Forty words, one sentence each.
The vocabulary you will meet in proposals, articles and sales calls, explained for the owner of a Monaco company, with what it means for you rather than for an engineer.
- AI agent
- Software that reads what arrives, does the repetitive part according to rules you set, and hands anything uncertain to a person. Unlike a chatbot, it acts: it files, drafts, books, matches, reports.
- Assistant
- The word we use for an agent that works alongside a person: it drafts, the person decides. Most of what we build is an assistant.
- Automation
- A task that runs on its own once triggered: an email arrives, lines are typed into the tool, a confirmation goes out. AI automation is automation that can read and write language, so it handles messages and documents, not only forms.
- Chatbot
- A program that answers questions in a chat window, usually from your own content. Useful for routine questions; it does not act on your tools unless it is also an agent.
- Large language model (LLM)
- The engine behind ChatGPT, Claude, Gemini and Copilot: a model trained on text that can read, write, summarise and translate. We use several, and swap them; the rules around them matter more than the brand.
- Prompt
- The instruction given to a model. In a built system the prompt is written once, tested, and versioned like code; nobody types it by hand each time.
- Hallucination
- A confident answer that is wrong: an invented price, a made-up reference. The reason every system we build cites its sources and stops when unsure, and why a person approves what leaves.
- Human in the loop
- The design where a person approves, corrects or rejects what the system proposes before it takes effect. Our default for anything that leaves the company.
- Guardrails
- The written limits a system works within: which prices it may quote, which messages it may send, what it must never do. Written into the software, not into a guideline.
- RAG (retrieval-augmented generation)
- The technique that lets an assistant answer from your own documents: it finds the relevant passage first, then writes from it, and can cite it. How an assistant knows your prices without being trained on them.
- Fine-tuning
- Training a model further on your own examples. Rarely needed for a small company; RAG and good rules do the job with none of the cost or the data risk.
- Training (on your data)
- When a provider uses what you send to improve its model. Public chatbots may; the contracts we use exclude it. Ask any provider in writing.
- Token
- The unit models are billed in, roughly three quarters of a word. Why model costs for a company of nine people are small, and why we show them separately in the monthly count.
- API
- The door through which software talks to other software. Outlook, HubSpot, Pennylane and WhatsApp Business all have one; an integration is built on it.
- Integration
- Connecting the assistant to the tools where the work lives, with its own identity and the minimum permissions. Most of the work in any project.
- Workflow
- The sequence of steps a process follows, with the decisions in between. Writing it down is the first thing an audit does; automating it is the second.
- OCR / document reading
- Turning a scanned or PDF document into text and fields a system can use: an invoice into lines, an ID into a name and a date. Checked against another source before anything is booked.
- Structured data
- Information in fields rather than in prose: a date in a date field, an amount in an amount field. Also the markup on a web page that tells search engines what the page is about.
- Log
- The record of everything the system did, with when, from which document and with what result. Yours, readable, the basis of the monthly count.
- Prototype
- Two weeks, fixed price, on a sample of your real data: the automation does the job on your cases so you decide on evidence. Our second step, after the free quick-check.
- Quick-check
- A free 30-minute call, then a written note with the two or three tasks to automate first, what each would take and a fixed price for the prototype. Where every project starts.
- AI audit
- The written picture of where a company’s hours go and which tasks to automate in what order. Ours asks every employee a seven-minute questionnaire and adds up the answers.
- Run (operations)
- The monthly step after the build: monitoring, fixing, counting the hours given back, and lining up the next task. A monthly amount you can stop.
- SaaS
- Software you rent monthly in the cloud. Fine for standard needs; for a process that is yours, a small tool you own is often cheaper over three years.
- Custom software
- A tool built for your process, owned by you, documented, maintained. What we build when no existing tool does the job.
- Cloud / EU hosting
- Where the software and data run. We host in the European Union by default and say where in the processing description; some clients run inside their own environment.
- Tenant
- Your own compartment in Microsoft 365 or Google Workspace. An assistant works inside it with a dedicated identity; nothing is copied out.
- Law 1.565
- Monaco’s data-protection law, aligned with European standards, supervised by the APDP. Applies to any AI system that processes personal data, which is most of them.
- APDP
- Monaco’s data-protection authority. The body a processing description is written for, and the one a client may ask you about.
- Law 1.362 / AML
- Monaco’s anti-money-laundering framework, supervised by the AMSF (formerly SICCFIN). Gives regulated professions due-diligence and reporting duties that AI can prepare but never decide.
- KYC
- Know your customer: the identity and background checks a regulated firm must run. Our product Trace does the verification part; the professional decides.
- Extended Monaco
- The Principality’s digital transition programme. bss works is an official partner; its support schemes can cover part of a project for Monaco companies.
- Fonds Bleu
- The Government fund that supports digital projects of Monaco businesses under conditions that change. We check eligibility in the quick-check.
- GEO (generative engine optimization)
- The work that makes ChatGPT, Perplexity, Gemini and Google AI Overviews name your business when a client asks. Plain pages, consistent facts, structured data, measured by citations.
- SEO
- Being found on Google. The same clean site that works for GEO works for SEO; AI makes several language versions affordable, a person keeps them true.
- Voice agent
- An assistant that answers the phone, understands the caller in their language, books or takes a message, and hands over to a person when needed. Our phone-answering guide covers it.
- WhatsApp Business API
- The official way for software to send and receive WhatsApp messages for a company, within rules on templates and consent. How an assistant handles WhatsApp.
- Model context / knowledge base
- The documents an assistant is allowed to read to answer: your price list, your FAQ, your procedures. Kept by you, cited by the assistant.
- Copilot (Microsoft 365)
- Microsoft’s assistant inside Word, Outlook and Teams. Good for drafting by hand; it does not run a process on its own. We often build alongside it.
- Agentic
- A fashionable adjective for software that takes several steps on its own toward a goal. Ask what it may do without a person before being impressed.
- Proof of concept
- A demonstration that something can work, usually on toy data. We prefer the prototype: two weeks on your real data, so the proof is about your business.
A word missing?
Ask it on the quick-check; we add the ones people ask.
ben@bss.mc · a reply within one working day