Engineering capabilities
- Embedded chat experience
- Approved knowledge retrieval
- Lead qualification
- Service-page recommendations
- Human handoff and transcript context
AI & Chatbots
Website Chatbot
Branded website assistants grounded in approved content, with lead capture, service routing and a clear path to a person.
Security & hardening
- Consent and privacy notices
- Input validation and abuse controls
- No browser-exposed model keys
- PII minimisation
- Retention limits
Website Chatbot reference flow
The chatbot is one controlled interface into existing operations, not a second ungoverned system of record.
-
01
Conversation
- Visitor request
- Intent and context
- Consent state
Only the information needed for the requested workflow is collected.
-
02
Control layer
- Validation
- Identity check
- Tool allowlist
- Approval rules
Every lookup or action is constrained before it reaches a business system.
-
03
Integration
- Website widget
- Approved knowledge
- Lead pipeline
- Analytics
Documented APIs and webhooks are used with least-privilege credentials.
-
04
Handoff
- Human owner
- Conversation summary
- Next action
- Audit log
A person receives enough verified context to continue without asking the visitor to start again.
Frequently asked questions
What is required for Website Chatbot?
A supported API or webhook, scoped credentials, agreed fields and a clear owner for exceptions. These are confirmed during discovery.
Will the chatbot be allowed to change records?
Only actions explicitly approved in scope. Read-only access is the default, and consequential changes can require confirmation or human approval.
What happens when the connected system is unavailable?
The visitor receives a clear fallback, the request is retained where permitted, and the failure is logged for an operator rather than silently lost.
Is conversation data copied everywhere?
No. Field allowlists and retention rules limit what is transferred and how long it remains in each system.
Scope the work before you commit budget
Send the environment details and get an engineered proposal with assumptions, risks and a rollback path.