Glossary
The jargon, translated.
Plain-English definitions of the AI, automation, and security terms that come up most in real engagements — no jargon for jargon's sake.
AI & Automation
- AI Agent
- Software that doesn't just answer a question, but takes multi-step action on your behalf — checking a system, making a decision, and doing something with the result, without a human clicking through each step.
- Automation vs. AI
- Automation runs the same steps every time a trigger fires; it doesn't think. AI adds judgment — it can handle exceptions and unstructured text the rules didn't anticipate. Most useful systems are both: automation for the repeatable parts, AI for the parts that need judgment.
- LLM (Large Language Model)
- The technology behind tools like Claude, ChatGPT, and Gemini. It's trained on huge amounts of text to predict and generate language — which is why it can read a document or draft an email, but doesn't "know" things the way a database does.
- RAG (Retrieval-Augmented Generation)
- A way of connecting an LLM to your actual documents or database before it answers, so it responds using your real information instead of guessing from what it was trained on.
- Hallucination
- When an AI states something confidently that isn't true. It's the single biggest reason AI systems need a verification layer before they touch anything customer-facing or compliance-sensitive.
- Prompt
- The instruction you give an AI system. In a production system, prompts are engineered and tested like code, not typed fresh every time by whoever's using the tool that day.
- Fine-tuning
- Additional training that adjusts a model's behavior for a specific task. Most business use cases don't need this — RAG and good prompting solve the same problem more cheaply and safely.
- Shadow AI
- Employees using AI tools on their own, without approval, policy, or oversight. It's usually already happening in your business whether you've noticed or not — exactly what our free AI Maturity Assessment is designed to surface.
Security & Compliance
- Zero Trust
- A security model that assumes no user, device, or system should be trusted by default, even inside your own network. Every request is verified, every time, rather than trusting anything once it's "inside."
- Data Governance
- The rules for who can access, use, or move your data, and how that's actually enforced and logged. Without enforcement, "we have an AI policy" is just a document nobody follows.
- PIPEDA
- Canada's federal private-sector privacy law. If you collect personal information from customers or employees, it applies to you — AI or not.
- Data Residency
- Where your data is physically stored. It matters because storing data outside Canada can expose it to another country's laws, including government access to that data.
- Encryption at Rest / In Transit
- Whether your data is scrambled while sitting in storage ("at rest") and while it's moving between systems ("in transit"). Both matter — having one without the other is a common, and commonly overlooked, gap.
- MFA (Multi-Factor Authentication)
- Requiring a second proof of identity beyond a password — a code, an app confirmation, a security key. It's the single cheapest control that stops the most common account breaches.
- Audit Logging
- A permanent record of who did what, and when, inside a system. Without it, you can't prove compliance, investigate an incident, or even know something went wrong until a client tells you.
- Security Clearance (GoC Level II)
- A government-issued check confirming a person can be trusted with sensitive but non-classified federal information. It's what lets a consultant start government work immediately instead of waiting months for vetting.
- Compliance vs. Security
- Compliance means you can prove you followed a required standard. Security means you're actually protected. You can be fully compliant and still get breached — the paperwork isn't the same thing as the protection.
- SLA (Service Level Agreement)
- A written commitment to a specific level of service — response time, uptime, resolution time — with consequences if it's missed. If a vendor won't put numbers in writing, that's the number worth paying attention to.
Have a term that's still not clicking?
Ask us directly, or take the free AI Maturity Assessment to see where these ideas actually apply to your business.