E-commerce
June 28, 2026
Your conversations dashboard shows 847 tickets this week. No one opens it on Friday night. Meanwhile, 23 customers write "open package", "weird smell", "grayed out pay button": a defective batch, a Safari checkout bug, a misconfigured promo.
SentiSum reminds us that most customer issues do not break out into a visible incident: they emerge through small variations in volume, sentiment, and vocabulary, easy to miss without proactive alerts in Slack (SentiSum, Slack silent signals 2026).
This guide #172 covers internal alerts triggered by conversational signals: rules, team routing, thresholds. Distinct from sentiment analysis (#139) (reading emotions) and conversational merchandising (#108) (catalog actions): here, it is about notifying the right teams at the right time.
Summary
Why turn conversations into internal alerts?
Support conversations are the primary sensor for friction in e-commerce. Without internal alerts, they remain a backlog consulted only after the fact.
The Limitation of the Passive Dashboard
Insight7 notes that thousands of conversations contain signals of risk, escalation, or opportunity, but without alert workflows, these signals get bogged down (Insight7, alert workflows 2026). A dashboard requires a human to go look at it. A Slack alert pushes information to where decisions are made.
What an Internal Alert Triggers
Logistic Ops: WISMO spike on carrier X
Product: 5 "defective batch" tickets within 2 hours
Tech: "pay button" keyword ×10 vs baseline
Marketing: launch promo code confusion
Support Lead: VIP + 3rd ticket in 7 days
Difference with Ticket Routing
Routing classifies an individual ticket into a queue. An internal alert signals a cross-cutting pattern across multiple tickets or conversations, before the volume becomes a crisis.
How does it differ from sentiment and merchandising content?
Three uses of conversations, three destinations.
#139 Sentiment
Sentiment analysis (#139) prioritizes the queue and reads emotional trends. Internal alerts #172 trigger a non-support team action (product, ops, dev).
#108 Merchandising
Merchandising (#108) transforms questions into collections and PDP enrichment. Alerts #172 also cover bugs, promos, logistics, compliance: any signal requiring a fast internal response.
#156 Klaviyo
Klaviyo segments (#156) feed customer marketing. Here: notify internal teams, not enrich a CRM profile.
Complementary Information
What types of internal alerts can be created from conversations?
Five families of e-commerce conversational alerts cover 90% of DTC needs.
1. Spike volume intent
Intent delivery_delay goes from 8% to 22% of tickets in 24 hours. Threshold: +10 points vs 7-day average or ×2 absolute. Route: #ops-logistics.
2. Critical keyword
Appearance of "scam", "dangerous", "food poisoning", "chargeback", "button not working". Threshold: ≥3 occurrences / hour. Route: #support-leads or #tech-front.
3. Product / batch anomaly
Same SKU + same complaint ("smell", "different color", "missing part") ≥5 tickets / 48 hours. Route: #product + #quality.
4. Customer repetition
Kustomer Signals detects the customer returning to the same subject for the 2nd or 3rd time (Kustomer, Signals 2026). Alert manager if LTV > threshold or threat of public review.
5. Agent or bot quality threshold
Bot handoff rate > 40% on a given intent, or CSAT < 3 on D-day promo macro. Route: #support-qa.
How do you build a library of signals to monitor?
The conversational signals library is your reference repository before any technical rules.
3-step method
Export 90 days of tickets + bot transcripts: top 30 intents and top 50 verbatim expressions
90 min workshop support + ops + product: "which signals should have alerted us sooner?"
Notion Document: signal | threshold | Slack channel | owner | action type
DTC E-commerce universal signals
LOG-*: carrier delay, damaged package, incorrect address
PRD-*: quality defect, suspicious batch, missing instructions
CHK-*: payment, promo, mobile checkout
PRM-*: promo code, sales, inconsistent marketing message
REG-*: allergen, ingredient, health claim (regulated vertical)
MVP Prioritization
Week 1: 5 spike intent alerts (WISMO, return, promo, checkout, quality). Week 2: 10 critical keywords. Week 3: repeated SKU alerts. Align codes with ticket taxonomy (#135).
