GenAI Demo Navigator

Customer

Concierge agent (AI)

Care team (human)

Quality & knowledge

1Customer makes contact 2 2

2Recognition & context assembly 1 1

3Intent triage & language 1 1

4Order & delivery answers 1 1

5Advice, membership & product answers 2 2

6Gated actions 2 2

7Warm handover 1 1

8Wrap-up & QA scoring 1 2

React Flow

Issue → AI potential map

Derived from this view’s pain points and potentials — the operating concept column is the production thinking.

Issue	Impact	AI pattern	Operating concept	KPI effect	Demo
Seasonal gifting peaks triple contact volume on known dates	cost, cx, qualitythe peak is forecastable but native-quality staffing cannot flex to match it	conversational-agent	Reads are free, writes are gated: the agent may read orders, shipments, memberships and the knowledge corpus at will, but return labels, membership changes and repair intakes are confirmed in-session or routed to a human. Escalation is warm — the transcript and context travel with the customer.	Cost per contact ↓ Peak queue time ↓	Glanzstein Concierge — Customer Care & Gift Advice Agentlive
Seasonal gifting peaks triple contact volume on known dates	cost, cx, qualitythe peak is forecastable but native-quality staffing cannot flex to match it	voice-agent	One agent definition serves both channels; voice adds caller identification by number and stricter confirmation before any write. Availability is the point — the peak does not respect boutique hours.	Peak abandonment rate ↓ Service hours ↑
Seasonal gifting peaks triple contact volume on known dates	cost, cx, qualitythe peak is forecastable but native-quality staffing cannot flex to match it	agent-assist	The human stays in front; the copilot works the systems. Every fact it surfaces cites its source, so a stylist can trust it in front of the customer — support, never replace.	Handle time ↓ First-answer accuracy ↑	Glanzstein Advisor — Boutique & Care Copilotplanned
Seasonal gifting peaks triple contact volume on known dates	cost, cx, qualitythe peak is forecastable but native-quality staffing cannot flex to match it	automated-qa	The rubric is versioned content, so what "on brand" means is inspectable and arguable. Scores drive coaching and knowledge fixes; they never auto-discipline a named employee — the human final word is a house rule.	QA coverage ↑ Brand-voice variance ↓	Glanzstein Care KPIs — Containment & QA DashboardplannedGlanzstein Admin — Content & Eval Consoleplanned
Native-quality service across ~130 markets is structurally expensive	cost, cx, qualitylanguage coverage is a step function in headcount, but not for an agent	conversational-agent	Reads are free, writes are gated: the agent may read orders, shipments, memberships and the knowledge corpus at will, but return labels, membership changes and repair intakes are confirmed in-session or routed to a human. Escalation is warm — the transcript and context travel with the customer.	Cost per contact ↓ Peak queue time ↓	Glanzstein Concierge — Customer Care & Gift Advice Agentlive
Native-quality service across ~130 markets is structurally expensive	cost, cx, qualitylanguage coverage is a step function in headcount, but not for an agent	voice-agent	One agent definition serves both channels; voice adds caller identification by number and stricter confirmation before any write. Availability is the point — the peak does not respect boutique hours.	Peak abandonment rate ↓ Service hours ↑
Kreis membership administration is high-volume procedure	cost, cxprocedural work with almost no judgement content, on the highest-LTV segment	conversational-agent	Reads are free, writes are gated: the agent may read orders, shipments, memberships and the knowledge corpus at will, but return labels, membership changes and repair intakes are confirmed in-session or routed to a human. Escalation is warm — the transcript and context travel with the customer.	Cost per contact ↓ Peak queue time ↓	Glanzstein Concierge — Customer Care & Gift Advice Agentlive
Where-is-my-order dominates the contact mix	cost, cxthe answer exists in systems; only the retrieval is manual	conversational-agent	Reads are free, writes are gated: the agent may read orders, shipments, memberships and the knowledge corpus at will, but return labels, membership changes and repair intakes are confirmed in-session or routed to a human. Escalation is warm — the transcript and context travel with the customer.	Cost per contact ↓ Peak queue time ↓	Glanzstein Concierge — Customer Care & Gift Advice Agentlive
