Glossary for business decision makers

The standard way two systems exchange data. A good API lets your team (or another vendor) plug in a new solution without involving the original vendor.

A chronological record of who changed what, and when, inside a system. Essential in finance, insurance, healthcare and public-sector solutions.

Processing in groups (e.g., once a day) instead of real time. Cheaper and easier to audit; fits finance and month-end close processes.

The structure of accounting accounts in the ERP. Any change (new GL code, tax category) ripples through every integration that posts financial data.

Adherence to legal, regulatory and security requirements (GDPR, ISO 27001, NIS2, industry rules). In AI projects it usually decides where data may be stored.

The system that stores customer information, sales opportunities and communication history (e.g., Salesforce, HubSpot, Microsoft Dynamics 365, Pipedrive).

Enterprise software that consolidates finance, inventory, procurement, manufacturing and HR in one place (e.g., SAP S/4HANA, Oracle NetSuite, Microsoft Dynamics, Odoo).

EU regulation on personal data protection. Governs how personal data may be collected, stored, transferred and used — including inside AI solutions.

A process where a human approves or corrects the AI's decision before it is committed to a system. Used for uncertain cases (e.g., when AI confidence is <80%).

An AI solution that automatically extracts structured fields from documents (invoices, contracts, waybills, forms) and feeds them into business systems. Replaces manual data entry.

A large language model (e.g., GPT, Claude, Gemini) trained on massive text corpora. Used for document understanding, generation, classification and AI agents.

A 2025 standard that lets an AI agent securely connect to business systems (CRM, ERP, document archives) and take rule-bound actions on its own.

An intermediate software layer that connects multiple systems and keeps integration logic in one place (MuleSoft, Workato, Azure Logic Apps, Boomi, n8n). Prevents the "spaghetti" code that appears once a document must reach several systems.

Optical Character Recognition turns images or PDFs into machine-readable text. It is only the first step of IDP: OCR recognises characters, while IDP understands the document's meaning.

The component that drives the flow: where and when to route data, what to do on error, how to retry a failed step.

A Salesforce mechanism that lets other systems react in real time to changes inside the CRM (e.g., when a user corrects a field extracted by AI).

A limited initial project that proves an AI solution works with your real data. Typical duration: 4–8 weeks.

A component that buffers requests and releases them at a controlled pace so the receiving system's rate limit or load is not exceeded.

An architecture where the AI retrieves relevant documents from your internal knowledge base before generating an answer. Lets the AI rely on real company data instead of hallucinating.

A cap a system sets on how many requests per minute/hour it accepts. Exceed it and requests get rejected — that is why a queue is needed to smooth the stream.

The most widely used API style on the web. Works with simple HTTP requests. In pull mode your system asks for data on its own schedule.

Return on investment — how many € come back per € invested. In AI projects measured in hours saved, errors reduced and faster closing cycles.

Software bots that follow rules and mimic clicks on a screen (e.g., UiPath, Power Automate). Often paired with IDP: IDP reads the document, RPA enters the result into systems that lack an API.

SAP's official integration layer. Without it, third-party systems cannot safely write into SAP S/4HANA — which is why SAP integrations are always pricier than other ERPs.

A service-level agreement fixing incident response times, resolution deadlines and system availability. Without an SLA a vendor has no formal commitment.

A situation where company data, processes and integrations are so tied to one vendor's platform that switching becomes technically or commercially prohibitive.

An automatic notification one system pushes to another the moment an event happens (e.g., an invoice has been processed). No waiting, no polling.

A cryptographic signature added to a webhook by the sending system. Lets the receiver verify the message came from the expected source and was not modified in transit.

Warehouse management software handling receiving, locations, picking and dispatch. Typically integrated with the ERP and e-commerce platform.