E-commerce
June 28, 2026
"Hello! Great choice! Use code LUXE10!" on a €2,800 ring, and the magic disappears. The customer closes the widget, calls the boutique, or leaves without saying a word.
Heeya reminds us that a jewelry bot must be restrained, attentive, never pushy: it guides with tact, not like a promo script (Heeya, luxury bot 2026). Farever Reach recommends designing the luxury concierge for routing by priority, not deflection: routine handled by AI, emotion and gifting handled by human with full context (Farever Reach, digital concierge 2026).
This guide #285 configures the luxury e-commerce AI chatbot: tone, discretion, trust, without the robotic feel. It complements luxury support (#284) (human) and premium bot (#213) (general advisory): here, focusing on the luxury vertical with brand-specific codes.
Summary
Why does a standard bot destroy the perception of luxury?
A poorly configured luxury bot costs more than no bot at all: it signals that the brand treats the customer as volume.
Five robot effects that kill trust
Retail enthusiasm: "Awesome!", emojis, unsolicited promo codes
Generic short answers: "Consult our help center"
Empty handoff: the customer repeats everything to the human concierge
Aggressive proactivity: popup at 3 seconds on a contemplative PDP
Material hallucination: invented gold/leather grain claim
What luxury expects from the bot
Relevance Digital: the luxury bot must replicate discretion, curation, measured engagement, not transactional speed (Relevance Digital, luxury shopping 2026). BCG: 62% of luxury customers fear losing human contact; AI must remain invisible or understated (BCG, luxury 2025).
Example DTC leather goods
Standard bot on a €1,650 bag: chat bounce 58%, CSAT 2.9. Redesign concierge #285: CSAT 4.5, assisted conversion +19%, "discount volume" complaints 0 over 90 days.
How does it differ from luxury support (#284) and premium bot (#213)?
Three neighboring contents, three layers.
Human luxury support (#284)
Luxury support (#284): concierge team, LUX-* macros, gift workflow. #285: bot configuration that respects the same codes.
Premium bot (#213)
Premium (#213): hands-off consultative, skincare/furniture. #285: gift discretion, routing by key challenge, silence, engraving, authenticity specific to luxury.
Prompts (#282) and governance (#142)
Prompts library (#282) provides the structure. Governance (#142) legal framework. #285: SYSTEM-LUXE prompt and luxury intents matrix.
Test grid (#283)
Test grid (#283): validate tone and escalation before luxury go-live.
What is the architecture for an AI luxury concierge?
Four-block luxury AI concierge architecture, in this build order.
Block 1: Tier rules (before the model)
Farever Reach: write first AI-only / AI-assisted / human-immediate by interaction type, not by customer tier alone. Example: wedding gift = human-immediate even for a first-time buyer.
Block 2: Contextual handoff
Payload: transcript, SKUs viewed, intent, customer tone (calm/frustrated), gift notes. A lossy handoff does more harm than a delay (Farever Reach).
Block 3: Conversation layer
In-house RAG: cheat sheets, care guides, craftsmanship details, return policy. Exclude sales promotions from the luxury bot corpus.
Block 4: Brand voice gate
Post-generation filter: blocklist promotional words, emojis, informal address (tutoiement). The model proposes; the gate decides (Farever Reach, luxury personalization 2026).
Silence principle
No proactive follow-ups if the customer is inactive for 60s+. Silence is a feature, not abandonment.
How do you write a luxury system prompt without sounding robotic?
Here is an adaptable luxury system prompt (excerpt #282).
SYSTEM-LUXE v1
"You are the digital assistant of [MAISON], a [jewelry/leather goods/fashion] house. Language: French, formal you (vouvoiement). First sentence: clearly state that you are an AI, with an immediate "speak to an advisor" option.
Tone: calm, attentive, full sentences. Never use: super, awesome, deal, promo, urgency, emoji. You answer, clarify, and suggest with restraint. One question at a time.
Sources: RAG passages and product sheets only. If material, warranty, or delivery time is missing: offer a human advisor, do not invent.
