glossary
the key ai terms for smbs.
What is a Company-GPT? What does RAG mean? How does the EU AI Act work? Eight central concepts, briefly explained — no marketing buzz, with pointers to the mid-market use case.
- company-gpt
A Company-GPT is a company-hosted, governance-compliant instance of a Large Language Model (LLM) with access to internal sources — documents, knowledge bases, CRM data.
Unlike private ChatGPT accounts on each desk, a Company-GPT is centrally managed, GDPR-aligned, with permissions, roles and audit logging. Processing happens in a defined tenant; there is no data leakage into personal accounts.
- voice agent
A Voice Agent is an AI system that conducts natural-language phone or chat conversations — taking inbound calls, making outbound calls, understanding intent and triggering actions in downstream systems (CRM, calendar, ticketing).
Unlike classic IVRs ("press 1 for …"), a modern Voice Agent understands free-form speech and reacts contextually. Current stacks (e.g. ElevenLabs for voice, GPT/Claude for reasoning) achieve sub-second latency — most callers cannot tell the difference from a human counterpart.
- ai solutions
AI Solutions are tailor-made AI applications for a concrete business problem — website, web app, internal tool, AI feature inside an existing product — as opposed to off-the-shelf SaaS tools.
The benefit: the solution fits your stack, your data and your processes. With an AI-assisted development toolchain, such custom applications now ship in weeks instead of quarters, at a fraction of classic agency cost.
- ai opportunity analysis
An AI opportunity analysis is a focused diagnosis identifying which AI use cases have the highest leverage in a given company — evaluated along data, adoption realism and compliance.
Typical formats: Quick Scan (1–2 weeks, broad evaluation) or Deep Assessment (3–4 weeks, in-depth evaluation of selected use cases including ROI estimates, stack recommendation and roadmap). Output is a compact concept document, not an 80-page PDF.
- mvp (minimum viable product)
In an AI context, an MVP is a runnable, focused first version of an AI solution — just broad enough to be tested in production and to make the core use case measurably work.
Typical AI-MVP timeline: 2–4 weeks from kickoff. Prerequisites are a clearly scoped use case, access to relevant data and an available counterpart on the customer side. An MVP is not a throwaway prototype — it runs on a production stack from day one and can be extended or replaced without restart.
- rag (retrieval-augmented generation)
RAG is a technique where a Large Language Model retrieves information from an external knowledge source (documents, database, website) and embeds it into its context before generating an answer.
The benefit: the LLM answers questions based on current, company-specific data — without needing to be retrained (fine-tuned). Per-answer source citations are possible, which reduces hallucinations and enables auditability. Standard architecture for Company-GPTs.
- eu ai act
The EU AI Act (Regulation (EU) 2024/1689) is the European regulation on artificial intelligence. It classifies AI systems by risk (prohibited, high, limited, minimal) and assigns graded obligations to providers and deployers.
Particularly relevant for SMBs: many everyday AI applications fall under "limited risk" (transparency obligations, Art. 50 — disclosing AI interaction) or "minimal risk" (no specific obligations). High-risk systems (HR scoring, credit scoring, critical infrastructure) require documented risk analysis, quality management and audits.
- gdpr + ai
When deploying AI systems under GDPR, organisations must demonstrate a legal basis for processing personal data (typically Art. 6 (1) (b) or (f)), conclude a data-processing agreement (Art. 28) with the LLM provider, and secure third-country transfers (US LLMs).
Current state (2026): the EU-U.S. Data Privacy Framework (DPF) is active for many US LLM providers (Microsoft, OpenAI, Anthropic, Google). Where highest caution is required, EU-only providers (Mistral, Aleph Alpha, Langdock) or self-hosted open-source models are an option. Transparency obligations toward staff and customers apply regardless of provider.
term missing? or just — ready to build?
30 minutes, no pitch deck. We help sort out what's relevant for your situation and tell you honestly whether — and how — it's worth doing.