E-commerce
June 28, 2026
Without structured tags, your helpdesk counts tickets but does not reveal customer friction points: false delivery promises, missing size guides, forgotten tracking. The volume of "returns" or "parcels" masks the actionable causes.
This guide describes an e-commerce support tagging method: friction-oriented taxonomy, agent workflow, quality control, trend reading, and loops back to content, bot, and ops. It differs from general feedback analysis and conversational analytics: here, classification is done ticket by ticket.
You will leave with a 3-level grid, helpdesk rules, and playbooks to transform tags into actions within 14 days.
Summary
Why tag support conversations in e-commerce?
Counting tickets is not enough. Two "late package" threads can hide different pain points: a carrier on strike, tracking that is never updated, or a 24-hour delivery promise on the product sheet. Without structured tagging, you treat the symptoms without seeing the causes.
Pain point vs. reason for contact
A reason describes what the client is asking for ("order tracking"). A pain point reveals the friction experienced ("tracking missing 5 days after announced shipment"). The taxonomy must capture both levels.
Concrete example
Fashion support 2,400 tickets/month, dominant "return" tag. Sub-tag "incorrect size, missing guide" = 18% of returns on a collection. Action: PDP size guide + bot macro. Linked return tickets: −34% in 8 weeks.
Who benefits
Management: priorities without guesswork
Product: targeted sheets and guides
Ops: visible carriers and packaging
Marketing: ad promises aligned with real pain points
Minimum volume
From 300 tagged tickets/month, trends by family become clear. Below that, start with only 5 families.
Convert over 2,000 customers on average per month with Qstomy.
The world’s 1st Shopify AI dedicated to customer conversion
Empowering 200+ e-commerce merchants
How to differentiate tagging support, NPS, and product reviews?
Do not duplicate four inconsistent classification systems.
Tagging conversations
Granular, operational, per ticket. Source: helpdesk, chat, escalated bot. Audience: support, product, ops.
NPS and CSAT
Aggregate post-experience score. A low NPS on "delivery" validates a spike in WISMO tags. See NPS e-commerce timing.
Public product reviews
Conversion and SEO signal. Cross-reference 1-2 star review themes with internal detractor tags.
Product insights
The article on support → product conversations addresses the SKU improvement backlog. Here: the tag grid that feeds these insights. See product insights from support.
Golden rule
One official support taxonomy. Other channels map to it ("size return review" = return tag + missing_size_guide).
How to build an irritant-oriented taxonomy?
Pattern Owl recommends a 3-level hierarchy: domain, theme, sub-theme, with 30 to 50 active themes in total. Fewer than 20 loses nuance; more than 60 creates overlaps (Pattern Owl, feedback taxonomy 2026).
Level 1: domains (10 to 15 max)
wismo: tracking, delay, lost package
product_prepurchase: size, stock, compatibility
order: modify, cancel, promo
return_exchange: timeframe, fees, label
payment: decline, refund, invoice
delivery: address, pickup point, damage
after_sales_product: defect, warranty, manual
account: login, data, GDPR
marketplace: Amazon, platform dispute
Level 2: pain point sub-tags
Wismo example: missing_tracking, blocked_scan, delay_exceeded_promise, package_not_received. Each sub-tag = possible action. Name in customer language: "runs small" rather than "negative fit discrepancy".
Cross-cutting tags
sentiment: neutral, frustrated, review_threat
sales_channel: shopify, amazon
resolved_1st_contact: yes/no
Avoiding expansion
New tag only if more than 20 unclassifiable tickets/month. Notion document: definition, 3 verbatim examples, counter-examples, action owner.
How to co-build the taxonomy in one day?
A taxonomy imposed without agents fails. Co-build it in a workshop.
Participants
Support lead, 2 senior agents, e-commerce, product or ops, optional data.
Morning: verbatim immersion
Export 200 random tickets over 30 days
Everyone reads 40 tickets, noting frustrating customer phrases
Clustering: group similar formulations (Miro or post-its)
Name clusters in neutral internal language
Afternoon: grid validation
Validate 10-15 level 1 families
Define 3-5 sub-tags per top-volume family
Write 2 lines of definition + 1 example per tag
Test on 30 tickets: two agents tag, measure agreement
Agreement threshold
Aim for over 85% agreement on the main tag after training. Below that, simplify the grid.
Post-launch iteration
Week 2: 30 min "confusing tags" stand-up. Week 4: merge twin tags. Invite logistics if 20% of tags point to packaging or wrong SKU prepared.
Which agent workflow to tag in less than 10 seconds?
Tagging must be quick or it will not be done.
Timing of the tag
At the closing of the ticket, not at the opening (intent can change). Mandatory before resolved status. Helpdesk blocks closing without a main tag.
Helpdesk fields
Main tag mandatory, sub-tag recommended, sentiment if escalated. Favorite drop-down list: 5 tags pinned per agent.
Pre-tag automation rules
Keyword "follow-up" + order shipped: pre-tag wismo, agent confirms sub-tag. Macro "WISMO tracking sent" automatically applies tags wismo + tracking_communique. See prioritize support requests.
Multi-topic tickets
Tag = main topic + secondary tag if two intents are equal. Maximum 3 tags. Marketplace: marketplace tag + platform. See centralize marketplace messages.
Bot to human handoff
Escalated ticket inherits bot tags; agent corrects if intent is poorly detected. Correction log feeds AI training.
How to combine manual tagging, rules, and AI?
Three complementary, not exclusive, modes.
Agent manual
Reference for nuances, disputes, and atypical cases. Indispensable for legal threats or GDPR.
Keyword rules
Inexpensive, rigid. Refine with AND/OR: return + label, not "return" alone.
