E-commerce
July 15, 2026
"Can I wash it in the machine?" "What product should I use to remove this wine stain?" "My sofa discolored after cleaning, is it covered?" These questions arrive post-purchase, often on a Sunday evening when the customer has already used the wrong product.
Chitika estimates that 41 to 58% of product tickets are deflected when the AI is trained on product docs and care instructions (Chitika, ticket reduction 2026).
This guide #347 formalizes the product care AI chatbot: when to answer from the care sheet, when to escalate to an agent. It complements support care ops (#346) (policy, human macros) with the angle of post-purchase care AI use cases, responsible AI, and escalation safeguards.
Summary
Why automate maintenance advice with a bot?
Bad care advice costs a ruined product, a warranty dispute, or a negative review. The bot does not replace the leather or textile expert; it cites the official protocol without improvising.
Three risks without a structured care bot
Care hallucination: LLM recommends bleach on silk
Late escalation: damage claim treated as a simple question
Agent inconsistency: three different answers for the same SKU
ADEME points out that correct maintenance extends textile lifespan (ADEME, textile care).
Angle #347
#346 documents policy care, CARE macros, and the responsibility matrix. #347 defines the dedicated care bot: care_* intents, reply vs. escalate rules, anti-hallucination.
Journey stage
Post-purchase Day 0 to Day 730: first wash, stain, storage, post-care damage complaint. Peak care tickets on weekends.
DTC Example
Fabric furniture brand: 85 care tickets/month, agents improvising. After bot care_how_clean + metafields: 64% self-resolved, care_damage_claim_rate 8%, care bot CSAT 4.7/5.
Responsible AI
A care bot must know how to say "I don't know" and escalate rather than recommending a product not listed in care_products_ok.
Care ticket volume
In furniture, leather goods, and outdoor catalogs, care is often in the top 5 post-delivery queries. A 24/7 care bot responds on Sunday when the customer discovers a stain, before they use the wrong product.
How does it differ from customer care support and other bots?
Seven neighboring contents, seven roles.
Support care ops (#346)
Guide #346: policy, 8 CARE macros, human routing. #347 implements the grounded AI care layer.
Lifespan (#345)
Lifespan (#345): how long the product lasts. #347: how to maintain it via bot, not waiting time.
Manual bots (#230)
Manual bots (#230): user manual PDF. #347 focuses on cleaning and storage protocols from metafield care.
Usage tickets (#230 reduce)
Usage (#230): intent "I cannot use it". #347 = preventive maintenance.
Troubleshooting (#229)
Troubleshooting (#229): functional breakdown. #347: aesthetic maintenance damage, not breakdown.
AI Governance (#124)
AI Governance (#124): global framework. #347 applies responsible care escalation to the maintenance case.
Promise #347
Intents care_*, answer/escalate tree, data metafields, care anti-hallucination, handoff damage claim, KPI care_bot_resolution.
Which care intents should the bot classify?
Map out the care bot intents before building the flows.
Ten post-purchase care intents
care_how_clean: standard cleaning protocol
care_stain: wine, oil, ink stain by type
care_storage: off-season storage
care_forbidden: customer asks for a forbidden method
care_frequency: how often to maintain
care_kit: compatible cleaning product to purchase
care_damage_claim: damage after maintenance or use
care_vs_warranty: defect or maintenance neglect
care_pro_cleaning: professional cleaning authorized
care_symbol_decode: translate GINETEX care symbols
Mandatory session fields
sku, material_type, care_protocol_source, stain_type, photos_received (bool), damage_claim (bool), escalation_reason. See taxonomy (#135).
Mining 90-day tickets
Export "wash", "clean", "stain", "maintenance", "damaged". Group by top 20 SKU. Prioritize care flows on 80% volume.
Priority verbatims
"Machine 40 or 60?", "coffee stain on sofa", "cracking leather", "winter storage for tents". Five formulations cover 65% of DTC furniture and leather goods care tickets.
When does the bot respond and when does it escalate?
The respond vs escalate care tree is the core of #347. Respond = documented protocol. Escalate = risk, dispute, or missing data.
Five decision gates
SKU Gate: product identified, care metafield present
Intent Gate: care_how_clean, stain, storage (not damage claim alone)
Protocol Gate: response exists in care_instructions
Risk Gate: no bleach/solvent on sensitive material
Exit Gate: bot response, handoff_agent, handoff_warranty
Respond (autonomous bot)
care_how_clean with complete metafield. care_stain if stain_protocol is documented. care_symbol_decode from GINETEX sheet. care_kit if care_products_ok is filled in.
Immediate escalation
care_damage_claim with photos requested then handoff. care_forbidden if customer insists after bot refusal. SKU without care metafield. Rare material undocumented (antique silk, exotic leather).
