Best AI Auto-Reply for Buyer Messages: Feishu + DeepSeek + Qianniu Full-Chain Automation
Build a completely free AI auto-reply system from scratch — handle buyer inquiries 24/7 and reduce human escalation by 60%
Your store traffic is finally up — but the nightmare is just beginning. Over 300 buyer inquiries flood in every day — sizing, shipping, returns, product features… the volume is overwhelming. Have you done the math? For a store averaging 200 orders a day, just replying to these messages costs $250 to $600 per month in customer service payroll. Worse still, slow response times immediately trigger bad reviews and refund rates.
Existing auto-reply tools aren't much better — most are overpriced and underpowered. Single-store plans run $7-$28 per month. Five sub-stores? That's over $100. And the intelligence level is laughable — most are just keyword matching. A buyer asks "Does this run big?" and the bot replies "Our sizing is standard" — robotic and useless.
But in 2026, the game has changed. Large language model APIs have crashed in price — DeepSeek costs just 2 RMB per million tokens. Processing 500 messages a day means monthly API costs under 20 RMB. Combine that with Feishu and Qianniu's free interfaces, and this system is essentially zero-cost. Spend half a day setting it up, and the AI handles 80% of your customer service workload.
Three-Layer Architecture: How Messages Flow
The auto-reply system has three clean layers, each with a distinct job.
Message Intake Layer — Taobao Qianniu Open Platform's message push interface. When a buyer sends a message on Wangwang, Qianniu instantly pushes it to your configured server address. This is the cheapest link in the chain — just register a Qianniu developer account and set up a callback URL.
Intelligence Layer — DeepSeek API handles two tasks: intent recognition and reply generation. DeepSeek's Chinese understanding is top-tier. It determines whether the buyer is asking about sizing, shipping, or returns, then pulls the most relevant answer from your knowledge base.
Message Sending Layer — Qianniu's send API pushes the AI-generated reply back to the buyer. Total latency: 2 to 4 seconds. Buyers will never know they're talking to an AI.
Can't code? No problem. I'll cover no-code platforms like Qingliu and Jijianyun later — drag, drop, and you're up and running.
Knowledge Base: The Make-or-Break Factor
90% of AI auto-reply quality depends on your knowledge base. Think of it as the AI's "exam syllabus" — content not in the syllabus can never be answered correctly.
Step 1: Categorize common questions. Export Qianniu chat histories from the last 3 months. Sort by topic and identify the top 20 most frequent questions. Prepare standard replies for each. Typical categories include: sizing inquiries (standard and special size charts, fit notes), shipping queries (carrier, delivery time, fees), return/exchange policies (no-reason return rules, who pays shipping, refund timeline), product features (usage, precautions, troubleshooting).
Step 2: Write the knowledge base document. Use Markdown — AI models love structured text. Each entry should include: question category, possible phrasings (3-5 variations), standard reply, and notes. Example:
## Category: Size Inquiry
### User Phrasings
- I'm 180cm, what size should I get?
- Does this run big or small?
- I usually wear L, what size for this?
### Standard Reply
Hi there! Based on your info, I'd recommend size XL. This product has a slim fit — go up one size if you prefer a looser style.
### Notes
If the user weighs over 90kg, recommend one size up. Sports suits run slightly looser than regular suits.
Step 3: Upload to Feishu Docs. Feishu supports document API access, so Feishu docs serve as your knowledge base storage. When the AI receives a user message, it reads the relevant info from Feishu before generating a reply. No extra database needed — just Feishu's built-in document management.
DeepSeek API Configuration: A Good System Prompt Is Half the Battle
DeepSeek is currently the best-value Chinese LLM at 1/50th the price of GPT-4. The key to configuring it is the System Prompt — this is your set of rules for the AI.
Here's my recommended system prompt:
You are a professional Taobao store AI customer service assistant. Your task is to find the best matching answer from the knowledge base based on the user's question.
Rules:
1. Only use information from the knowledge base — never fabricate information
2. Be friendly and warm — start with "亲" or "您好"
3. If no matching answer is found in the knowledge base, reply: "亲，抱歉，这个问题我需要转给人工客服为您处理，马上为您转接"
4. Don't ask for users' personal information
5. Keep replies under 50 characters — concise is key
6. For returns/exchanges/refunds, guide users to human customer service
7. Reply tone should be professional and warm, not mechanical
Parameters: Set temperature to 0.3 — the sweet spot between creativity and accuracy. Too high and replies become erratic; too low and they sound robotic. Set max_tokens to 200 to keep replies concise.
Setting Up Your Feishu Bot: Step by Step
Feishu's built-in bot feature is the infrastructure for this system. It supports webhook triggers and auto-replies — completely free.
Step 1: Create a Feishu app. Open Feishu Open Platform, create an enterprise self-built app. Enable "Message & Group" and "Document" read permissions.