Helpdesk classification AI
Gorgias detects intent and sentiment (Return/Status, Shipping/Delivery-Issue, etc.); WHEN/IF/THEN rules apply tags and actions (eesel AI, tagging ecommerce 2026). Zendesk Intelligent Triage classifies topic, sentiment, language, entities.
Recommended hybrid
AI suggests, human validates for 90 days. Then auto-tag if confidence is greater than 90% on stable categories (simple wismo). AI is profitable starting from ~1,500 historically tagged tickets/month.
Pre-purchase bot
The chatbot logs intent before escalation. See helpdesk vs chatbot vs KB.
How to maintain tag quality over time?
An unmaintained taxonomy becomes noise within six months.
Weekly Audit
Lead support reviews 20 closed tickets: correct tag? relevant sub-tag? Individual feedback, not punishment.
Inter-agent Agreement
Monthly: two agents re-tag the same sample. Agreement below 80%: 30-minute recalibration session on confusing tags.
Tag Drift
Symptom: "miscellaneous" or "other" rises to the top. Cause: new product, campaign, carrier. Add a sub-tag or automation rule.
Agent Onboarding
Day 1: read taxonomy. Week 1: tag with a mentor. Quiz on 10 dummy tickets before going solo.
Changelog
Support Slack: "New sub-tag promo_not_applied since Monday, here is the definition". Quarterly calibration: review definitions of the 5 most used tags.
How to read tags to identify the real pain points?
Numbers become actionable with the right interpretations.
Volume vs Growth
Volume: wismo is often number 1, which is normal
Growth: +40% month-over-month on promise_delay_exceeded = ops alert
Handling time: slow tag = broken process
CSAT by tag: poorly managed pain point despite macro
Repeat contact: customer returns within 48 hours
Impact / effort matrix
X-axis: tag ticket volume. Y-axis: revenue impact (average basket, return rate). Prioritize high-volume + high-impact quadrant.
Associated verbatim
Every week, read 5 raw verbatims from the number 1 growing tag. Numbers without customer words lead in the wrong direction.
Cross-referencing SKU and seasonality
Tag size_guide + jeans collection + 40% tickets = targeted product action. Compare BFCM vs off-peak tags to distinguish structural pain points from logistical peaks.
Automatic alert
Slack notification if tag promise_delay_exceeded is +50% vs previous week.
How do you turn an annoying tag into concrete action?
An untagged pain point without an owner is a decorative statistic.
4-step action loop
Tag exceeds threshold (50 tickets/week or +25% vs baseline)
Owner assigned (product, ops, CX)
Action documented with date
Measure tag volume 30 days after
Examples per pain point
tracking_manquant : shipping webhook + WISMO bot
guide_taille : PDP table + size quiz
promo_non_appliquee : clarify checkout + macro
emballage_casse : packaging ops + proactive support
Content and bot link
Top 5 tags: 5 priority knowledge base articles. Bot intents aligned with tags. See reducing WISMO Shopify and segmenting by funnel.
Avoid over-correcting
A tag spike 1 week post-SKU launch does not equal a return policy redesign. Wait for 2 stable weeks.
Which dashboard and which weekly review ritual?
The tags dashboard must fit on a single screen.
Essential widgets
Top 10 tags of the week
Top 5 growing tags
CSAT by main tag
Resolution time by tag
Share of other / untagged (target below 5%)
Tools
Gorgias Analytics, Zendesk Explore, Looker, BigQuery export. Gorgias displays tag reports by frequency and evolution (Gorgias, tag insights). Data analytics page.
30 min weekly review
Support lead + e-commerce: top 3 tags, 1 action decided, owner assigned. 5-line Slack minutes.
Tagging KPI
Tickets tagged upon closing above 95%, audit agreement above 85%, average tag time below 10 s. See KPI chatbot.
How does Qstomy unify chat and helpdesk tags?
Qstomy automatically classifies pre-purchase and post-purchase conversational intents. After escalation, the helpdesk ticket inherits the same tag vocabulary as the bot.
Cross-channel view
Unresolved bot intent "delivery time": escalation tagged delivery + delay_exceeded_promise. Analytics aggregates chat and email under the same pain point, without counting the same customer twice.
Quantified DTC Scenario
Cosmetics brand, 1,800 tickets/month, helpdesk tags only, 22% classified as "other". After V1 taxonomy + Qstomy on chat with aligned intents: "other" drops to 6%, early detection of the guide_usage tag (+62% within 3 weeks post-launch of a serum), product page correction on D+5, related tickets down 41% in 6 weeks.
Explore Shopify integration, customer support and request a demo.
Which playbooks should be launched this week?
Playbook 1: export 200 tickets
Pull 200 random tickets over 30 days. Each senior agent rates 10 irritating customer phrases. Group them into clusters. This is the basis of your level 2.
Playbook 2: V1 taxonomy in 10 families
Validate 10 level 1 domains, 3 sub-tags per top 3 current volume families. Publish a one-page PDF + Notion with definitions.
Playbook 3: blocked closing without tag
Configure mandatory main tag field upon resolution. Favorites per agent. Measure % tagged over 7 days: target 95%.
Playbook 4: WISMO pre-tag rule
Automation: tracked + shipped → WISMO pre-tag. Agent confirms sub-tag in 1 click. Linked macro applies consistent tags.
Playbook 5: Monday 30-min review
Dashboard: top 3 tags, top 1 growth, 5 verbatims. Decide 1 action owner + date. Measure irritating tag volume at D+30.
Useful links
Global Analytics : conversations analytics
Automation : support automation errors
NPS : e-commerce NPS
Product Insights : support product insights
Enzo
June 28, 2026
