Supermarket Chain Resolves 94% of Customer Issues Without Human Agents
A national supermarket chain deployed AI agents across Telegram and WhatsApp to handle returns, store queries, and投诉 — cutting response time from 4 hours to under 2 minutes.
The chain's customer service team handled 2,000+ daily inquiries across phone, email, and social media. Common issues — checking receipts, processing returns, finding store hours — consumed 70% of agent time. Peak seasons caused 4+ hour wait times and a 22% abandonment rate.
We deployed Oido AI agents on Telegram and WhatsApp that handle the full lifecycle of common customer issues: receipt lookup via the POS system MCP server, return eligibility checking, store locator queries, and complaint escalation. Human agents only接手 complex cases.
The Problem
The supermarket chain's customer service was built for a pre-digital era. Customers could call a central number during business hours, email, or visit a store in person. As the chain grew to 120 locations, volume outpaced capacity.
The breakdown was predictable: peak hours (11 AM — 2 PM) saw 45-minute phone queues. Email response time drifted to 4+ hours. Social media messages were often missed entirely. The abandonment rate hit 22% during holiday seasons, and abandoned complaints frequently escalated to public negative reviews.
The Solution
We deployed a single AI agent with multiple skill paths, accessible via Telegram and WhatsApp — channels customers already had open on their phones.
Skill paths
Receipt lookup & returns — The agent connects to the POS system via a custom MCP server, looks up the transaction by order ID or loyalty card number, checks the return policy, and generates a return code the customer can use at any store.
Store locator & hours — The agent queries a Postgres database of store locations and operating hours, returning results sorted by distance from the customer's location (requested once, stored for the session).
Complaint intake & escalation — The agent logs structured complaints directly into the existing CRM via n8n. If the customer expresses high emotion or requests a human, the ticket is flagged as priority and assigned to a human agent within the SLA.
Deployment
The system was rolled out in phases across 8 weeks. Phase 1 (Telegram only, 3 stores) ran for 2 weeks alongside the existing team. Phase 2 expanded to WhatsApp and all 120 stores. Phase 3 added the escalation workflow and human handoff.
The Results
| Metric | Before | After |
|---|---|---|
| First response time | 4 hours | 90 seconds |
| Resolution rate (no human) | 0% | 94% |
| Customer satisfaction | 76% | 91% |
| Support headcount | 18 FTE | 6 FTE |
| Coverage | 9 AM — 6 PM | 24/7 |
| Peak abandonment rate | 22% | 3% |
The chain is now expanding the agents to handle order tracking, loyalty program queries, and personalized offer delivery based on purchase history.