Company

What do the main AI and automation terms mean?

Most AI and automation terms describe one of three things: the technology, such as a large language model or retrieval; the way work moves through a firm, such as a workflow, handoff or process map; or the rules around it, such as UK GDPR and a DPIA. This glossary defines each one in plain English.

Conversations about AI in a professional services firm quickly fill with vocabulary from three worlds: software, practice management and data protection. The terms below are the ones a partner or operations director is most likely to meet when a supplier, a colleague or a regulator raises the subject. Each definition is short and plain, and where Aldbry has written about a term in more depth, the definition links to it.

A to C

Adoption: The point at which a team uses a new process or tool as the normal way of doing a job, without being reminded. A licence being bought or a training session being held is not adoption. See getting a team to adopt AI.
Agentic AI: AI that carries out a sequence of steps towards a goal, such as searching files, drafting a document and filing it, rather than answering a single question. It acts within limits a person sets, and in client work its output should still be reviewed before anything leaves the firm. See custom AI systems.
AI policy: A written statement of how a firm's people may and may not use AI tools: which tools are approved, what information can go into them, and who checks the output. See writing an AI policy.
AI readiness: How prepared a firm is to put AI into real work, judged on things like how well its processes are documented, how consistent its templates are and whether its data is organised and governed. See the AI readiness checklist.
Audit trail: A record of who did what, when and on what basis, kept as the work happens so it can be checked later by a reviewer, a client or a regulator. See audit trails.
Automation: Having a system carry out steps of a process by rule, without a person doing each one by hand. It is older than AI and often works best alongside it. See process automation consultancy.
Capacity: The hours a team has available for work in a given period. Freeing capacity means the same people can take on more chargeable work without the firm hiring. See the capacity calculator.
Charge-out rate: The hourly rate at which a fee earner's time is billed to clients. It is the usual basis for putting a value on hours lost to unbillable work, and it can be tested in the unbillable hours calculator.
Compliance evidence: The documents and records a firm keeps to show it has met a regulatory, contractual or quality obligation, such as checks completed, approvals given and training done. See compliance evidence.
Copilot: Microsoft's name for its AI assistants, most often meaning Microsoft 365 Copilot, which works inside Word, Outlook, Teams and Excel using the files and emails a user can already access. See what Copilot is actually good for.

D to F

Data controller: Under UK GDPR, the organisation that decides why and how personal data is processed, and which carries primary responsibility for it. A firm is usually the controller of the personal data it holds on its clients and staff. See data protection and AI on client work.
Data processor: An organisation that processes personal data on a controller's behalf and on its instructions, such as a software supplier hosting a firm's files. UK GDPR requires a written contract between controller and processor setting out what the processor may do.
Day rates: A charging model in which a supplier bills for each day worked, so the final cost depends on how long the work takes rather than on what is delivered. Aldbry does not charge day rates. See fixed fee versus day rates.
Document automation: Producing documents such as letters, reports and engagement terms from templates that fill themselves with the right details, instead of copying and editing a previous version. See document automation.
DPIA: A data protection impact assessment: a documented review of the risks a use of personal data poses to individuals and how those risks will be reduced. UK GDPR requires one where processing is likely to result in a high risk to individuals.
Fee earner: Anyone in a firm whose time is billed to clients, such as a solicitor, accountant, surveyor or consultant. Time they spend on work that cannot be billed is the main cost automation targets. See the partner bottleneck.
Fixed fee: A price agreed in writing before work starts that does not change if the work takes longer than planned. Aldbry works only on a fixed fee, set by the process being rebuilt. See AI consultancy pricing.
Full-time equivalent: A way of counting staff as if all worked full time, so two people on half hours count as one. It is useful for expressing hours saved as a share of a person's working week. See automation versus hiring.

G to K

Generative AI: AI that produces new content, such as text, summaries or images, in response to an instruction. Most AI tools now sold to firms, including chat assistants, are generative AI built on large language models.
Go-live: The day a new process stops being tested and becomes how the job is actually done, on live work. In an Aldbry rebuild it falls within 30 days of kickoff. See how we work.
Hallucination: When an AI model states something that sounds plausible but is false, such as an invented figure, citation or clause. It is the main reason AI output on client work needs a named reviewer. See the risk in copy and edit.
Handoff: The point in a process where work passes from one person or system to another. Handoffs are where delays, rekeying and lost detail tend to gather. See the work done twice.
Human in the loop: A design in which a person checks and approves an AI system's output at set points before it is acted on or sent. It keeps professional judgement, and accountability, with the firm. See reviewing AI output on client work.
Knowledge base: An organised collection of a firm's documents, precedents, policies and past work that people or an AI system can search for answers. See knowledge retrieval.

