Automation terms, without the jargon.

The kind of AI that reads and writes text. It powers tools that draft replies, summarize documents, classify enquiries, or answer questions from your business information.

When it matters: When a vendor says "AI", they usually mean an LLM. Ask what it is being asked to do, what information it sees, and where a human still checks the result.

A way to give AI the right business information before it answers. Instead of guessing, it looks up the relevant policy, product detail, email history, or knowledge-base article first.

When it matters: This matters whenever the answer must be based on your business, not on the AI model's general knowledge. It is one of the main ways to reduce confident but wrong answers.

An AI system that can choose its next step instead of following a fixed checklist. That flexibility can be useful, but it also makes the system harder to control.

When it matters: If a vendor proposes an "agent", ask what it is allowed to do, what it is not allowed to do, and what happens when it gets something wrong.

How much information the AI can look at in one request. A bigger window lets it read more, but it usually costs more and can slow the workflow down.

When it matters: A good system gives the AI only the information it needs for that task, rather than sending your whole knowledge base every time.

Extra training that teaches an AI model to behave better on a narrow task. It is useful sometimes, but most business workflows do not need it.

When it matters: If a vendor proposes fine-tuning immediately, ask why a simpler lookup-based system is not enough. For most SMEs, the simpler approach is cheaper and easier to maintain.

When AI gives a confident answer that is wrong.

When it matters: Any workflow that touches customers, money, or compliance needs checks against this. The simplest protection is to make the AI answer from your approved information and keep human approval on risky steps.

A repeatable sequence of steps that runs when something happens: a form is submitted, an email arrives, a booking changes, or a report is due.

When it matters: For most business problems, a clear workflow is safer and cheaper than an open-ended AI agent. It does the same job the same way every time.

A workflow that can run again without creating duplicates. If a step fails and retries, it should not send the same email twice or create the same order twice.

When it matters: Reliable automation needs retries. Ask what happens if a step fails halfway through and runs again.

A way for one tool to tell another tool that something just happened. When a customer fills out your contact form, a webhook can alert your CRM immediately.

When it matters: Webhooks are the difference between systems updating in real time and systems updating ten minutes later. Most modern integration work runs on webhooks.

A structured way for software tools to exchange information. If a tool has an API, another system can usually read from it, update it, or trigger actions inside it.

When it matters: When evaluating a tool, "does it have an API" often means "can this be connected to the rest of the business". Modern SaaS usually says yes. Older industry tools often make this harder.

Two ways for a system to notice that something changed. Webhooks are instant alerts. Polling means checking every few minutes to see whether anything new happened.

When it matters: Reliable systems often use both: webhooks for speed and polling as a backup. Ask how the workflow recovers if an alert is missed.

The one system that is treated as correct for a specific piece of information. If systems disagree, this one wins.

When it matters: Before connecting tools, decide where customer data, bookings, invoices, inventory, and tasks officially live. Without that, automation spreads messy data faster.

Self-hosted means the software runs in an account or server you control. Cloud means the software vendor runs it for you.

When it matters: Self-hosting can lower cost and increase control, but someone must maintain it. Cloud is easier, but you depend more on the vendor.

The billing unit for what an AI model reads and writes. More input and longer answers cost more.

When it matters: You do not need to calculate tokens yourself, but you should ask for an expected monthly AI usage cost. It is usually small, but it should not be a surprise.

Thailand's personal data protection law. It matters when your workflow handles customer, staff, patient, or applicant information.

When it matters: Most SME workflows can use standard business software responsibly, but sensitive sectors need more care around access, storage, and human review.

A contract between you and any tool that processes your customers' data. Spells out what they can and cannot do with that data. Most major SaaS tools have a standard DPA you can sign.

When it matters: Before pushing customer data into any new tool, sign their DPA. It is usually free, takes minutes, and is the difference between defensible and indefensible PDPA compliance.