AI & automation glossary
Every term a decision-maker meets in an AI or automation project, defined in plain language and in the context of real operations. Free to quote with a link to the source.
Foundations
AI agents & models
Agentic AI
Agentic AI is the design approach in which AI systems plan and execute multi-step work autonomously, using tools and feedback, instead of generating a single response. It shifts the unit of automation from an answer to a completed task.
AI agent
An AI agent is a program that pursues a goal on its own: it decides which steps to take, calls tools or APIs to take them, checks the result and repeats until the goal is reached or it hands over. Unlike a chatbot it acts, rather than only answering.
Fine-tuning
Fine-tuning further trains an existing model on company-specific examples so it adopts a style, format or classification behaviour. It changes how a model responds, not what current facts it knows, and is usually the last option after prompting and retrieval.
Large language model (LLM)
A large language model is a statistical model trained on very large text corpora that predicts likely continuations of text. It can classify, summarise, translate and draft, and it is the reasoning component inside most AI agents — but it holds no live company data by itself.
Prompt engineering
Prompt engineering is the practice of specifying, in writing, what a model must do: the task, the rules, the input format and the required output structure. In production systems prompts are versioned artefacts, tested against real cases like any other code.
Retrieval-augmented generation (RAG)
Retrieval-augmented generation is a method in which relevant company documents are searched first and the retrieved passages are given to the language model as context. The model answers from those passages, which makes answers current, source-linked and auditable.
Process automation
Business process automation (BPA)
Business process automation is the systematic execution of a recurring business process by software, from trigger to result, including the data, rules and hand-offs involved. Unlike single-task tools it targets the whole chain: intake, decision, execution and record.
Hyperautomation
Hyperautomation describes the combined use of process analysis, workflow automation, integrations and AI agents to automate as much of an end-to-end process chain as is economically sensible — rather than automating isolated single tasks.
Intelligent document processing (IDP)
Intelligent document processing extracts structured data from invoices, delivery notes, contracts or forms, validates it against master data and posts it into the target system. It combines OCR for reading with models for interpretation and rules for verification.
Process audit
A process audit is a structured review of recurring workflows: which steps exist, which systems and people are involved, how long each step takes, how often it fails, and what a case costs. It produces the evidence base for any automation decision.
Process mining
Process mining reconstructs how a process actually runs by analysing timestamps and event logs from existing systems. It reveals real variants, loops, waiting times and rework — usually a very different picture from the documented process.
Robotic process automation (RPA)
Robotic process automation uses software robots that operate existing applications through their user interface — clicking, typing, copying between screens. It automates rule-based routine work without changing the underlying systems, and breaks when those interfaces change.
Straight-through processing (STP)
Straight-through processing is the share of cases that pass from intake to completion without any manual intervention. It is the central operational metric of an automation programme, because it directly translates into cost and cycle time per case.
Workflow automation
Workflow automation routes a task through its defined steps, owners and conditions automatically: who gets it, what must be checked, what happens on approval or rejection, and what is recorded. It removes coordination work rather than performing the work itself.
Systems & engineering
API integration
An API integration connects two systems through their programming interfaces so data moves automatically and reliably. It is the stable alternative to manual re-entry, CSV exchange and interface robots, and it is the backbone of every serious automation.
Custom software
Custom software is built around a specific company's processes instead of forcing those processes into a standard product. It is chosen when the workflow is a competitive differentiator, when integration depth matters, or when licence and workaround costs exceed the build.
ERP integration
ERP integration connects operational software and AI agents to the system of record for orders, articles, customers, stock and accounting, so automation reads current data and writes results back into the books instead of into a parallel spreadsheet.
Legacy system
A legacy system is business-critical software that is technically outdated but still runs core operations. It typically lacks modern interfaces, which makes it the main constraint — and often the main cost driver — in automation projects.
Model Context Protocol (MCP)
The Model Context Protocol is an open standard for connecting AI assistants to external tools and data sources through a uniform interface, so the same integration can be reused across models and applications instead of being rebuilt for each one.
OCR (optical character recognition)
OCR converts scans, photos and PDFs into machine-readable text. It is the entry step of document automation and delivers characters only — the meaning, validation and posting logic are added by the layers above it.
Orchestration
Orchestration is the coordination layer that decides which agent, model or service handles which step, in what order, with which data, and what happens on failure. It is what turns individual agents into a dependable operational system.
Vector database
A vector database stores text, images or records as numeric embeddings and finds the entries closest in meaning to a query. It is the search layer behind retrieval-augmented generation, enabling semantic lookups where keyword search fails.
Security & governance
AI governance
AI governance is the set of rules defining which AI systems a company operates, on what data, with which permissions, under whose responsibility, and how outcomes are reviewed. It makes AI use auditable towards management, auditors and regulators.
Data residency
Data residency defines in which jurisdiction company data is stored and processed. For AI systems it covers the hosting region of the application, the database, the model endpoint and any logs, plus whether prompts may be used for provider training.
EU AI Act
The EU AI Act is the European regulation that classifies AI systems by risk and attaches obligations accordingly — from transparency duties for ordinary business applications to strict requirements for high-risk uses such as employment decisions or creditworthiness.
Guardrails
Guardrails are the technical limits around an AI system: allowed tools, value ranges, spending and volume caps, forbidden actions and mandatory approvals. They turn an unpredictable model into a system that fails safely and visibly.
Hallucination
A hallucination is a confident but false output from a language model — an invented figure, clause or reference. It is a property of statistical text generation, which is why production systems verify outputs against source data instead of trusting them.
Human in the loop
Human in the loop means a defined point at which a person reviews, approves or corrects the AI system's work. The handover is designed in advance for specific triggers — value thresholds, low confidence, unknown counterparties — rather than left to chance.
Cost & operations
Automation KPI
Automation KPIs are the figures that prove whether an automated process is working: touchless completion rate, cycle time, error and rework rate, cost per case, and volume handled per period. They are defined before the build, not after.
Cost per case
Cost per case is the total cost of processing one unit of work — an invoice, a quote, an enquiry — including labour, software, model usage and error correction. It is the cleanest metric for comparing manual processing with an automated one.
Total cost of ownership (TCO)
Total cost of ownership is the full cost of a system over its lifetime: build or licence, integration, hosting, model usage, support, changes and internal effort. For AI systems the usage-based model cost and the maintenance of integrations are the items most often underestimated.