Step 2: Create a bot. Add bot functionality to your app. Set message reception mode to "Receive Messages."
Step 3: Build the knowledge base. Create a "CS Knowledge Base" document in Feishu. Write out the categorized questions and replies.
Step 4: Configure auto-reply logic. Set up event callbacks in the Feishu app. When a user sends a message, it triggers a callback that sends the message to your backend. Your backend calls DeepSeek and sends the reply back through Feishu's API.
Step 5: Test and launch. Send yourself test messages. Confirm all question types get correct replies, then go live.
No-Code Solutions for Non-Technical Users
Can't write code? Use no-code tools like Qingliu or Jijianyun. Less flexible, but achieves about 80% of the effect.
In Qingliu, create a "Customer Service Auto-Reply" flow: Set up a webhook trigger (new Qianniu message) → Add a "Condition Judgment" node → Call DeepSeek API → Send the reply back via HTTP request.
Setup takes about 2 hours. Data flow isn't as smooth as a custom build, but it works for daily use. Updates are just a few clicks.
Jijianyun is similar — it has built-in connectors for Qianniu and Feishu. Just drag and drop. No manual API coding needed.
Real-World Taobao Qianniu Integration
Qianniu Open Platform provides complete IM message APIs. You mainly need two: the message push callback and the message send API.
Register as a developer on Qianniu Open Platform. Create an app and select the "Message Service" API package. Configure the message callback URL with your backend address. Qianniu will push all buyer messages there.
The push data structure is simple: sender nickname, receiver nickname, message content, message type (text/image/emoji), and session ID. We only process text messages. For images, reply: "I've received your image — please describe your issue in text."
The send API needs two parameters: session ID and message content. Rate limit is 10 calls per second — plenty for small and medium sellers.
Scaling for Million-Order Volume
When orders exceed 1,000 per day, the basic setup may struggle. Here are the key optimizations:
Caching layer: Cache high-frequency replies in Redis. When a user asks a common question, the AI returns cached results without calling the API. Hit rate can exceed 60%.
Multi-model fallback: Use DeepSeek as primary, GPT-4o-mini as backup. Auto-switch on timeouts or errors.
Human escalation circuit breaker: After 3 consecutive failed answers, auto-escalate to a human agent. This prevents the AI from making things worse.
Real-time monitoring: Configure a Feishu dashboard with alerts when the message queue exceeds 100 or API success rate drops below 95%.
Data Analysis and Continuous Optimization
Once live, ongoing optimization matters more than the initial setup. Build a data-driven iteration loop:
Daily: Check the "AI Reply Quality Report" — total messages, successful AI replies, human escalations, average response time. The key metric is "AI resolution rate" — percentage where the AI replied successfully and the user didn't follow up. Above 80% is healthy.
Weekly: Review "Unresolved Case Analysis." Extract questions the AI couldn't answer. Figure out if the issue is missing knowledge or unclear phrasing. Update accordingly. Expect 2-3 percentage points improvement every two weeks.
Monthly: Run a "User Satisfaction Survey." Randomly sample 100 AI-replied conversations. Score each on a 5-point scale. Investigate anything below 3.
FAQ
Q: Won't buyers notice the AI replies feel fake? A: With a well-written system prompt and 2-4 seconds of response delay, most buyers won't realize they're talking to AI. The key is a complete knowledge base — no nonsense answers.
Q: What if the AI gets it wrong? A: Monitor manually for the first 3 days after launch. Intervene immediately on errors and update the knowledge base. After 3 days, the system stabilizes. Also configure the circuit breaker — auto-escalate after consecutive failures.
Q: Do different platforms require different setups? A: Qianniu, Pinduoduo Merchant, and Douyin all use the same logic — configure a message push callback. Douyin requires replies within 30 seconds, which demands faster response times. Shopify connects through Zendesk or Tidio. For Alibaba International Station, use GPT-4o-mini instead of DeepSeek for better English handling.
Q: What's the total cost? A: DeepSeek API ≈ 9 RMB/month. Server ≈ 50 RMB (or zero with serverless). Feishu and Qianniu APIs are free. Total: 9-100 RMB/month. Compared to thousands for human agents, that's a 99%+ savings.
Q: What's the pre-launch checklist? Checklist: Knowledge base covers 80%+ of common questions; every reply is human-reviewed; products and shipping info are up-to-date; API keys and permissions are correct; message push callbacks work; manual takeover and emergency shutoff are configured.
Summary
AI auto-reply isn't about replacing human customer service — it's about deploying limited resources where they matter most. Routine questions about sizing, shipping, and returns are AI's sweet spot. Human agents get freed from repetition to focus on real complex issues and emotional escalations.
This Feishu + DeepSeek + Qianniu solution is near-zero cost and highly effective. Spend half a day building your knowledge base, and the next day you'll see workloads drop while reply quality improves. Cheaper, better, and less stressful — what's not to love?