High-intent gift shoppers abandon without advice	cx, costadvice capacity, not demand, is the limiting factor on conversion	conversational-agent	Suggestions come only from the live catalog with availability checked at answer time — the agent never recommends what it cannot promise. Advice stays advice: checkout remains the customer's explicit act.	Advised-session conversion ↑ Advice availability ↑	Glanzstein Concierge — Customer Care & Gift Advice Agentlive
Product and care knowledge turns over with every drop	quality, risk, speednobody can tell a stale answer from a current one at the counter	conversational-agent	Reads are free, writes are gated: the agent may read orders, shipments, memberships and the knowledge corpus at will, but return labels, membership changes and repair intakes are confirmed in-session or routed to a human. Escalation is warm — the transcript and context travel with the customer.	Cost per contact ↓ Peak queue time ↓	Glanzstein Concierge — Customer Care & Gift Advice Agentlive
Product and care knowledge turns over with every drop	quality, risk, speednobody can tell a stale answer from a current one at the counter	agent-assist	The human stays in front; the copilot works the systems. Every fact it surfaces cites its source, so a stylist can trust it in front of the customer — support, never replace.	Handle time ↓ First-answer accuracy ↑	Glanzstein Advisor — Boutique & Care Copilotplanned
Product and care knowledge turns over with every drop	quality, risk, speednobody can tell a stale answer from a current one at the counter	automated-qa	The rubric is versioned content, so what "on brand" means is inspectable and arguable. Scores drive coaching and knowledge fixes; they never auto-discipline a named employee — the human final word is a house rule.	QA coverage ↑ Brand-voice variance ↓	Glanzstein Care KPIs — Containment & QA DashboardplannedGlanzstein Admin — Content & Eval Consoleplanned
Product and care knowledge turns over with every drop	quality, risk, speednobody can tell a stale answer from a current one at the counter	knowledge-ops	Knowledge is edited, evaluated and published as a unit — the eval suite runs before a change goes live, so a seasonal drop is a reviewed release rather than a scramble across channels.	Time to publish a drop's knowledge ↓ Stale-answer incidents ↓	Glanzstein Admin — Content & Eval Consoleplanned
Returns and exchanges take manual touches per parcel	cost, speed, cxevery day of refund latency creates repeat contact volume	conversational-agent	Reads are free, writes are gated: the agent may read orders, shipments, memberships and the knowledge corpus at will, but return labels, membership changes and repair intakes are confirmed in-session or routed to a human. Escalation is warm — the transcript and context travel with the customer.	Cost per contact ↓ Peak queue time ↓	Glanzstein Concierge — Customer Care & Gift Advice Agentlive
Returns and exchanges take manual touches per parcel	cost, speed, cxevery day of refund latency creates repeat contact volume	outbound-agent	Messages are generated from case facts and templates under a named sender; deadline watchdogs escalate a stalled case before the promise date is breached rather than apologising after. What leaves the house is logged with the state that produced it.	Status-chasing contacts ↓ Promise-date breaches ↓	Glanzstein Atelier — Agentic Repair & Aftercare Desklive
Repair intake loops for weeks before a technician sees the piece	speed, cx, costcycle time is dominated by intake completeness, not repair effort	outbound-agent	Messages are generated from case facts and templates under a named sender; deadline watchdogs escalate a stalled case before the promise date is breached rather than apologising after. What leaves the house is logged with the state that produced it.	Status-chasing contacts ↓ Promise-date breaches ↓	Glanzstein Atelier — Agentic Repair & Aftercare Desklive

Figures marked “actual” come from FY2025 public reporting; ranges marked “illustrative” are industry-plausible benchmarks, not company data.