The Problem
Brand: Mid-size fashion D2C brand, 300+ orders/day, selling across Shopify + Amazon.
Support load: 500+ customer queries/day across WhatsApp, email, and Instagram DMs.
Team: 4 full-time support agents (₹18K/month each = ₹72K/month)
Average response time: 3-4 hours. Customer satisfaction: 3.2/5.
The Fix: AI Chatbot + Human Hybrid
Tool Selected: Interakt (WhatsApp) + Tidio (Website)
Why this combination: Interakt handles WhatsApp (70% of queries), Tidio handles website chat (20%), email remains human-managed (10%).
Implementation Timeline: 3 Weeks
Week 1: Data preparation
- Exported 3 months of support conversations (15,000+ queries)
- Categorized by type: order tracking (38%), returns/exchange (22%), sizing (15%), product queries (12%), complaints (8%), other (5%)
- Created FAQ document: 50 questions with detailed answers
- Mapped common conversation flows for each category
Week 2: Bot setup and training
- Configured Interakt chatbot with the FAQ knowledge base
- Built automated flows: order tracking (connects to courier API), return initiation, size recommendation
- Set up escalation rules: transfer to human if bot confidence is low, or customer asks for human
- Tested with team members pretending to be customers — fixed edge cases
Week 3: Gradual rollout
- Day 1-3: Bot handles 20% of queries (random routing). Humans monitor all bot responses.
- Day 4-7: Bot handles 50%. Human review on bot-resolved conversations to catch errors.
- Day 8-14: Bot handles 80%. Humans only get escalated queries.
- Day 15+: Full deployment. Bot as first point of contact for ALL queries.
What AI Handles vs What Humans Handle
| Query Type | AI Handles? | How |
|---|---|---|
| ‘Where is my order?’ | Yes (100%) | Bot asks for order number → fetches tracking from API → sends status |
| ‘How to return?’ | Yes (90%) | Bot collects return reason → checks eligibility → initiates return ticket |
| ‘What size should I order?’ | Yes (85%) | Bot asks height/weight/usual size → recommends based on size chart |
| ‘Do you have X in blue?’ | Yes (80%) | Bot searches catalog → shows available options |
| ‘Product is damaged’ | Partial | Bot collects photos and details → creates ticket → routes to human for resolution |
| Payment/refund issues | No | Immediately routes to human agent |
| Angry/escalated customer | No | Bot detects negative sentiment → immediate human handoff |
The Results (After 3 Months)
| Metric | Before AI | After AI | Change |
|---|---|---|---|
| Queries handled by humans | 500/day | 180/day | -64% |
| Average response time | 3-4 hours | Under 2 minutes | -98% |
| Support team size | 4 agents | 2 agents (₹36K/month saved) | -50% |
| Customer satisfaction | 3.2/5 | 4.1/5 | +28% |
| Monthly support cost | ₹72K | ₹28K (agents) + ₹5K (tools) | -54% |
| Resolution rate (first contact) | 65% | 82% | +26% |
Key Lessons
- AI handles volume, humans handle complexity — The goal isn’t replacing humans, it’s freeing them for high-value interactions.
- Training data quality matters more than AI model quality — A well-trained simple bot outperforms a poorly trained advanced one.
- Always offer human handoff — The #1 customer complaint with chatbots is being stuck without a way to reach a person.
- Monitor weekly — Review bot conversations weekly. Customers ask new questions. Keep updating the knowledge base.
- Hindi/Hinglish support is mandatory — 45% of queries came in Hindi or Hinglish. The bot needed to understand both.
Want AI support that cuts your bill 60%?
A well-trained simple bot — built for the 80% of repeat questions, with proper Hindi/Hinglish handling and clean human-handoff — beats a poorly-trained advanced one. The implementation is 3-4 weeks if we run it. The 60% support-cost reduction holds when bot accuracy stays above 88% and human handoff stays below 12%. We’ve done it for 200+ Indian D2C brands. ₹385Cr+ revenue processed. 4.5x average ROI. 98% retention.
The Shopify build is ₹50,000 fixed-price with no AMC — bug fixes for what we ship are included for the lifetime of the store.
Start a WhatsApp chat: Message the Growww Tech team on WhatsApp →
Related reading:
Leave a Reply
You must be logged in to post a comment.