Escalate after 1 turn
Customer describes stain not listed in stain_protocol. Mixed care + warranty question (#62). Customer threatens chargeback or negative review.
Max turns rule
4-6 questions max for preventive care. Damage claim: photos mandatory before handoff, no additional care advice.
Tree documentation
Each CARE-BOT branch: intent, condition to respond, condition to escalate, equivalent macro #346, handoff payload.
Respond / escalate matrix
Notion table: Intent | Required metafield | Respond if | Escalate if. Example: care_stain | stain_protocol | stain type listed | unknown stain or damage. Each row validated by product manager before bot activation.
Which data sources should the care bot read?
The care bot only advises from verifiable sources, never from LLM memory alone.
Five Shopify sources
Metafield product.care_instructions: rich text or JSON protocol
Metafield product.care_forbidden: list of forbidden products/methods
Metafield product.stain_protocol: JSON by stain type
Metafield product.care_products_ok: compatible kit references
Help center /care: RAG chunks indexed by SKU
Shopify documents product metafields to enrich the catalog (Shopify, metafields 2025).
GINETEX symbols
Table of washing, bleaching, drying, and ironing pictograms. The bot translates symbols into customer instructions (GINETEX, symbols).
Indexed care video
60-second cleaning video transcript → care_how_clean RAG chunks. The bot cites timestamps or video links, never loose paraphrasing.
Warranty policy (#62)
The care bot reads the warranty policy for routing via care_vs_warranty, but does not decide on replacements alone. Handoff based on hypothesis defect/negligence.
Sync care sheet #346
Metafield = export of 6-field care sheet (#346). One single source, bot and agents aligned. See Shopify bot data.
How can I prevent the bot from inventing maintenance advice?
The anti-hallucination care protects the customer and the brand against dangerous advice.
Five strict rules
Product Whitelist: bot only mentions care_products_ok
Method Blacklist: bleach, solvent if care_forbidden
No "try this": documented protocol or escalation
No guaranteed result: old stain = managed expectations
Fallback: "undocumented protocol" + handoff
System care prompt
Instruction: "You never recommend a cleaning product outside of care_products_ok. You never invent a washing temperature. You escalate if care_damage_claim or undocumented material." See anti-hallucination.
Regression tests
25 scenarios: proto OK, wine stain, bleach requested on silk, damage claim, SKU without metafield, care_pro forbidden. Target care hallucination rate 0%.
Refusal of care_forbidden
Customer asks for bleach on leather: bot cites care_forbidden, proposes alternative CARE-PROTO, never acquiesces.
Audit logging
Each care_how_clean logs sku, source metafield, decision. Weekly review of refusal escalations to enrich care sheets.
Which bot flows for post-purchase care?
The care bot intervenes at multiple touchpoints post-delivery.
Proactive flow D+7
Email or bot message: "Here is how to maintain your [SKU]" with care_how_clean link. Reduces preventive care tickets by 20 to 35%.
Post-purchase widget flow
Order / customer account page: "Maintenance questions?" Pre-filled care_how_clean intent with order line SKU.
Incident stain flow
Customer "wine stain" → care_stain → stain_protocol JSON → numbered steps. If stain > 48 h: reframe expectations, no absolute promise.
Damage claim flow
"Product damaged after cleaning" → mandatory photos → collection of followed protocol → handoff_agent with payload CARE-DAMAGE (#346).
Cross-sell kit flow
After care_how_clean: "Recommended care kit: [link]" if care_products_ok. No cross-sell of untested products.
Care handoff payload
sku, intent, care_protocol_sent, photos[], stain_type, client_admits_wrong_product (bool), escalation_reason.
Example of validated bot response
"For your Model Oslo sofa in polyester fabric: vacuum dry, clean fresh stain with damp cloth and mild soap, air dry. Prohibited: bleach, machine wash, solvent. Complete guide: [link]. Care kit: [URL]."
Instagram / WhatsApp flow
Customer sends stain photo: bot asks for SKU or order_id, identifies stain_type if visible, applies stain_protocol or escalates. Limit advice if photo is blurry or product is unidentifiable.
How do I connect the care bot to Shopify?
The Shopify care bot setup relies on metafields aligned with sheet #346.
Technical Checklist
Create product.care_* metafields namespace
Populate care_instructions for the top 50 high-volume care SKUs
Index help center /maintenance in RAG by SKU
Configure care_* intents and Section 4 tree
Write anti-hallucination care prompt in Section 6
Test 25 regression scenarios
Activate proactive D+7 flow in Klaviyo or Flow
PDP and Account Widget
"Product Care" button with care_how_clean intent pre-filled with product_id.
Order Line Context
Bot reads order history if customer is logged in: purchased SKU → direct care metafield, fewer product errors.
Phased Rollout
Phase 1: top 20 SKUs representing 80% of care tickets. Phase 2: complete catalog. Phase 3: advanced care_stain if stain_protocol is enriched.