Gift, engraving, authenticity, dispute, press request: immediate [ESCALATE:concierge] with a summary of the context."
Few-shot welcome PDP
Proactive 90s (AOV > threshold): "A question about this piece? I can provide information about the materials, dimensions, or personalization options."
To avoid: "Hello! Need help with your purchase? -10% today!"
Few-shot recommendation
Client: "Wedding anniversary gift, flexible budget." Bot: "Congratulations. Would you prefer a discreet piece to wear daily, or a more symbolic model? I can also arrange gift wrapping with no price visible."
Few-shot elegant refusal
Client: "Do you offer -15% like on Instagram?" Bot: "Our collections are offered at the price displayed on the website. I can provide information about materials or personalization options if you wish."
Which intents for automated, assisted, or immediate human interaction?
Luxury bot intents matrix: three columns, binary rules.
AI-only (routine, low stakes)
WISMO with reliable tracking
Boutique hours, documented care
Dimensions/weight from product sheet
Standard text return policy (non-dispute)
AI-assisted (draft → optional validation)
Comparison of 2 pieces in the same range
Size guide with cautious phrasing
Look completion suggestion (1 recommendation max)
Human-immediate (zero client auto-reply)
Surprise gift, handwritten note, sender anonymity
Engraving, customization, workshop repair
Authenticity, certificate, dispute serial number
Frustrated customer, mention of lawyer/chargeback
Explicit "talk to someone" request
Tier rule per interaction
Intent stakes > LTV score. Message "it is for a wedding proposal this Saturday" routes to human even without a VIP tag.
Direct routing example
Message: "engagement ring Saturday, inside engraving possible?" → human-immediate lux_gift + lux_engrave. Bot collects desired text (80 char. max) + event date, then escalates without promising undocumented workshop turnaround times.
AI-only example
"What is the shipping time for order #4421?" → tracking lookup → 3-line response + tracking link. Zero upsell.
How to manage gifts and discretion via the bot?
The gift and discretion intents are not handled by an auto-bot; the bot qualifies and transfers.
LUX-GIFT bot flow (qualification)
Bot detects gift intent → "I am putting you in touch with our gift team to prepare this with discretion."
Minimal collection: requested date, gift wrap yes/no, handwritten note yes/no
[ESCALATE:concierge] + payload section 3
Customer message: "[Advisor first name] is taking over your request; you won't have to repeat the details."
Gift bot prohibitions
Promising "without invoice" if ops is not confirmed
Asking for the recipient's name in a public Instagram DM chat
Mentioning order amount in response
Alignment #284
Human-side LUX-GIFT-01 macros pick up the thread. See gift order.
Gift qualification bot script (copy)
"I understand that discretion is essential. I am putting you in touch with our gift team. Would you like signature wrapping and a handwritten note? If so, please enter the text (80 characters maximum)."
How can we make the widget proactive without breaking the user experience?
The luxury widget proactivity signals availability, not sales urgency.
Display Rules
Delay: 60-90 s on PDP AOV > luxury threshold, not upon landing
Frequency: 1 invitation/session, zero if the customer closed the widget
Pages: PDP hero, signature collection; never checkout
Mobile: discreet bubble, no fullscreen
Authorized Proactive Messages
“Any questions about this piece?” · “I can help you with materials or personalization.” · “An advisor is available if you prefer to chat.”
Forbidden Messages
Countdown, promo code, “Only 2 left in stock!”, “Excellent choice!” Heeya: understated proactivity after contemplation, not interruption (Heeya).
Post-Conversation Silence
Customer: “I am thinking about it” → “Take all the time you need. I remain available.” No Day+1 bot follow-up. Optional brand email only if explicit opt-in.
How to build a flawless contextual handoff?
The luxury handoff is the most critical asset in the system.
CONCIERGE-HANDOFF Payload (Internal JSON)
customer_id, first name, language
skus_viewed: last 3
intent_primary + emotional_signal
summary_2_lines: what the customer wants
gift_flags: packaging, note, event date
bot_responses_given: short list
Standard customer message
“I am handing you over to Élise from our concierge team. She has your conversation and your order #8842. You will be contacted within 30 minutes during business hours.”