How to set thresholds and rules without alert fatigue?
A poorly calibrated alert ends up on Slack mute. Conversation alert thresholds must be actionable.
Practical threshold formulas
Relative spike: intent > 7-day average + 2 standard deviations
Absolute spike: ≥10 tickets same intent / 4 h (adjust to volume)
Keyword: ≥3 mentions / hour OR ≥1 if VIP LTV > €500
SKU cluster: ≥5 tickets same SKU + same quality tag / 48 h
Unmatched bot: same unmatched question ≥15× / 24 h
Anti-noise safeguards
2-hour cooldown window: a spike alert does not trigger again until acknowledged. Active hours: CHK-* alerts 7 days a week; merch_compare alerts during business hours only. Minimum volume: no spike alert if intent baseline < 5 tickets/day (statistical noise).
Pre-production testing
Replay 30 historical days: how many alerts would have fired? Target: 2 to 8 actionable alerts / week, not 40 ignored notifications. Quantum Metric recommends linking each alert to an analyzable segment to avoid dead ends (Quantum Metric, experience alerts).
How to route alerts to the right Slack teams?
The e-commerce Slack alert routing follows a signal → channel → owner matrix.
DTC stack type channels
#cx-alerts: volume spikes, CSAT, VIP escalations
#ops-logistique: WISMO, carrier, damaged package
#produit-qualité: batch defect, quality returns, instructions
#tech-front: checkout, promo bug, site error
#marketing-comms: promo confusion, ads vs site messaging
Alert message format
Title: [SPIKE ALERT] Intent delivery_delay +14 pts. Body: volume 34 tickets / 24 hrs vs baseline 12, top carrier mentioned, 2 anonymized verbatims, link to filtered Gorgias dashboard, "Acknowledged by @owner" button. Delighted shows the benefit of pushing feedback directly into Slack for team visibility (Delighted, Slack integration).
Escalation if not acknowledged
No acknowledgment within 30 min (CHK-*, REG-*) or 2 hrs (LOG-*): ping @channel-leads. Weekend: senior support on-call on #cx-alerts only.
How do you connect Gorgias, the bot, and Slack technically?
Three architectures to connect conversations and internal alerts.
Option A: native helpdesk rules
Gorgias Rule: IF tag intent = checkout_bug AND count last 4 h > 5 THEN HTTP POST Slack webhook. Limit: few complex rules, no native multi-ticket correlation.
Option B: middleware (Zapier, Make, n8n)
Trigger: new tagged ticket or end of bot conversation. Action: increment Redis/Airtable counter, IF threshold THEN Slack. Target latency < 2 min. Ideal MVP for 5 to 15 alerts.
Option C: VoC platform (SentiSum, Enterpret)
Unified ingestion of tickets + surveys + reviews, ML anomaly detection, Slack push (Enterpret, 2026 feedback integrations). Investment justified starting from ~2,000 tickets/month or complex multi-channel stack.
Qstomy bot / helpdesk pipeline
End of conversation webhook: payload intent, SKU, URL, sentiment, summarized transcript. Worker evaluates rules from library section 4. Log each fired alert in audit table (date, rule, acknowledgment).
Which ops scenarios should be triggered on a daily basis?
Eight ops alert scenarios to set up as a priority on a Shopify DTC store.
Logistics and fulfillment
Carrier delay: WISMO spike + carrier name in verbatim
Damaged parcel: ≥5 tickets / 24 h same intent
Post-shipping address error: cluster before ops cutoff
Product and quality
Suspect batch: same SKU + "smell / color / broken"
Instructions or allergen: key-word REG-* vertical food/beauty
Tech and promo
Mobile checkout: "greyed out pay", "Safari", "Apple Pay"
Launch promo: code + "not working" spike D0-D+2
Support management
VIP + negative sentiment + 2nd ticket 7 d → lead alert. Completes post-redesign support (#171) and flash sale support (#169).