Shopify Flow Alerts
If care_escalation_rate > 50% on SKU: alert ops to enrich care_instructions metafield. If care_hallucination_incident is reported: temporarily disable intent.
Which KPIs should be measured on the customer care bot?
Without care bot KPIs, it is impossible to prove ROI and detect excessive escalations.
Seven Key Metrics
care_bot_resolution: resolved without agent / care bot tickets
care_escalation_rate: handoffs / care sessions
care_hallucination_incidents: incorrect advice reported / month
care_damage_handoff_time: claim → agent < SLA
care_kit_conversion_bot: kit purchases post-bot link
CSAT intent care_bot: satisfaction post-bot response
repeat_care_contact: customer returns for the same topic
DTC Benchmark
Target care_bot_resolution > 60%, care_escalation_rate < 40%, care_hallucination_incidents 0, CSAT care bot > 4.5/5.
Correlation #346
Compare overall care_deflection_rate (#346) vs care_bot_resolution. Bot must increase deflection without an increase in care_damage_claim_rate.
Monthly Review
Top 5 escalation_reason: metafield missing, stain unknown, damage claim. Enrich care sheets based on the dominant cause.
A/B Test Flow D+7
Measure the impact of proactive message care_how_clean on care_tickets and repeat_care_contact. Often -25% preventive care tickets at D+30.
What edge cases and escalations should be anticipated?
Five care bot edge cases require escalation, not LLM improvisation.
Customer has already used bleach
Damage is done. Bot does not advise "rinse and repeat" without a protocol. Immediate damage claim escalation + photos.
Old stain (> 72 h)
Bot applies stain_protocol with a disclaimer of reduced effectiveness. Do not promise removal. Escalate if customer demands replacement.
Undocumented material
Vintage silk, exotic leather: product expert handoff, no generic "damp cloth" advice.
Mixed care + warranty
"Damaged AND under warranty": bot collects photos + purchase date, warranty handoff (#62) with care payload.
Professional cleaning prohibited
care_pro_cleaning = no on file: bot refuses, cites policy #346. Escalate if customer disputes.
Allergic to care kit product
Safety issue: escalate if skin reaction is reported. Bot does not recommend untested alternatives.
Customer insists after bot refusal
Second request for bleach on delicate textile: immediate handoff, do not restart protocol. Tag care_forbidden_escalated for training audit.
How does Qstomy handle product care advice?
Qstomy processes care_* intents from metafields care_instructions, stain_protocol and policy #346.
Bot care capabilities
care_how_clean: SKU protocol from metafield
care_stain: stain_protocol branch by type
care_forbidden: method refusal + alternative
care_damage_claim: photos + structured handoff
Smart escalation: automatic Section 4 rules
Encrypted DTC Scenario
Leather goods brand, 72 care tickets/month, heterogeneous agent responses.
After Qstomy care bot + metafields: 67% care tickets auto-resolved, care_escalation_rate 28%, care_hallucination_incidents 0, care bot CSAT 4.8/5, care_kit_conversion +22%.
The bot cites official protocol; the agent handles damage disputes with pre-collected photos and care history.
Explore AI support, Shopify, request a demo.
What is the checklist for launching the care bot?
Care bot checklist (10 steps)
Audit care tickets 90 days (#346)
Populate care metafields for top 50 SKUs
Document answer vs. escalate tree in section 4
Draft care anti-hallucination prompt in section 6
Configure care_* intents
Index help center / RAG maintenance
Test 25 regression scenarios
Activate D+7 proactive flow
Weekly care_bot_resolution dashboard
Monthly escalation_reason review and metafields enrichment
In short
#347 = AI care bot, not policy ops (#346)
Answer if protocol is documented, escalate if damage or doubt
10 care_* intents: clean, stain, damage, warranty
Anti-hallucination: whitelist care_products_ok
care_bot_resolution KPI: target > 60%
FAQ
Difference with care support #346?
#346 = team, macros, policy. #347 = AI bot for when to answer and when to escalate.
Can the bot handle a damage claim alone?
No. Collects photos + agent handoff with payload. No replacement decision.
Is a metafield required per SKU?
Yes for grounded care_how_clean. Without metafield: escalate, no improvisation.
Link to lifespan #345?
#345 = how long. #347 = bot maintenance to achieve it.
Bot or human for complex stain?
Bot if stain_protocol is documented. Human if stain is unknown or damage claim.
Going further
Test five care scenarios on staging: standard protocol, stain, bleach forbidden, damage claim, SKU without metafield.
Share this #347 guide with product and ops: a care bot without care_instructions metafields = maximum hallucination risk.
Sync care_instructions with each material change: a stale protocol = useless escalations and declining care bot CSAT.
Enzo
July 15, 2026
