E-commerce
June 26, 2026
Automating customer service without a structure often amounts to putting a chatbot on top of an already chaotic inbox. Responses go out faster, but categories remain unclear, escalations happen too late, and the root causes are not addressed.
The right promise is simpler: first organize requests, understand volumes, write the rules, and then automate repetitive and documented cases.
This article #25 is different from other content about customer service automation. It lays down the e-commerce customer service strategy before AI: taxonomy, ownership, SLA, channels, SOP, self-service, and the Qstomy pilot.
Summary
Why structure customer service before AI?
Automation amplifies what already exists. If your policies contradict each other, if your tags are unusable, or if no one knows when to escalate, AI does not fix the problem: it makes it more visible.
Clarity: knowing which patterns genuinely fill the support inbox
Priority: handling urgent or sensitive requests first
Quality: responding with the same rules, regardless of the channel
ROI: automating repetitive requests, not exceptions
Carti points out that a good Shopify support workflow must distinguish between instant responses, context-rich escalations, and cases that require human judgment (Carti, customer service workflow).
How to audit 90 days of requests?
Start with facts, not with intuition.
Export tickets, emails, chats, Instagram DMs, WhatsApp, and marketplace requests
Add channel, date, status, first message, first response time
Group by motive: WISMO, return, refund, size, payment, product
Calculate volume, resolution time, and customer risk
Identify repetitive and documentable themes
If you don't have a helpdesk yet, a spreadsheet is enough. The goal is to know which 5 to 8 categories represent the majority of the volume and which ones cost the most agent time.
Add a "probable cause" column. A WISMO ticket can come from a carrier delay, a tracking email not received, or a tracking page that is too hidden. The customer service strategy must not only classify symptoms; it must indicate what to fix upstream.
What customer service taxonomy should be created?
A useful taxonomy must be simple enough to be used every day, but precise enough to drive decisions.
Recommended Structure
Level 1: domain. Example: Delivery, Return, Product, Payment, Account, Dispute. Level 2: sub-theme. Example: Delivery > Delay, Return > Label, Product > Defective.
Practical Rules
Limit: 30 to 50 active tags are often enough
Definition: one clear sentence per tag
Usage: a tag must trigger reporting, an owner, or an action
Review: monthly cleaning at first, then quarterly
Yektoo recommends tagging tickets first to identify top intents before changing tools or adding automation (Yektoo, support workflows).
Example: "Return" alone is too broad. "Return > label", "Return > refund delay" and "Return > worn item refused" allow you to create macros, measure pain points, and know what can go into self-service.
How do I assign owners and routing?
Each category must have an owner. Otherwise, the ticket always goes back to the most available person, often the founder.
Front-end support: WISMO, FAQ, simple returns, customer accounts
Ops: lost parcel, picking error, damage, carrier
Product: recurring defect, unclear product sheet, sizing
Finance: complex refund, chargeback, invoice
Manager: VIP, legal threat, sensitive public review
Create a "who handles what" page: category, owner, SLA, macro, escalation. It is better than an oral rule that nobody applies during peak periods.
What SLAs should you set without overpromising?
A SLA is not a marketing catchphrase. It is an operating rule that the team can actually meet.
Chat: quick response if the channel is open, otherwise clear hours
Email: realistic first response, for example same business day or 24 hours
VIP: stricter priority if justified by customer value
Dispute: rapid human escalation
Resolution: resolution time handled separately from the first response
Gorgias allows you to create SLAs by channel, tag, or ticket field, and then track first response and resolution in an SLA report (Gorgias, SLA policies). Start simple: first response, resolution, escalation.
Do not copy a benchmark if your team cannot meet it. An internal SLA of 24 hours met 95% of the time is better than a 4-hour promise met only half the time. Then, reduce the timeframe once your routing and macros handle the workload.
How to centralize the channels?
A customer service strategy cannot ignore the channels where customers actually write.
