status: complete audience: both chapter: 06 last_updated: 2026-04 contributors: [alexwill87, claude-cockpit] lang: en

03 -- E-commerce

For: Online seller managing a shop (Shopify, WooCommerce, PrestaShop or equivalent) Setup time: 3 to 5 days Difficulty: Intermediate

Context

An online seller manages a shop with a few hundred product references. They sell through their own shop and potentially on marketplaces. They work alone or with a part-time assistant. Their daily routine: processing orders, answering customer questions, monitoring stock, and making replenishment decisions.

They're looking for a system that automates repetitive tasks without depending on an additional SaaS costing 200 EUR/month.

Problem

• Customer questions arrive via email, contact forms, and marketplaces -- slow or forgotten responses
• Stock tracking is manual: stockouts discovered too late, overstocking on other products
• Sales reports are generated by hand in spreadsheets
• No proactive alerts on trends (product taking off, product stagnating)
• Time spent on operations prevents work on strategy

Configuration

Infrastructure

Agents

Customer Support Agent: - Reads incoming customer messages (via API or export) - Classifies by category: order tracking / product question / complaint / other - Generates an appropriate response draft - Escalates complaints to the seller with a summary

Stock Agent: - Queries the shop API daily - Alerts when a product falls below the reorder threshold - Detects slow-moving products (no sales for X days) - Generates a supplier order suggestion

Reporting Agent: - Generates a daily report: sales, average cart value, best-selling products - Generates a weekly report with trends and comparison to previous week - Flags anomalies (sudden sales drop, unusual spike)

Setup

Day 1-2: Infrastructure and shop connection

1. Provision the VPS and install OpenClaw
2. Configure API access to the e-commerce platform
3. Test reading orders and stock via the API
4. Create the file structure for response templates

Day 3: Customer Support Agent

1. Write the system prompt with the shop's tone and common FAQs
2. Configure reading of incoming messages
3. Define classification rules (tracking / question / complaint)
4. Test on 20 real messages
5. Adjust the prompt based on results

Day 4: Stock and Reporting Agent

1. Configure reorder thresholds by product category
2. Set up daily stock alerts
3. Configure daily sales report
4. Test the weekly report on previous week's data

Day 5: Stabilization

1. Configure notifications (email or webhook)
2. Test the complete flow over a real day
3. Adjust thresholds and templates based on feedback
4. Document procedures for a potential assistant

Result

After two weeks of use:

• Customer response time divided by 3: response drafts are ready in seconds. The seller reviews, adjusts if necessary, and sends
• Zero surprise stockouts: alerts arrive 5 to 7 days before stockout, leaving time to order
• Automated reporting: the daily report arrives each morning. No more spreadsheet manipulation
• Trend detection: the agent flagged a product whose sales had tripled in a week, enabling anticipated replenishment
• 2 hours freed per day: time recovered from support and reporting is reinvested in product strategy

Lessons Learned

1. FAQs cover 80% of questions. Invest time writing 20 to 30 standard responses in the system prompt to eliminate most support work.

2. Stock thresholds must be by category, not universal. A fast-moving product needs a higher threshold than a niche product.

3. The daily report must be short. 5 lines maximum. If the report is too long, it won't be read. Details should be accessible on demand.

4. Don't automate complaints. A poorly handled complaint costs a customer. The agent summarizes and escalates, the seller responds personally.

5. Test on real data, not fictional data. Edge cases (stockouts, canceled orders, returns) only reveal themselves with production data.

Common Mistakes

Template -- Customer support agent system prompt

You are the support assistant for [SHOP NAME].

Context:
- Online shop selling [PRODUCT TYPE]
- Clientele primarily [COUNTRY/REGION]
- Tone: professional, warm, concise

Rules:
- You generate response drafts, never final responses
- Always use formal address with customers
- Never promise refunds or goodwill gestures without seller validation
- Escalate any complaint with a factual summary
- Include the order number in each response

Message classification:
1. Order tracking -> check status, respond with tracking
2. Product question -> answer from product sheet and FAQs
3. Complaint -> summarize the problem, escalate to seller
4. Other -> classify and flag

FAQs:
- Delivery time: [X business days]
- Return policy: [conditions]
- Payment methods: [list]

Verification

• [ ] The shop's API is accessible from the VPS
• [ ] The support agent correctly classifies a sample of 20 messages
• [ ] Stock alerts trigger at the right threshold on a test product
• [ ] The daily report is generated and sent each morning
• [ ] Complaints are escalated and not auto-responded to
• [ ] Notifications arrive on the chosen channel (email or webhook)

An online seller doesn't need a support team. They need a well-configured agent and human review.