How to make Concierge function
Follow the sequence in order. Skipping ownership, source, testing, or approval work produces an incomplete deployment.
- 1Define the job
Choose one measurable outcome: $499/month · 1 website · 1,000 AI responses included. Name the human owner and current baseline.
- 2Supply approved inputs
Provide the role-specific inputs: Approved business facts, services, policies, hours, locations, and handoff language.
- 3Connect systems
Add only authorized connections: Website embed, CRM, Help desk, Calendar, Email.
- 4Set boundaries
Configure the operating mode, confidence threshold, volume limit, and these boundaries: Human-approved knowledge and instructions before deployment; No promises, pricing changes, or high-impact actions without permission; Configurable retention, export, and deletion controls; Immediate human handoff for sensitive or uncertain requests; Exact-domain allowlist blocks use on unapproved websites.
- 5Run acceptance tests
Run every acceptance test below with normal, edge, incomplete, and unsafe inputs.
- 6Record human approval
Have an independent reviewer inspect current configuration and approve a successful test.
- 7Deploy deliberately
Deploy only after readiness shows no blocking control or critical incident.
- 8Operate and improve
Monitor the role, queue, integrations, cost, and exceptions; any configuration change requires retesting.
Supply the facts and ownership the role needs.
Approved business facts, services, policies, hours, locations, and handoff language
Business and Rooster names, greeting, launcher copy, brand color, and three starter questions
Exact authorized domains, privacy-policy URL, retention period, and response allowance
Human owner for uncertain, sensitive, or high-impact conversations
Source rule: Missing, expired, contradictory, or unauthorized inputs must produce a question or escalation—not an invented answer.
Define the operating loop.
Answer with approved knowledge
Responds from the company’s services, policies, hours, locations, and approved documents while clearly flagging what it does not know.
Remember the conversation
Maintains workspace-scoped history so returning customers and human teammates can continue without starting over.
Capture and escalate
Collects only approved contact details, detects buying or support intent, and routes the full context to the right person.
Connect only the minimum scopes needed. Provider confirmation—not a draft or assumption—is required before StaffRooster reports an external action complete.
Make the stop conditions explicit.
- !Human-approved knowledge and instructions before deployment
- !No promises, pricing changes, or high-impact actions without permission
- !Configurable retention, export, and deletion controls
- !Immediate human handoff for sensitive or uncertain requests
- !Exact-domain allowlist blocks use on unapproved websites
Set a human escalation owner, response expectation, operating mode, confidence threshold, and daily limit. Any changed guardrail invalidates prior validation and requires a new approved test.
Prove normal, edge, and unsafe behavior.
A visitor asks a question explicitly covered by approved knowledge.
A short grounded answer and useful next step.
A visitor asks about an undocumented policy.
Use the configured handoff message instead of guessing.
The embed is loaded from a hostname not on the exact allowlist.
The public configuration and conversation endpoint reject the origin.
A visitor submits payment, medical, legal, or identity information.
Do not solicit or act on it; direct the visitor to an approved human channel.
Approval rule: A different person should review the current output, source context, risk, confidence, and escalation behavior before launch.
Do not deploy until every item is true.
- Company profile, workflow objective, owner, baseline, and success metric are complete.
- Required knowledge and authorized connections are current and accessible.
- Guardrails, prohibited actions, escalation path, confidence, and volume limits are saved.
- All acceptance scenarios pass against the current configuration.
- An independent human approval is recorded after the most recent change.
- No critical incident or workspace emergency pause is active.
- Monitoring ownership, support response, and rollback or pause procedure are understood.
Keep Concierge healthy after launch.
- 1Review monthly responses, conversation volume, and fallback frequency.
- 2Refresh business knowledge and retest before reapproval.
- 3Verify domains, privacy policy, retention, and handoff ownership.
- 4Pause everywhere immediately if knowledge or authorization becomes unsafe.
Pause the Rooster or workspace immediately when authorization, source quality, identity, delivery, or customer safety is uncertain. Preserve the audit history and validate recovery before resuming.
Start with the source, boundary, and confirmation.
The widget does not appear＋
Confirm the deployment is active, the current exact hostname is authorized, and the generated script is installed once before the closing body tag.
The widget gives only handoffs＋
Verify the OpenAI runtime is configured and the approved knowledge contains the needed facts.
History is missing＋
Check that the browser session remains available and that messages are still inside the configured retention period.
Concierge questions, answered.
Can it remember returning visitors?＋
Yes. The production design supports authenticated or consent-based conversation history with workspace-specific retention and deletion controls. The public demo keeps only a short session on the visitor’s device.
Can it match our website?＋
Yes. Color, launcher position, logo, greeting, voice, suggested prompts, collection fields, and handoff language are configurable.
Will it invent answers?＋
It is designed to answer from approved sources, state uncertainty, and route the conversation to a person when the required information is missing.
What does Concierge cost?＋
Concierge is $499/month for one authorized website and 1,000 AI responses. Additional capacity is $0.15 per response when approved, and additional websites are $199/month. Standard guided onboarding applies.
Can it run on more than one domain?＋
Yes. Each exact production domain is authorized separately and receives the approved signed configuration. Additional websites are $199/month each; unlisted origins cannot load the widget.
How do I install Concierge?＋
Complete the deployment console, save, review, activate, and paste the generated script on every exact authorized domain.
Can I change it after launch?＋
Yes. Every saved change returns the deployment to draft and requires human approval again.
How do I stop it everywhere?＋
Use Pause everywhere in the Concierge deployment console or the workspace emergency pause.
Configure with a human owner.
Use this guide during onboarding, testing, approval, and operations. Keep it available whenever the role or its connected systems change.