L to O

Large language model: The type of AI model behind tools such as ChatGPT and Copilot. It is trained on very large amounts of text to predict and produce language, which lets it draft, summarise and answer questions, but it has no built-in knowledge of your firm or your clients.
Lawful basis: The legal ground an organisation relies on to process personal data under UK GDPR. There are six: consent, contract, legal obligation, vital interests, public task and legitimate interests, and one must apply before processing begins.
Merge field: A placeholder in a template, such as a client name or matter reference, that is filled automatically from a record when the document is produced. See preparing templates for automation.

P to R

Pilot: A limited, real use of a new process by a small group on live work, to prove it holds up before it is rolled out. Pilots that never move past this stage are a common way AI projects stall. See why AI rollouts fail.
Process map: A step-by-step picture of how a job is really done: every input, template, handoff, check and decision, and who owns each. See how to document a process.
Prompt: The instruction or question given to an AI model. Its wording, and the context supplied with it, largely decide the quality of what comes back.
Prompt library: A shared, maintained set of tested prompts for a firm's recurring tasks, so people start from wording that works rather than writing their own each time. See AI training for teams.
Proof of concept: A small test built to show that an idea can work technically, usually on sample material rather than live work. It answers whether something is possible, not whether a team will use it every day. See training versus implementation.
Realisation: The proportion of the value of recorded time that a firm actually bills, or collects, once write-offs and discounts are taken off. It sits alongside utilisation as a measure of how well time turns into fees. See measuring return on AI.
Retrieval (RAG): Retrieval-augmented generation: an AI system first finds the relevant passages in a firm's own documents, then writes its answer from them. It grounds answers in your material and makes it possible to show the source behind each one. See knowledge retrieval.

S to Z

Scope creep: The gradual growth of a piece of work beyond what was agreed, usually without the timetable or fee being revisited. See scope documents.
Shadow AI: Staff using AI tools the firm has not approved, often personal accounts, for work tasks. The risk is client information leaving the firm's control without anyone knowing. See shadow AI.
Special category data: Personal data that UK GDPR treats as more sensitive, such as health, racial or ethnic origin, religious beliefs, trade union membership, genetic and biometric data, and sex life or sexual orientation. Processing it needs a lawful basis and an additional condition.
Standard operating procedure: A written, step-by-step description of how a recurring task should be done, so it is done the same way whoever does it. See standard operating procedures.
Template: A standard document with fixed wording and set places for details that change, used as the starting point for letters, reports and proposals. Templates drift out of date unless someone owns them. See why templates rot.
UK GDPR: The UK's general data protection law, kept in UK law after Brexit and read alongside the Data Protection Act 2018. It is enforced by the ICO and governs any use of personal data, including in AI tools.
Unbillable time: Hours fee earners spend on work that cannot be charged to a client, such as write-ups, reporting, chasing and admin. See how to calculate unbillable hours.
Utilisation: The share of a fee earner's available hours spent on chargeable work. Cutting unbillable time is one of the most direct ways to raise it without longer days.
Workflow: The sequence of steps, people and systems a piece of work passes through from start to finish. See workflow automation.

Questions people ask

What is the difference between AI and automation?

Automation is any process that runs by rule without a person doing each step, such as filling a template from a form. AI handles the parts that need language or judgement, such as drafting commentary or reading an unfamiliar document. Most useful systems in a firm combine the two, with a person reviewing the result.

Is Copilot the same as ChatGPT?

No. They are separate products from different companies, although both are assistants built on large language models. Microsoft 365 Copilot works inside Word, Outlook, Teams and Excel and can draw on the files and emails a user already has access to. Neither one, on its own, changes how a process in the firm runs.

Do we need a DPIA before using AI on client work?

Not in every case. UK GDPR requires one where processing is likely to result in a high risk to individuals, and client files that hold personal data, particularly special category data, can meet that bar. The ICO publishes guidance on when one is needed, and many firms complete one as a matter of course before a new AI use goes live.

Which of these terms matters most when choosing a first process?

Unbillable time. The best first process is usually the one where fee earners lose the most hours each week to work nobody can charge for. The audit estimates those hours, prices them at your own charge-out rate and names the process to tackle first.

Which job is costing your firm the most?

Ten questions, about three minutes. You see the hours and the cost before you speak to anyone.

Find out which job is costing you most