost, measured rather than estimated.
Usage-based cost	Billing that scales with how much you use, which requires caps, monitoring and forecasting to stay predictable.
Evaluation set	A collection of real examples with known good answers, used to judge a system before and after every change.
Drift	The decline in an AI system’s accuracy over time as data, behaviour or models change, found by monitoring.
Adoption	The share of the people it was built for who actually use it, which decides whether any other measure matters.
Vendor lock-in	The cost of leaving a provider, created mostly by where your data, prompts and workflows live rather than by the model.
Quarterly review	The rhythm of checking each AI use against its measure and retiring what does not earn its place.

Using the glossary

What this means for you

Thirty terms cover the planning, governance, risk and money vocabulary of AI strategy. Use them to read proposals and board papers for the decisions they contain, check that every planned use has an owner, a baseline, a measure, data terms and an error path, and treat any term that names no decision as decoration. This is general information rather than legal advice.