Three kinds of AI chatbot we build
Customer-facing
Embedded on your website, in Messenger or WhatsApp. Answers FAQs, looks up orders, books appointments, hands off to human support when uncertain.
Internal staff Q&A
Lives in Microsoft Teams or Slack. Answers policy, compliance, product or HR questions from your internal documents, with full per-user audit logs.
Sales qualification
Captures inbound enquiry intent, asks qualifying questions, books a meeting or routes to your sales team, with the conversation summary attached to the CRM record.
How we make sure the chatbot stays accurate
The biggest concern most NZ businesses have about AI chatbots is hallucination, the model confidently making things up. We design every chatbot with the same defences:
- Retrieval-augmented generation (RAG). Every answer is grounded in passages retrieved from your documents at query time. The model is instructed to use only what's retrieved.
- Refuse-and-escalate. When the chatbot can't find a confident answer, it says so clearly and offers a handoff to a human, rather than guessing.
- Source citations. Every answer can show the document and section it came from, so users (and auditors) can verify.
- Confidence routing. Low-confidence answers are flagged and routed for human review before being sent.
- Tone and policy guardrails. Hard rules about what the bot won't say (pricing it can't quote, advice it can't give, competitors it won't mention).
- Continuous evaluation. A test set of representative questions runs against every model/prompt change before deployment.
Where chatbots commonly live
- Website widget. Embedded in any HTML page. Custom-styled to match your brand.
- Microsoft Teams. Internal Q&A bots with SSO, channel and DM access.
- Slack. Internal Q&A or workflow-trigger bots.
- Facebook Messenger / Instagram DM. Customer-facing for retail and service businesses.
- WhatsApp Business. High-volume customer support and order updates.
- Zendesk / Freshdesk. Inside your existing helpdesk as a triage/draft-reply assistant.
- Voice (phone). AI voice agent for inbound call routing and FAQ. Available where the use case justifies it.
Integration examples
- Retail: Customer asks "where's my order?" → bot looks up Shopify by email → returns tracking link → logs interaction to CRM.
- Insurance: Claimant asks about a policy clause → bot retrieves from policy PDF → cites section number → escalates to human if claim-specific.
- Logistics: Internal bot answers "what are the duty rules for shipping X to Y?" from your internal compliance docs, with full audit log per query.
- Professional services: Inbound enquiry asks about pricing → bot qualifies (industry, size, timeline), books a meeting in HubSpot, attaches transcript.
Frequently asked questions
What kinds of AI chatbots do you build?
Three main categories: (1) customer-facing web/messaging chatbots that answer FAQs, triage support tickets and book appointments; (2) internal staff chatbots that answer policy, product or compliance questions from your internal documents with full audit logs; (3) sales-qualification chatbots that capture lead intent and pass qualified leads to your CRM.
Where can the chatbot live?
Anywhere your customers or staff already are: embedded in your website, inside Microsoft Teams or Slack, on Facebook Messenger, on WhatsApp Business, on Zendesk or Freshdesk, or as a standalone app.
Will it hallucinate?
We design every chatbot with retrieval-augmented generation (RAG), answers are grounded in your documents, not the model's pre-training. The bot is also instructed to refuse and escalate when it cannot find an answer, rather than guessing.
Can it integrate with our CRM and ticketing system?
Yes. Customer-facing bots can lookup the user in HubSpot/Salesforce, check open tickets in Zendesk/Freshdesk, query orders in Shopify, and write new tickets, opportunities or contact records back. Every interaction is logged.
What about data residency and privacy?
Default deployment is AWS Sydney (ap-southeast-2) for ANZ data residency. We can architect on-prem or NZ-only deployments where required. PII is redacted from prompts before they hit the LLM, and full conversation logs are retained per your retention policy.
Want to scope a chatbot for your business?
A free 30-minute strategy call is enough to identify whether a chatbot will pay back, what kind to build, and what it will realistically cost.
Get in touch