An AI chatbot for business is software that accepts a question in natural language and returns an answer grounded in content or systems the operator has approved. The chatbot vs AI agent line matters from the first scoping call. An AI agent, on Sigi’s AI agent development page, is different: it completes multi-step tasks through tools and APIs with controls, approvals, and auditability. Copilots and generation features sit on generative AI development. Retrieval-heavy assistants sit on RAG assistants.
Decide the job before picking a model
- FAQ deflection: a knowledge base chatbot that answers from reviewed content. Fail closed when the answer is not in the corpus. A no-code chatbot for your website is often enough at this level.
- Authenticated help: an LLM chatbot for customer support may read account or order state through APIs, with the same permissions a support agent would have.
- Human handoff: collect context and open a ticket or chat with a person. Many “chatbots” should stop here.
- Agent work: create or update records, trigger workflows, or take payments. That is the AI-agent product, not a widget on a marketing site.
How to build an AI chatbot in five evaluable steps
- Write the allowed questions and the forbidden actions. If the bot must never invent a price or a clinical instruction, that is a product rule, not a prompt afterthought.
- Choose retrieval before “fine-tune everything.” A RAG chatbot (retrieval-augmented generation) searches owned documents and then writes the answer; Sigi’s RAG page exists because many business chatbots are exactly that.
- Put a human fallback on day one. Unanswered questions are a dataset, not a failure to hide.
- Measure answer faithfulness against the corpus, not only user thumbs-up; that is how hallucinated answers get caught. Cost and reliability work is the subject of Sigi’s LLM optimization page.
- Promote tool use only after the Q&A path is boringly correct. Sigi’s agent page starts with workflow mapping, then a small prototype, then phased rollout with permission boundaries.
What has to be designed besides the chat window
- Source of truth: help centre, policy PDFs, product catalog, or live APIs — named, versioned, and reviewable.
- Identity: an anonymous chatbot on your website versus signed-in support with audit logs.
- Languages and tone: ServiPR is bilingual because the market is; a bot should match how customers already write, but that is a requirement, not a ServiPR-AI claim.
- Regulated content: Allfor Care shows how UK care data is permissioned. A care chatbot that quotes clinical records would inherit those access rules. Sigi has not published such a bot.
Related Sigi reading
Capability pages: AI agents, generative AI, RAG assistants, and LLM optimization. If you want a partner for custom AI chatbot development, start on Sigi’s AI development services page. For a regulated records system rather than a bot, see CQC and UK GDPR for care software. For product cost context (mobile, labelled estimates), see how much it costs to build a mobile app.