Email: history and detailed requests
On-site Chat: pre-purchase, checkout, mild emergency
Instagram DM: quick questions and near-public complaints
WhatsApp: mobile-first markets, follow-up and relationship building
Marketplace: SLAs imposed by the platform
Centralizing does not mean responding everywhere with the same tone. It means a single source of truth: same policies, same order history, same escalation rules.
Which SOPs should be documented first?
Before automating, write down the procedures that the team is already more or less applying manually.
WISMO: where to find tracking, when to reassure, when to escalate
Simple return: conditions, portal link, refund timeframe
Defective product: photos requested, replacement, refund
Order modification: before or after shipping
Refund: amount, timeframe, payment method, exception
ScaleOps emphasizes SOPs for Shopify: without written procedures, it is impossible to delegate properly, and even less to automate (ScaleOps, Shopify support SOP). Each SOP must state when to use it, when not to use it, and when to hand over to a senior human.
A useful SOP often fits on one page: context, eligibility criteria, agent steps, customer message, Shopify link to check, escalation cases. If it requires ten minutes of reading, it will not be used during a peak in orders.
What self-service content should you create?
Self-service is often more cost-effective than a sophisticated first bot.
FAQs: delivery, returns, payment, account, products
Order tracking: status, carrier, estimated time
Returns portal: conditions and steps without email
Notifications: shipped, delayed, delivered, return received
Macros: validated responses for agents and future bot
Talk Shop advises starting with the most frequent categories and tracking FRT, FCR, and CSAT rather than over-optimizing handling time too early (Talk Shop, customer service ecommerce).
What must remain human?
A good customer service strategy also defines the limits of automation.
Disputes: angry customer, legal threat, public review
Chargebacks: financial risk and evidence to prepare
Safety: regulated product, allergy, sensitive data
VIP: high LTV, influencer, large order
Exceptions: commercial gesture, out-of-policy refund
Atlas Media Group recommends leaving AI on tier 1 and routing complaints, disputes, and emotional signals to a human with full context (Atlas, ecommerce customer service strategy).
What roadmap should be followed over 90 days?
Weeks 1-2: visibility
90-day audit, top intents, channels, response times, initial categories.
Weeks 3-5: structure
Taxonomy V1, owners, internal SLAs, macros for the top ten issues.
Weeks 6-8: self-service
FAQ, order tracking, returns portal, notifications, better-routed contact page.
Weeks 9-12: AI pilot
Automate 3 to 5 repetitive, low-risk intents: WISMO, simple return, delivery FAQ, refund status, simple product questions.
Maintain a monitoring period. Read resolved conversations, escalations, and bot rejections. This quality control is used to correct content before scaling up automation.
How does Qstomy fit into this strategy?
Qstomy works best when your requests are already organized. It can integrate your taxonomy, validated contents, and handoff rules.
DTC Shopify Scenario
Store with 1,100 tickets/month. Audit: 28% WISMO, 16% returns, 12% product questions, 8% refunds. After taxonomy and macros, Qstomy is launched on five documented intents. Pilot hypothesis: 420 conversations resolved without human intervention, automatic handoff on disputes and VIPs, stable CSAT, 25% fewer WISMO tickets in 45 days.
The bot then becomes a conversational layer of your customer support strategy, not a shortcut to avoid structuring the team.
See AI customer support, human handoff, chatbot KPIs and Shopify integration.
Which playbooks should be applied this week?
Playbook 1: Useful tags
Take 100 recent tickets. Create 8 to 12 tags maximum. If a tag does not change reporting, routing, or macro, delete it.
Test these tags for a week with the team before freezing them in your helpdesk.
Playbook 2: Owner matrix
For each main category, write: owner, target resolution time, macro, escalation, ticket example.
Playbook 3: Internal SLA
Set three rules: email response, chat response, dispute escalation. Measure for two weeks before making public promises.
Playbook 4: AI pilot
Choose a single, well-documented intent. Launch Qstomy in monitoring mode, compare resolution, escalations, and CSAT.
Useful links
Automation: automate customer service
FAQ: e-commerce FAQ page
Post-purchase: automated post-purchase support
Enzo
June 26, 2026