SLA concierge bot → human
Human-immediate: bot ack < 10 s, human takeover < 30 min store hours. WISMO AI-only: response < 15 s.
Which anti-promo and anti-bot guardrails should be activated?
Fifteen luxury bot guardrails in blocklist + hard-coded rules.
Word/phrase blocklist
super, great, amazing, deal, promo, -XX %, last chance, hurry, limited stock (unless sourced), hey, hello, you have, cheap
Behavioral rules
Max 2 product recommendations/conversation without explicit request
Max 120 words/response unless comparison is requested
Zero coupons generated by LLM
"Talk to an advisor" button visible at all times
Line 1 AI disclosure (EU AI Act)
Pre-production tests
15 tone adversarial cases + 10 gift/engraving cases via grid #283. Brand voice score dim_tone: 100% required on luxury P0.
Material anti-hallucination
Gold claim, carat, leather grain, provenance: cite-or-decline mandatory. See anti-hallucination (#123).
Tone test pack (10 verbatims)
Angry customer delayed package → empathy before status
"Is it real leather?" → cite sheet or escalate
Promo code requested → polite refusal without lecturing
"I am comparing with [competitor]" → focus focus on our house, no denigration
Customer silent for 2 min → no spam follow-ups
Which KPIs should be tracked beyond deflection?
Luxury bot KPIs prioritize relationship, not deflection.
Primary KPIs (monthly)
High-stakes time-to-human: median minutes, target < 30
Handoff context completeness: % complete payload, target > 95%
Luxury CSAT widget: AOV segment > threshold, target > 4.3
12 m repeat purchase: clients who used bot vs. non-users
Tone complaint rate: verbatims "discount/robot", target < 1%
Secondary KPIs
Assisted conversion, bot AOV (monitor 30-day refunds). Forever Reach: useful deflection in ops, poor primary luxury metric.
Monthly Review (30 min)
10 worst CSAT transcripts + 5 best. Adjust prompt or routing. Link bot KPIs (#11).
How does Qstomy deploy an AI luxury concierge?
Qstomy deploys the luxury concierge model: routing intents, tone guardrails, enriched handoff.
Luxury capabilities
SYSTEM-LUXE template: voice blocks + integrated blocklist
Intent lux_gift / lux_repair / lux_auth: human-immediate hard rule
Configurable proactivity: 90s delay, 1×/session
Brand voice gate: post-generation filter for promo/emoji
Concierge payload: Gorgias sidebar sync
In-house RAG: exclude promo chunks
Quantified DTC scenario
Fine jewelry, AOV €1,950, 280 bot conversations/month. Before #285: CSAT 3.1, handoff without context 62%, assisted conversion 8.4%. After SYSTEM-LUXE + Qstomy routing: CSAT 4.6, complete context 97%, assisted conversion 11.2%, 12m repeat bot segment +24% vs control.
See AI support, assisted sales, Shopify, demo.
Which playbooks to launch your luxury bot?
Playbook 1: Tier Intents Matrix (3 hrs)
List 15 intents. Classify AI-only / assisted / human-immediate section 5. Lead concierge + ops validation.
Playbook 2: SYSTEM-LUXE Prompt (4 hrs)
Paste section 4, add 5 in-house few-shots. Test 20 real verbatims. Section 9 blocklist active.
Playbook 3: Handoff Payload (2 hrs)
Configure section 8 JSON in Gorgias/Zendesk. Simulate 5 gift escalations. Agent must never re-ask for context.
Playbook 4: Widget Proactivity (1 hr)
Set 90s PDP AOV > threshold. Test iOS mobile. Verify zero checkout popups.
Playbook 5: Run Luxe Scorecard (4 hrs)
25 cases #283 including 10 tone + 5 gift. Threshold: zero P0 dim_tone failures. Soft go-live 15% luxe PDP traffic.
Useful Links
This week: replace your proactive message with the section 7 phrasing. Measure widget bounce rate 7 days before/after.
Enzo
June 28, 2026