How do you measure the effectiveness of the alert system?
Conversational alert KPIs prove that the system prevents crises, not that it spams Slack.
Leading KPIs
Time to acknowledge: alert → acknowledgment delay (target < 30 min critical)
Time to action: acknowledgment → documented action (patch, email, ops)
False positive rate: non-actionable alerts / total (target < 25%)
Alerts / week: stable 2 to 8, not infinite growth
Lagging KPIs
Incidents prevented: checkout bug detected before loss of conversion (cross-reference GA4). Post-alert ticket volume: decrease in relevant intent within 72 hours post-fix. Quality batch cost: early recall vs public negative reviews. CSAT intent spike: return to baseline within 7 days.
30-minute monthly review
Top 5 alerts triggered, top 3 actions initiated, 2 rules to adjust or delete. See products generating tickets, Data & Analytics Qstomy.
What mistakes sabotage internal alerts?
Five conversational alert anti-patterns kill adoption in a matter of weeks.
Frequent Mistakes
Too many alerts: low thresholds, saturated single channel
No owner: Slack message with no designated owner
Alert without context: "spike detected" without verbatim or ticket links
No closed loop: alert acknowledged, never a measured fix
Channel silos: Instagram DM and chatbot excluded from monitoring
Quick Fix
Cut 50% of the least actioned rules after 30 days. Add a mandatory "action taken" field to acknowledgment. Centralize all sources into a single ingestion before rules (SleekFlow insists: insights come from all conversations, not surveys alone (SleekFlow, CX intelligence 2026)).
How does Qstomy trigger internal alerts from conversations?
Qstomy detects conversational signals and pushes configurable alerts to Slack, email, or webhook.
Alerting Features
Intent + keyword rules: editable no-code library
Spike detection: comparison against a 7-day rolling baseline
SKU Cluster: grouping of product complaints
Multi-channel routing: Slack, email, Jira webhook
Enriched payload: verbatims, URL, LTV, summarized transcript
Audit trail: history of alerts and acknowledgements
Quantified DTC Scenario
DTC food brand, 1,900 conversations/month, Qstomy alerts on REG-allergen and WISMO spike. Week 3: cluster alert "bizarre smell" on granola batch SKU-442, 7 tickets / 36 hours. Product flagged within 20 mins, batch recalled on Day+1. Without an alert: estimated 45 quality tickets + 12 avoided 1★ reviews. MTTA for critical alerts 18 mins, false positives 11%, quality intent volume post-recall -62% within 5 days.
Explore AI support, Shopify, request a demo.
Which operational playbooks should be used to launch alerts this week?
Playbook 1: Signals Library (2 h)
90-day ticket export. Support + Ops workshop: list 15 signals, thresholds, Slack channel, owner. Shared Notion document.
Playbook 2: 5 MVP Alerts (3 h)
Configure WISMO spike, checkout keyword, SKU quality cluster, D-day promo, VIP repetition. Test with historical tickets.
Playbook 3: Slack Channels (30 min)
Create #cx-alerts, #ops-logistics, #product-quality, #tech-front. Post routing matrix. Appoint channel owners.
Playbook 4: Message Format + Acknowledgment (1 h)
Slack template section 6. Mandatory "action taken" field upon acknowledgment. Auto-escalation if silent for 30 min.
Playbook 5: W+4 Review (45 min)
Count fired alerts, false positives, triggered actions. Mute 2 noisy rules, add 1 missed signal.
Playbook 6: Monthly Debrief
KPI section 9, 1 documented prevented incident, updated signals library.
Useful Linking
Customer conversations speak before your dashboards. Well-calibrated internal alerts turn this whisper into action before the crisis hits.
Enzo
June 28, 2026
